diff --git a/.github/workflows/rust-ci.yml b/.github/workflows/rust-ci.yml
index 7918629165..103e8cf6bd 100644
--- a/.github/workflows/rust-ci.yml
+++ b/.github/workflows/rust-ci.yml
@@ -9,7 +9,7 @@ on:
 # CI builds in debug (dev) for faster signal.
 
 jobs:
-  # --- Detect what changed (always runs) -------------------------------------
+  # --- Detect what changed to detect which tests to run (always runs) -------------------------------------
   changed:
     name: Detect changed areas
     runs-on: ubuntu-24.04
@@ -84,8 +84,8 @@ jobs:
         run: cargo shear
 
   # --- CI to validate on different os/targets --------------------------------
-  lint_build_test:
-    name: ${{ matrix.runner }} - ${{ matrix.target }}${{ matrix.profile == 'release' && ' (release)' || '' }}
+  lint_build:
+    name: Lint/Build — ${{ matrix.runner }} - ${{ matrix.target }}${{ matrix.profile == 'release' && ' (release)' || '' }}
     runs-on: ${{ matrix.runner }}
     timeout-minutes: 30
     needs: changed
@@ -94,6 +94,11 @@ jobs:
     defaults:
       run:
         working-directory: codex-rs
+    env:
+      # Speed up repeated builds across CI runs by caching compiled objects.
+      RUSTC_WRAPPER: sccache
+      CARGO_INCREMENTAL: "0"
+      SCCACHE_CACHE_SIZE: 10G
 
     strategy:
       fail-fast: false
@@ -159,20 +164,83 @@ jobs:
             ~/.cargo/registry/index/
             ~/.cargo/registry/cache/
             ~/.cargo/git/db/
-          key: cargo-home-${{ matrix.runner }}-${{ matrix.target }}-${{ matrix.profile }}-${{ hashFiles('**/Cargo.lock') }}
+          key: cargo-home-${{ matrix.runner }}-${{ matrix.target }}-${{ matrix.profile }}-${{ hashFiles('**/Cargo.lock') }}-${{ hashFiles('codex-rs/rust-toolchain.toml') }}
+          restore-keys: |
+            cargo-home-${{ matrix.runner }}-${{ matrix.target }}-${{ matrix.profile }}-
 
-      - name: Restore target cache (except gnu-dev)
-        id: cache_target_restore
-        if: ${{ !(matrix.target == 'x86_64-unknown-linux-gnu' && matrix.profile != 'release') }}
+      # Install and restore sccache cache
+      - name: Install sccache
+        uses: taiki-e/install-action@0c5db7f7f897c03b771660e91d065338615679f4 # v2
+        with:
+          tool: sccache
+          version: 0.7.5
+
+      - name: Configure sccache backend
+        shell: bash
+        run: |
+          set -euo pipefail
+          if [[ -n "${ACTIONS_CACHE_URL:-}" && -n "${ACTIONS_RUNTIME_TOKEN:-}" ]]; then
+            echo "SCCACHE_GHA_ENABLED=true" >> "$GITHUB_ENV"
+            echo "Using sccache GitHub backend"
+          else
+            echo "SCCACHE_GHA_ENABLED=false" >> "$GITHUB_ENV"
+            echo "SCCACHE_DIR=${{ github.workspace }}/.sccache" >> "$GITHUB_ENV"
+            echo "Using sccache local disk + actions/cache fallback"
+          fi
+
+      - name: Restore sccache cache (fallback)
+        if: ${{ env.SCCACHE_GHA_ENABLED != 'true' }}
+        id: cache_sccache_restore
         uses: actions/cache/restore@v4
         with:
-          path: ${{ github.workspace }}/codex-rs/target/
-          key: cargo-target-${{ matrix.runner }}-${{ matrix.target }}-${{ matrix.profile }}-${{ hashFiles('**/Cargo.lock') }}
+          path: ${{ github.workspace }}/.sccache/
+          key: sccache-${{ matrix.runner }}-${{ matrix.target }}-${{ matrix.profile }}-${{ hashFiles('**/Cargo.lock') }}-${{ github.run_id }}
+          restore-keys: |
+            sccache-${{ matrix.runner }}-${{ matrix.target }}-${{ matrix.profile }}-${{ hashFiles('**/Cargo.lock') }}-
+            sccache-${{ matrix.runner }}-${{ matrix.target }}-${{ matrix.profile }}-
+
+      - if: ${{ matrix.target == 'x86_64-unknown-linux-musl' || matrix.target == 'aarch64-unknown-linux-musl'}}
+        name: Prepare APT cache directories (musl)
+        shell: bash
+        run: |
+          set -euo pipefail
+          sudo mkdir -p /var/cache/apt/archives /var/lib/apt/lists
+          sudo chown -R "$USER:$USER" /var/cache/apt /var/lib/apt/lists
+
+      - if: ${{ matrix.target == 'x86_64-unknown-linux-musl' || matrix.target == 'aarch64-unknown-linux-musl'}}
+        name: Restore APT cache (musl)
+        id: cache_apt_restore
+        uses: actions/cache/restore@v4
+        with:
+          path: |
+            /var/cache/apt
+          key: apt-${{ matrix.runner }}-${{ matrix.target }}-v1
 
       - if: ${{ matrix.target == 'x86_64-unknown-linux-musl' || matrix.target == 'aarch64-unknown-linux-musl'}}
         name: Install musl build tools
+        env:
+          DEBIAN_FRONTEND: noninteractive
+        shell: bash
         run: |
-          sudo apt install -y musl-tools pkg-config && sudo rm -rf /var/lib/apt/lists/*
+          set -euo pipefail
+          sudo apt-get -y update -o Acquire::Retries=3
+          sudo apt-get -y install --no-install-recommends musl-tools pkg-config
+
+      - name: Install cargo-chef
+        if: ${{ matrix.profile == 'release' }}
+        uses: taiki-e/install-action@0c5db7f7f897c03b771660e91d065338615679f4 # v2
+        with:
+          tool: cargo-chef
+          version: 0.1.71
+
+      - name: Pre-warm dependency cache (cargo-chef)
+        if: ${{ matrix.profile == 'release' }}
+        shell: bash
+        run: |
+          set -euo pipefail
+          RECIPE="${RUNNER_TEMP}/chef-recipe.json"
+          cargo chef prepare --recipe-path "$RECIPE"
+          cargo chef cook --recipe-path "$RECIPE" --target ${{ matrix.target }} --release --all-features
 
       - name: cargo clippy
         id: clippy
@@ -191,20 +259,6 @@ jobs:
           find . -name Cargo.toml -mindepth 2 -maxdepth 2 -print0 \
             | xargs -0 -n1 -I{} bash -c 'cd "$(dirname "{}")" && cargo check --profile ${{ matrix.profile }}'
 
-      - uses: taiki-e/install-action@0c5db7f7f897c03b771660e91d065338615679f4 # v2
-        with:
-          tool: nextest
-          version: 0.9.103
-
-      - name: tests
-        id: test
-        # Tests take too long for release builds to run them on every PR.
-        if: ${{ matrix.profile != 'release' }}
-        continue-on-error: true
-        run: cargo nextest run --all-features --no-fail-fast --target ${{ matrix.target }} --cargo-profile ci-test
-        env:
-          RUST_BACKTRACE: 1
-
       # Save caches explicitly; make non-fatal so cache packaging
       # never fails the overall job. Only save when key wasn't hit.
       - name: Save cargo home cache
@@ -217,33 +271,193 @@ jobs:
             ~/.cargo/registry/index/
             ~/.cargo/registry/cache/
             ~/.cargo/git/db/
-          key: cargo-home-${{ matrix.runner }}-${{ matrix.target }}-${{ matrix.profile }}-${{ hashFiles('**/Cargo.lock') }}
+          key: cargo-home-${{ matrix.runner }}-${{ matrix.target }}-${{ matrix.profile }}-${{ hashFiles('**/Cargo.lock') }}-${{ hashFiles('codex-rs/rust-toolchain.toml') }}
 
-      - name: Save target cache (except gnu-dev)
-        if: >-
-          always() && !cancelled() &&
-          (steps.cache_target_restore.outputs.cache-hit != 'true') &&
-          !(matrix.target == 'x86_64-unknown-linux-gnu' && matrix.profile != 'release')
+      - name: Save sccache cache (fallback)
+        if: always() && !cancelled() && env.SCCACHE_GHA_ENABLED != 'true'
         continue-on-error: true
         uses: actions/cache/save@v4
         with:
-          path: ${{ github.workspace }}/codex-rs/target/
-          key: cargo-target-${{ matrix.runner }}-${{ matrix.target }}-${{ matrix.profile }}-${{ hashFiles('**/Cargo.lock') }}
+          path: ${{ github.workspace }}/.sccache/
+          key: sccache-${{ matrix.runner }}-${{ matrix.target }}-${{ matrix.profile }}-${{ hashFiles('**/Cargo.lock') }}-${{ github.run_id }}
+
+      - name: sccache stats
+        if: always()
+        continue-on-error: true
+        run: sccache --show-stats || true
+
+      - name: sccache summary
+        if: always()
+        shell: bash
+        run: |
+          {
+            echo "### sccache stats — ${{ matrix.target }} (${{ matrix.profile }})";
+            echo;
+            echo '```';
+            sccache --show-stats || true;
+            echo '```';
+          } >> "$GITHUB_STEP_SUMMARY"
+
+      - name: Save APT cache (musl)
+        if: always() && !cancelled() && (matrix.target == 'x86_64-unknown-linux-musl' || matrix.target == 'aarch64-unknown-linux-musl') && steps.cache_apt_restore.outputs.cache-hit != 'true'
+        continue-on-error: true
+        uses: actions/cache/save@v4
+        with:
+          path: |
+            /var/cache/apt
+          key: apt-${{ matrix.runner }}-${{ matrix.target }}-v1
 
       # Fail the job if any of the previous steps failed.
       - name: verify all steps passed
         if: |
           steps.clippy.outcome == 'failure' ||
-          steps.cargo_check_all_crates.outcome == 'failure' ||
-          steps.test.outcome == 'failure'
+          steps.cargo_check_all_crates.outcome == 'failure'
         run: |
-          echo "One or more checks failed (clippy, cargo_check_all_crates, or test). See logs for details."
+          echo "One or more checks failed (clippy or cargo_check_all_crates). See logs for details."
+          exit 1
+
+  tests:
+    name: Tests — ${{ matrix.runner }} - ${{ matrix.target }}
+    runs-on: ${{ matrix.runner }}
+    timeout-minutes: 30
+    needs: changed
+    if: ${{ needs.changed.outputs.codex == 'true' || needs.changed.outputs.workflows == 'true' || github.event_name == 'push' }}
+    defaults:
+      run:
+        working-directory: codex-rs
+    env:
+      RUSTC_WRAPPER: sccache
+      CARGO_INCREMENTAL: "0"
+      SCCACHE_CACHE_SIZE: 10G
+
+    strategy:
+      fail-fast: false
+      matrix:
+        include:
+          - runner: macos-14
+            target: aarch64-apple-darwin
+            profile: dev
+          - runner: ubuntu-24.04
+            target: x86_64-unknown-linux-gnu
+            profile: dev
+          - runner: ubuntu-24.04-arm
+            target: aarch64-unknown-linux-gnu
+            profile: dev
+          - runner: windows-latest
+            target: x86_64-pc-windows-msvc
+            profile: dev
+          - runner: windows-11-arm
+            target: aarch64-pc-windows-msvc
+            profile: dev
+
+    steps:
+      - uses: actions/checkout@v5
+      - uses: dtolnay/rust-toolchain@1.90
+        with:
+          targets: ${{ matrix.target }}
+
+      - name: Restore cargo home cache
+        id: cache_cargo_home_restore
+        uses: actions/cache/restore@v4
+        with:
+          path: |
+            ~/.cargo/bin/
+            ~/.cargo/registry/index/
+            ~/.cargo/registry/cache/
+            ~/.cargo/git/db/
+          key: cargo-home-${{ matrix.runner }}-${{ matrix.target }}-${{ matrix.profile }}-${{ hashFiles('**/Cargo.lock') }}-${{ hashFiles('codex-rs/rust-toolchain.toml') }}
+          restore-keys: |
+            cargo-home-${{ matrix.runner }}-${{ matrix.target }}-${{ matrix.profile }}-
+
+      - name: Install sccache
+        uses: taiki-e/install-action@0c5db7f7f897c03b771660e91d065338615679f4 # v2
+        with:
+          tool: sccache
+          version: 0.7.5
+
+      - name: Configure sccache backend
+        shell: bash
+        run: |
+          set -euo pipefail
+          if [[ -n "${ACTIONS_CACHE_URL:-}" && -n "${ACTIONS_RUNTIME_TOKEN:-}" ]]; then
+            echo "SCCACHE_GHA_ENABLED=true" >> "$GITHUB_ENV"
+            echo "Using sccache GitHub backend"
+          else
+            echo "SCCACHE_GHA_ENABLED=false" >> "$GITHUB_ENV"
+            echo "SCCACHE_DIR=${{ github.workspace }}/.sccache" >> "$GITHUB_ENV"
+            echo "Using sccache local disk + actions/cache fallback"
+          fi
+
+      - name: Restore sccache cache (fallback)
+        if: ${{ env.SCCACHE_GHA_ENABLED != 'true' }}
+        id: cache_sccache_restore
+        uses: actions/cache/restore@v4
+        with:
+          path: ${{ github.workspace }}/.sccache/
+          key: sccache-${{ matrix.runner }}-${{ matrix.target }}-${{ matrix.profile }}-${{ hashFiles('**/Cargo.lock') }}-${{ github.run_id }}
+          restore-keys: |
+            sccache-${{ matrix.runner }}-${{ matrix.target }}-${{ matrix.profile }}-${{ hashFiles('**/Cargo.lock') }}-
+            sccache-${{ matrix.runner }}-${{ matrix.target }}-${{ matrix.profile }}-
+
+      - uses: taiki-e/install-action@0c5db7f7f897c03b771660e91d065338615679f4 # v2
+        with:
+          tool: nextest
+          version: 0.9.103
+
+      - name: tests
+        id: test
+        continue-on-error: true
+        run: cargo nextest run --all-features --no-fail-fast --target ${{ matrix.target }} --cargo-profile ci-test
+        env:
+          RUST_BACKTRACE: 1
+
+      - name: Save cargo home cache
+        if: always() && !cancelled() && steps.cache_cargo_home_restore.outputs.cache-hit != 'true'
+        continue-on-error: true
+        uses: actions/cache/save@v4
+        with:
+          path: |
+            ~/.cargo/bin/
+            ~/.cargo/registry/index/
+            ~/.cargo/registry/cache/
+            ~/.cargo/git/db/
+          key: cargo-home-${{ matrix.runner }}-${{ matrix.target }}-${{ matrix.profile }}-${{ hashFiles('**/Cargo.lock') }}-${{ hashFiles('codex-rs/rust-toolchain.toml') }}
+
+      - name: Save sccache cache (fallback)
+        if: always() && !cancelled() && env.SCCACHE_GHA_ENABLED != 'true'
+        continue-on-error: true
+        uses: actions/cache/save@v4
+        with:
+          path: ${{ github.workspace }}/.sccache/
+          key: sccache-${{ matrix.runner }}-${{ matrix.target }}-${{ matrix.profile }}-${{ hashFiles('**/Cargo.lock') }}-${{ github.run_id }}
+
+      - name: sccache stats
+        if: always()
+        continue-on-error: true
+        run: sccache --show-stats || true
+
+      - name: sccache summary
+        if: always()
+        shell: bash
+        run: |
+          {
+            echo "### sccache stats — ${{ matrix.target }} (tests)";
+            echo;
+            echo '```';
+            sccache --show-stats || true;
+            echo '```';
+          } >> "$GITHUB_STEP_SUMMARY"
+
+      - name: verify tests passed
+        if: steps.test.outcome == 'failure'
+        run: |
+          echo "Tests failed. See logs for details."
           exit 1
 
   # --- Gatherer job that you mark as the ONLY required status -----------------
   results:
     name: CI results (required)
-    needs: [changed, general, cargo_shear, lint_build_test]
+    needs: [changed, general, cargo_shear, lint_build, tests]
     if: always()
     runs-on: ubuntu-24.04
     steps:
@@ -252,7 +466,8 @@ jobs:
         run: |
           echo "general: ${{ needs.general.result }}"
           echo "shear  : ${{ needs.cargo_shear.result }}"
-          echo "matrix : ${{ needs.lint_build_test.result }}"
+          echo "lint   : ${{ needs.lint_build.result }}"
+          echo "tests  : ${{ needs.tests.result }}"
 
           # If nothing relevant changed (PR touching only root README, etc.),
           # declare success regardless of other jobs.
@@ -264,4 +479,10 @@ jobs:
           # Otherwise require the jobs to have succeeded
           [[ '${{ needs.general.result }}' == 'success' ]] || { echo 'general failed'; exit 1; }
           [[ '${{ needs.cargo_shear.result }}' == 'success' ]] || { echo 'cargo_shear failed'; exit 1; }
-          [[ '${{ needs.lint_build_test.result }}' == 'success' ]] || { echo 'matrix failed'; exit 1; }
+          [[ '${{ needs.lint_build.result }}' == 'success' ]] || { echo 'lint_build failed'; exit 1; }
+          [[ '${{ needs.tests.result }}' == 'success' ]] || { echo 'tests failed'; exit 1; }
+
+      - name: sccache summary note
+        if: always()
+        run: |
+          echo "Per-job sccache stats are attached to each matrix job's Step Summary."
diff --git a/.github/workflows/rust-release.yml b/.github/workflows/rust-release.yml
index 5beadd55c4..fa3e22894e 100644
--- a/.github/workflows/rust-release.yml
+++ b/.github/workflows/rust-release.yml
@@ -53,6 +53,10 @@ jobs:
     defaults:
       run:
         working-directory: codex-rs
+    env:
+      RUSTC_WRAPPER: sccache
+      CARGO_INCREMENTAL: "0"
+      SCCACHE_CACHE_SIZE: 10G
 
     strategy:
       fail-fast: false
@@ -88,14 +92,80 @@ jobs:
             ~/.cargo/registry/index/
             ~/.cargo/registry/cache/
             ~/.cargo/git/db/
-            ${{ github.workspace }}/codex-rs/target/
-          key: cargo-${{ matrix.runner }}-${{ matrix.target }}-release-${{ hashFiles('**/Cargo.lock') }}
+          key: cargo-${{ matrix.runner }}-${{ matrix.target }}-release-${{ hashFiles('**/Cargo.lock') }}-${{ hashFiles('codex-rs/rust-toolchain.toml') }}
+          restore-keys: |
+            cargo-${{ matrix.runner }}-${{ matrix.target }}-release-
+
+      - name: Install sccache
+        uses: taiki-e/install-action@0c5db7f7f897c03b771660e91d065338615679f4 # v2
+        with:
+          tool: sccache
+          version: 0.7.5
+
+      - name: Configure sccache backend
+        shell: bash
+        run: |
+          set -euo pipefail
+          if [[ -n "${ACTIONS_CACHE_URL:-}" && -n "${ACTIONS_RUNTIME_TOKEN:-}" ]]; then
+            echo "SCCACHE_GHA_ENABLED=true" >> "$GITHUB_ENV"
+            echo "Using sccache GitHub backend"
+          else
+            echo "SCCACHE_GHA_ENABLED=false" >> "$GITHUB_ENV"
+            echo "SCCACHE_DIR=${{ github.workspace }}/.sccache" >> "$GITHUB_ENV"
+            echo "Using sccache local disk + actions/cache fallback"
+          fi
+
+      - name: Restore sccache cache (fallback)
+        if: ${{ env.SCCACHE_GHA_ENABLED != 'true' }}
+        id: cache_sccache_restore
+        uses: actions/cache/restore@v4
+        with:
+          path: ${{ github.workspace }}/.sccache/
+          key: sccache-${{ matrix.runner }}-${{ matrix.target }}-release-${{ hashFiles('**/Cargo.lock') }}-${{ github.run_id }}
+          restore-keys: |
+            sccache-${{ matrix.runner }}-${{ matrix.target }}-release-${{ hashFiles('**/Cargo.lock') }}-
+            sccache-${{ matrix.runner }}-${{ matrix.target }}-release-
+
+      - if: ${{ matrix.target == 'x86_64-unknown-linux-musl' || matrix.target == 'aarch64-unknown-linux-musl'}}
+        name: Prepare APT cache directories (musl)
+        shell: bash
+        run: |
+          set -euo pipefail
+          sudo mkdir -p /var/cache/apt/archives /var/lib/apt/lists
+          sudo chown -R "$USER:$USER" /var/cache/apt /var/lib/apt/lists
+
+      - if: ${{ matrix.target == 'x86_64-unknown-linux-musl' || matrix.target == 'aarch64-unknown-linux-musl'}}
+        name: Restore APT cache (musl)
+        id: cache_apt_restore
+        uses: actions/cache/restore@v4
+        with:
+          path: |
+            /var/cache/apt
+          key: apt-${{ matrix.runner }}-${{ matrix.target }}-v1
 
       - if: ${{ matrix.target == 'x86_64-unknown-linux-musl' || matrix.target == 'aarch64-unknown-linux-musl'}}
         name: Install musl build tools
+        env:
+          DEBIAN_FRONTEND: noninteractive
+        shell: bash
         run: |
-          sudo apt-get update
-          sudo apt-get install -y musl-tools pkg-config
+          set -euo pipefail
+          sudo apt-get -y update -o Acquire::Retries=3
+          sudo apt-get -y install --no-install-recommends musl-tools pkg-config
+
+      - name: Install cargo-chef
+        uses: taiki-e/install-action@0c5db7f7f897c03b771660e91d065338615679f4 # v2
+        with:
+          tool: cargo-chef
+          version: 0.1.71
+
+      - name: Pre-warm dependency cache (cargo-chef)
+        shell: bash
+        run: |
+          set -euo pipefail
+          RECIPE="${RUNNER_TEMP}/chef-recipe.json"
+          cargo chef prepare --recipe-path "$RECIPE"
+          cargo chef cook --recipe-path "$RECIPE" --target ${{ matrix.target }} --release --all-features
 
       - name: Cargo build
         run: cargo build --target ${{ matrix.target }} --release --bin codex --bin codex-responses-api-proxy
@@ -327,6 +397,40 @@ jobs:
             zstd -T0 -19 --rm "$dest/$base"
           done
 
+      - name: Save APT cache (musl)
+        if: always() && !cancelled() && (matrix.target == 'x86_64-unknown-linux-musl' || matrix.target == 'aarch64-unknown-linux-musl') && steps.cache_apt_restore.outputs.cache-hit != 'true'
+        continue-on-error: true
+        uses: actions/cache/save@v4
+        with:
+          path: |
+            /var/cache/apt
+          key: apt-${{ matrix.runner }}-${{ matrix.target }}-v1
+
+      - name: Save sccache cache (fallback)
+        if: always() && !cancelled() && env.SCCACHE_GHA_ENABLED != 'true'
+        continue-on-error: true
+        uses: actions/cache/save@v4
+        with:
+          path: ${{ github.workspace }}/.sccache/
+          key: sccache-${{ matrix.runner }}-${{ matrix.target }}-release-${{ hashFiles('**/Cargo.lock') }}-${{ github.run_id }}
+
+      - name: sccache stats
+        if: always()
+        continue-on-error: true
+        run: sccache --show-stats || true
+
+      - name: sccache summary
+        if: always()
+        shell: bash
+        run: |
+          {
+            echo "### sccache stats — ${{ matrix.target }} (release)";
+            echo;
+            echo '```';
+            sccache --show-stats || true;
+            echo '```';
+          } >> "$GITHUB_STEP_SUMMARY"
+
       - name: Remove signing keychain
         if: ${{ always() && matrix.runner == 'macos-15-xlarge' }}
         shell: bash
diff --git a/codex-rs/app-server-protocol/src/protocol.rs b/codex-rs/app-server-protocol/src/protocol.rs
index f164ce5601..65a233a376 100644
--- a/codex-rs/app-server-protocol/src/protocol.rs
+++ b/codex-rs/app-server-protocol/src/protocol.rs
@@ -5,7 +5,7 @@ use crate::JSONRPCNotification;
 use crate::JSONRPCRequest;
 use crate::RequestId;
 use codex_protocol::ConversationId;
-use codex_protocol::account::Account;
+use codex_protocol::account::PlanType;
 use codex_protocol::config_types::ForcedLoginMethod;
 use codex_protocol::config_types::ReasoningEffort;
 use codex_protocol::config_types::ReasoningSummary;
@@ -236,6 +236,22 @@ client_request_definitions! {
     },
 }
 
+#[derive(Debug, Clone, PartialEq, Deserialize, Serialize, JsonSchema, TS)]
+#[serde(tag = "type", rename_all = "camelCase")]
+#[ts(tag = "type")]
+pub enum Account {
+    #[serde(rename = "apiKey", rename_all = "camelCase")]
+    #[ts(rename = "apiKey", rename_all = "camelCase")]
+    ApiKey { api_key: String },
+
+    #[serde(rename = "chatgpt", rename_all = "camelCase")]
+    #[ts(rename = "chatgpt", rename_all = "camelCase")]
+    ChatGpt {
+        email: Option<String>,
+        plan_type: PlanType,
+    },
+}
+
 #[derive(Serialize, Deserialize, Debug, Clone, PartialEq, JsonSchema, TS)]
 #[serde(rename_all = "camelCase")]
 pub struct GetAccountResponse {
@@ -1246,6 +1262,35 @@ mod tests {
         Ok(())
     }
 
+    #[test]
+    fn account_serializes_fields_in_camel_case() -> Result<()> {
+        let api_key = Account::ApiKey {
+            api_key: "secret".to_string(),
+        };
+        assert_eq!(
+            json!({
+                "type": "apiKey",
+                "apiKey": "secret",
+            }),
+            serde_json::to_value(&api_key)?,
+        );
+
+        let chatgpt = Account::ChatGpt {
+            email: Some("user@example.com".to_string()),
+            plan_type: PlanType::Plus,
+        };
+        assert_eq!(
+            json!({
+                "type": "chatgpt",
+                "email": "user@example.com",
+                "planType": "plus",
+            }),
+            serde_json::to_value(&chatgpt)?,
+        );
+
+        Ok(())
+    }
+
     #[test]
     fn serialize_list_models() -> Result<()> {
         let request = ClientRequest::ListModels {
diff --git a/codex-rs/app-server/src/codex_message_processor.rs b/codex-rs/app-server/src/codex_message_processor.rs
index fc79d20968..4f8c066f7e 100644
--- a/codex-rs/app-server/src/codex_message_processor.rs
+++ b/codex-rs/app-server/src/codex_message_processor.rs
@@ -323,7 +323,11 @@ impl CodexMessageProcessor {
             }
         }
 
-        match login_with_api_key(&self.config.codex_home, &params.api_key) {
+        match login_with_api_key(
+            &self.config.codex_home,
+            &params.api_key,
+            self.config.cli_auth_credentials_store_mode,
+        ) {
             Ok(()) => {
                 self.auth_manager.reload();
                 self.outgoing
@@ -367,6 +371,7 @@ impl CodexMessageProcessor {
                 config.codex_home.clone(),
                 CLIENT_ID.to_string(),
                 config.forced_chatgpt_workspace_id.clone(),
+                config.cli_auth_credentials_store_mode,
             )
         };
 
diff --git a/codex-rs/app-server/src/message_processor.rs b/codex-rs/app-server/src/message_processor.rs
index a2c192cf2a..7693cc2ff9 100644
--- a/codex-rs/app-server/src/message_processor.rs
+++ b/codex-rs/app-server/src/message_processor.rs
@@ -37,7 +37,11 @@ impl MessageProcessor {
         feedback: CodexFeedback,
     ) -> Self {
         let outgoing = Arc::new(outgoing);
-        let auth_manager = AuthManager::shared(config.codex_home.clone(), false);
+        let auth_manager = AuthManager::shared(
+            config.codex_home.clone(),
+            false,
+            config.cli_auth_credentials_store_mode,
+        );
         let conversation_manager = Arc::new(ConversationManager::new(
             auth_manager.clone(),
             SessionSource::VSCode,
diff --git a/codex-rs/app-server/tests/common/auth_fixtures.rs b/codex-rs/app-server/tests/common/auth_fixtures.rs
index 594b621d31..071a920b89 100644
--- a/codex-rs/app-server/tests/common/auth_fixtures.rs
+++ b/codex-rs/app-server/tests/common/auth_fixtures.rs
@@ -6,6 +6,7 @@ use base64::Engine;
 use base64::engine::general_purpose::URL_SAFE_NO_PAD;
 use chrono::DateTime;
 use chrono::Utc;
+use codex_core::auth::AuthCredentialsStoreMode;
 use codex_core::auth::AuthDotJson;
 use codex_core::auth::save_auth;
 use codex_core::token_data::TokenData;
@@ -108,7 +109,11 @@ pub fn encode_id_token(claims: &ChatGptIdTokenClaims) -> Result<String> {
     Ok(format!("{header_b64}.{payload_b64}.{signature_b64}"))
 }
 
-pub fn write_chatgpt_auth(codex_home: &Path, fixture: ChatGptAuthFixture) -> Result<()> {
+pub fn write_chatgpt_auth(
+    codex_home: &Path,
+    fixture: ChatGptAuthFixture,
+    cli_auth_credentials_store_mode: AuthCredentialsStoreMode,
+) -> Result<()> {
     let id_token_raw = encode_id_token(&fixture.claims)?;
     let id_token = parse_id_token(&id_token_raw).context("parse id token")?;
     let tokens = TokenData {
@@ -126,5 +131,5 @@ pub fn write_chatgpt_auth(codex_home: &Path, fixture: ChatGptAuthFixture) -> Res
         last_refresh,
     };
 
-    save_auth(codex_home, &auth).context("write auth.json")
+    save_auth(codex_home, &auth, cli_auth_credentials_store_mode).context("write auth.json")
 }
diff --git a/codex-rs/app-server/tests/suite/archive_conversation.rs b/codex-rs/app-server/tests/suite/archive_conversation.rs
index 6dcfefdbb8..2d8f746a69 100644
--- a/codex-rs/app-server/tests/suite/archive_conversation.rs
+++ b/codex-rs/app-server/tests/suite/archive_conversation.rs
@@ -1,5 +1,4 @@
-use std::path::Path;
-
+use anyhow::Result;
 use app_test_support::McpProcess;
 use app_test_support::to_response;
 use codex_app_server_protocol::ArchiveConversationParams;
@@ -9,45 +8,37 @@ use codex_app_server_protocol::NewConversationParams;
 use codex_app_server_protocol::NewConversationResponse;
 use codex_app_server_protocol::RequestId;
 use codex_core::ARCHIVED_SESSIONS_SUBDIR;
+use std::path::Path;
 use tempfile::TempDir;
 use tokio::time::timeout;
 
 const DEFAULT_READ_TIMEOUT: std::time::Duration = std::time::Duration::from_secs(10);
 
 #[tokio::test(flavor = "multi_thread", worker_threads = 2)]
-async fn archive_conversation_moves_rollout_into_archived_directory() {
-    let codex_home = TempDir::new().expect("create temp dir");
-    create_config_toml(codex_home.path()).expect("write config.toml");
+async fn archive_conversation_moves_rollout_into_archived_directory() -> Result<()> {
+    let codex_home = TempDir::new()?;
+    create_config_toml(codex_home.path())?;
 
-    let mut mcp = McpProcess::new(codex_home.path())
-        .await
-        .expect("spawn mcp process");
-    timeout(DEFAULT_READ_TIMEOUT, mcp.initialize())
-        .await
-        .expect("initialize timeout")
-        .expect("initialize request");
+    let mut mcp = McpProcess::new(codex_home.path()).await?;
+    timeout(DEFAULT_READ_TIMEOUT, mcp.initialize()).await??;
 
     let new_request_id = mcp
         .send_new_conversation_request(NewConversationParams {
             model: Some("mock-model".to_string()),
             ..Default::default()
         })
-        .await
-        .expect("send newConversation");
+        .await?;
     let new_response: JSONRPCResponse = timeout(
         DEFAULT_READ_TIMEOUT,
         mcp.read_stream_until_response_message(RequestId::Integer(new_request_id)),
     )
-    .await
-    .expect("newConversation timeout")
-    .expect("newConversation response");
+    .await??;
 
     let NewConversationResponse {
         conversation_id,
         rollout_path,
         ..
-    } = to_response::<NewConversationResponse>(new_response)
-        .expect("deserialize newConversation response");
+    } = to_response::<NewConversationResponse>(new_response)?;
 
     assert!(
         rollout_path.exists(),
@@ -60,19 +51,15 @@ async fn archive_conversation_moves_rollout_into_archived_directory() {
             conversation_id,
             rollout_path: rollout_path.clone(),
         })
-        .await
-        .expect("send archiveConversation");
+        .await?;
     let archive_response: JSONRPCResponse = timeout(
         DEFAULT_READ_TIMEOUT,
         mcp.read_stream_until_response_message(RequestId::Integer(archive_request_id)),
     )
-    .await
-    .expect("archiveConversation timeout")
-    .expect("archiveConversation response");
+    .await??;
 
     let _: ArchiveConversationResponse =
-        to_response::<ArchiveConversationResponse>(archive_response)
-            .expect("deserialize archiveConversation response");
+        to_response::<ArchiveConversationResponse>(archive_response)?;
 
     let archived_directory = codex_home.path().join(ARCHIVED_SESSIONS_SUBDIR);
     let archived_rollout_path =
@@ -90,6 +77,8 @@ async fn archive_conversation_moves_rollout_into_archived_directory() {
         "expected archived rollout path {} to exist",
         archived_rollout_path.display()
     );
+
+    Ok(())
 }
 
 fn create_config_toml(codex_home: &Path) -> std::io::Result<()> {
diff --git a/codex-rs/app-server/tests/suite/auth.rs b/codex-rs/app-server/tests/suite/auth.rs
index aa6cc1bb7d..72912362fe 100644
--- a/codex-rs/app-server/tests/suite/auth.rs
+++ b/codex-rs/app-server/tests/suite/auth.rs
@@ -1,5 +1,4 @@
-use std::path::Path;
-
+use anyhow::Result;
 use app_test_support::McpProcess;
 use app_test_support::to_response;
 use codex_app_server_protocol::AuthMode;
@@ -11,6 +10,7 @@ use codex_app_server_protocol::LoginApiKeyParams;
 use codex_app_server_protocol::LoginApiKeyResponse;
 use codex_app_server_protocol::RequestId;
 use pretty_assertions::assert_eq;
+use std::path::Path;
 use tempfile::TempDir;
 use tokio::time::timeout;
 
@@ -71,125 +71,99 @@ forced_login_method = "{forced_method}"
     std::fs::write(config_toml, contents)
 }
 
-async fn login_with_api_key_via_request(mcp: &mut McpProcess, api_key: &str) {
+async fn login_with_api_key_via_request(mcp: &mut McpProcess, api_key: &str) -> Result<()> {
     let request_id = mcp
         .send_login_api_key_request(LoginApiKeyParams {
             api_key: api_key.to_string(),
         })
-        .await
-        .unwrap_or_else(|e| panic!("send loginApiKey: {e}"));
+        .await?;
 
     let resp: JSONRPCResponse = timeout(
         DEFAULT_READ_TIMEOUT,
         mcp.read_stream_until_response_message(RequestId::Integer(request_id)),
     )
-    .await
-    .unwrap_or_else(|e| panic!("loginApiKey timeout: {e}"))
-    .unwrap_or_else(|e| panic!("loginApiKey response: {e}"));
-    let _: LoginApiKeyResponse =
-        to_response(resp).unwrap_or_else(|e| panic!("deserialize login response: {e}"));
+    .await??;
+    let _: LoginApiKeyResponse = to_response(resp)?;
+    Ok(())
 }
 
 #[tokio::test(flavor = "multi_thread", worker_threads = 2)]
-async fn get_auth_status_no_auth() {
-    let codex_home = TempDir::new().unwrap_or_else(|e| panic!("create tempdir: {e}"));
-    create_config_toml(codex_home.path()).unwrap_or_else(|err| panic!("write config.toml: {err}"));
+async fn get_auth_status_no_auth() -> Result<()> {
+    let codex_home = TempDir::new()?;
+    create_config_toml(codex_home.path())?;
 
-    let mut mcp = McpProcess::new_with_env(codex_home.path(), &[("OPENAI_API_KEY", None)])
-        .await
-        .expect("spawn mcp process");
-    timeout(DEFAULT_READ_TIMEOUT, mcp.initialize())
-        .await
-        .expect("init timeout")
-        .expect("init failed");
+    let mut mcp = McpProcess::new_with_env(codex_home.path(), &[("OPENAI_API_KEY", None)]).await?;
+    timeout(DEFAULT_READ_TIMEOUT, mcp.initialize()).await??;
 
     let request_id = mcp
         .send_get_auth_status_request(GetAuthStatusParams {
             include_token: Some(true),
             refresh_token: Some(false),
         })
-        .await
-        .expect("send getAuthStatus");
+        .await?;
 
     let resp: JSONRPCResponse = timeout(
         DEFAULT_READ_TIMEOUT,
         mcp.read_stream_until_response_message(RequestId::Integer(request_id)),
     )
-    .await
-    .expect("getAuthStatus timeout")
-    .expect("getAuthStatus response");
-    let status: GetAuthStatusResponse = to_response(resp).expect("deserialize status");
+    .await??;
+    let status: GetAuthStatusResponse = to_response(resp)?;
     assert_eq!(status.auth_method, None, "expected no auth method");
     assert_eq!(status.auth_token, None, "expected no token");
+    Ok(())
 }
 
 #[tokio::test(flavor = "multi_thread", worker_threads = 2)]
-async fn get_auth_status_with_api_key() {
-    let codex_home = TempDir::new().unwrap_or_else(|e| panic!("create tempdir: {e}"));
-    create_config_toml(codex_home.path()).unwrap_or_else(|err| panic!("write config.toml: {err}"));
+async fn get_auth_status_with_api_key() -> Result<()> {
+    let codex_home = TempDir::new()?;
+    create_config_toml(codex_home.path())?;
 
-    let mut mcp = McpProcess::new(codex_home.path())
-        .await
-        .expect("spawn mcp process");
-    timeout(DEFAULT_READ_TIMEOUT, mcp.initialize())
-        .await
-        .expect("init timeout")
-        .expect("init failed");
+    let mut mcp = McpProcess::new(codex_home.path()).await?;
+    timeout(DEFAULT_READ_TIMEOUT, mcp.initialize()).await??;
 
-    login_with_api_key_via_request(&mut mcp, "sk-test-key").await;
+    login_with_api_key_via_request(&mut mcp, "sk-test-key").await?;
 
     let request_id = mcp
         .send_get_auth_status_request(GetAuthStatusParams {
             include_token: Some(true),
             refresh_token: Some(false),
         })
-        .await
-        .expect("send getAuthStatus");
+        .await?;
 
     let resp: JSONRPCResponse = timeout(
         DEFAULT_READ_TIMEOUT,
         mcp.read_stream_until_response_message(RequestId::Integer(request_id)),
     )
-    .await
-    .expect("getAuthStatus timeout")
-    .expect("getAuthStatus response");
-    let status: GetAuthStatusResponse = to_response(resp).expect("deserialize status");
+    .await??;
+    let status: GetAuthStatusResponse = to_response(resp)?;
     assert_eq!(status.auth_method, Some(AuthMode::ApiKey));
     assert_eq!(status.auth_token, Some("sk-test-key".to_string()));
+    Ok(())
 }
 
 #[tokio::test(flavor = "multi_thread", worker_threads = 2)]
-async fn get_auth_status_with_api_key_when_auth_not_required() {
-    let codex_home = TempDir::new().unwrap_or_else(|e| panic!("create tempdir: {e}"));
-    create_config_toml_custom_provider(codex_home.path(), false)
-        .unwrap_or_else(|err| panic!("write config.toml: {err}"));
+async fn get_auth_status_with_api_key_when_auth_not_required() -> Result<()> {
+    let codex_home = TempDir::new()?;
+    create_config_toml_custom_provider(codex_home.path(), false)?;
 
-    let mut mcp = McpProcess::new(codex_home.path())
-        .await
-        .expect("spawn mcp process");
-    timeout(DEFAULT_READ_TIMEOUT, mcp.initialize())
-        .await
-        .expect("init timeout")
-        .expect("init failed");
+    let mut mcp = McpProcess::new(codex_home.path()).await?;
+    timeout(DEFAULT_READ_TIMEOUT, mcp.initialize()).await??;
 
-    login_with_api_key_via_request(&mut mcp, "sk-test-key").await;
+    login_with_api_key_via_request(&mut mcp, "sk-test-key").await?;
 
     let request_id = mcp
         .send_get_auth_status_request(GetAuthStatusParams {
             include_token: Some(true),
             refresh_token: Some(false),
         })
-        .await
-        .expect("send getAuthStatus");
+        .await?;
 
     let resp: JSONRPCResponse = timeout(
         DEFAULT_READ_TIMEOUT,
         mcp.read_stream_until_response_message(RequestId::Integer(request_id)),
     )
-    .await
-    .expect("getAuthStatus timeout")
-    .expect("getAuthStatus response");
-    let status: GetAuthStatusResponse = to_response(resp).expect("deserialize status");
+    .await??;
+    let status: GetAuthStatusResponse = to_response(resp)?;
     assert_eq!(status.auth_method, None, "expected no auth method");
     assert_eq!(status.auth_token, None, "expected no token");
     assert_eq!(
@@ -197,76 +171,60 @@ async fn get_auth_status_with_api_key_when_auth_not_required() {
         Some(false),
         "requires_openai_auth should be false",
     );
+    Ok(())
 }
 
 #[tokio::test(flavor = "multi_thread", worker_threads = 2)]
-async fn get_auth_status_with_api_key_no_include_token() {
-    let codex_home = TempDir::new().unwrap_or_else(|e| panic!("create tempdir: {e}"));
-    create_config_toml(codex_home.path()).unwrap_or_else(|err| panic!("write config.toml: {err}"));
+async fn get_auth_status_with_api_key_no_include_token() -> Result<()> {
+    let codex_home = TempDir::new()?;
+    create_config_toml(codex_home.path())?;
 
-    let mut mcp = McpProcess::new(codex_home.path())
-        .await
-        .expect("spawn mcp process");
-    timeout(DEFAULT_READ_TIMEOUT, mcp.initialize())
-        .await
-        .expect("init timeout")
-        .expect("init failed");
+    let mut mcp = McpProcess::new(codex_home.path()).await?;
+    timeout(DEFAULT_READ_TIMEOUT, mcp.initialize()).await??;
 
-    login_with_api_key_via_request(&mut mcp, "sk-test-key").await;
+    login_with_api_key_via_request(&mut mcp, "sk-test-key").await?;
 
     // Build params via struct so None field is omitted in wire JSON.
     let params = GetAuthStatusParams {
         include_token: None,
         refresh_token: Some(false),
     };
-    let request_id = mcp
-        .send_get_auth_status_request(params)
-        .await
-        .expect("send getAuthStatus");
+    let request_id = mcp.send_get_auth_status_request(params).await?;
 
     let resp: JSONRPCResponse = timeout(
         DEFAULT_READ_TIMEOUT,
         mcp.read_stream_until_response_message(RequestId::Integer(request_id)),
     )
-    .await
-    .expect("getAuthStatus timeout")
-    .expect("getAuthStatus response");
-    let status: GetAuthStatusResponse = to_response(resp).expect("deserialize status");
+    .await??;
+    let status: GetAuthStatusResponse = to_response(resp)?;
     assert_eq!(status.auth_method, Some(AuthMode::ApiKey));
     assert!(status.auth_token.is_none(), "token must be omitted");
+    Ok(())
 }
 
 #[tokio::test(flavor = "multi_thread", worker_threads = 2)]
-async fn login_api_key_rejected_when_forced_chatgpt() {
-    let codex_home = TempDir::new().unwrap_or_else(|e| panic!("create tempdir: {e}"));
-    create_config_toml_forced_login(codex_home.path(), "chatgpt")
-        .unwrap_or_else(|err| panic!("write config.toml: {err}"));
+async fn login_api_key_rejected_when_forced_chatgpt() -> Result<()> {
+    let codex_home = TempDir::new()?;
+    create_config_toml_forced_login(codex_home.path(), "chatgpt")?;
 
-    let mut mcp = McpProcess::new(codex_home.path())
-        .await
-        .expect("spawn mcp process");
-    timeout(DEFAULT_READ_TIMEOUT, mcp.initialize())
-        .await
-        .expect("init timeout")
-        .expect("init failed");
+    let mut mcp = McpProcess::new(codex_home.path()).await?;
+    timeout(DEFAULT_READ_TIMEOUT, mcp.initialize()).await??;
 
     let request_id = mcp
         .send_login_api_key_request(LoginApiKeyParams {
             api_key: "sk-test-key".to_string(),
         })
-        .await
-        .expect("send loginApiKey");
+        .await?;
 
     let err: JSONRPCError = timeout(
         DEFAULT_READ_TIMEOUT,
         mcp.read_stream_until_error_message(RequestId::Integer(request_id)),
     )
-    .await
-    .expect("loginApiKey error timeout")
-    .expect("loginApiKey error");
+    .await??;
 
     assert_eq!(
         err.error.message,
         "API key login is disabled. Use ChatGPT login instead."
     );
+    Ok(())
 }
diff --git a/codex-rs/app-server/tests/suite/codex_message_processor_flow.rs b/codex-rs/app-server/tests/suite/codex_message_processor_flow.rs
index 5c010bf034..1feda42841 100644
--- a/codex-rs/app-server/tests/suite/codex_message_processor_flow.rs
+++ b/codex-rs/app-server/tests/suite/codex_message_processor_flow.rs
@@ -1,5 +1,4 @@
-use std::path::Path;
-
+use anyhow::Result;
 use app_test_support::McpProcess;
 use app_test_support::create_final_assistant_message_sse_response;
 use app_test_support::create_mock_chat_completions_server;
@@ -32,26 +31,27 @@ use codex_protocol::protocol::Event;
 use codex_protocol::protocol::EventMsg;
 use pretty_assertions::assert_eq;
 use std::env;
+use std::path::Path;
 use tempfile::TempDir;
 use tokio::time::timeout;
 
 const DEFAULT_READ_TIMEOUT: std::time::Duration = std::time::Duration::from_secs(10);
 
 #[tokio::test(flavor = "multi_thread", worker_threads = 4)]
-async fn test_codex_jsonrpc_conversation_flow() {
+async fn test_codex_jsonrpc_conversation_flow() -> Result<()> {
     if env::var(CODEX_SANDBOX_NETWORK_DISABLED_ENV_VAR).is_ok() {
         println!(
             "Skipping test because it cannot execute when network is disabled in a Codex sandbox."
         );
-        return;
+        return Ok(());
     }
 
-    let tmp = TempDir::new().expect("tmp dir");
+    let tmp = TempDir::new()?;
     // Temporary Codex home with config pointing at the mock server.
     let codex_home = tmp.path().join("codex_home");
-    std::fs::create_dir(&codex_home).expect("create codex home dir");
+    std::fs::create_dir(&codex_home)?;
     let working_directory = tmp.path().join("workdir");
-    std::fs::create_dir(&working_directory).expect("create working directory");
+    std::fs::create_dir(&working_directory)?;
 
     // Create a mock model server that immediately ends each turn.
     // Two turns are expected: initial session configure + one user message.
@@ -61,20 +61,15 @@ async fn test_codex_jsonrpc_conversation_flow() {
             Some(&working_directory),
             Some(5000),
             "call1234",
-        )
-        .expect("create shell sse response"),
-        create_final_assistant_message_sse_response("Enjoy your new git repo!")
-            .expect("create final assistant message"),
+        )?,
+        create_final_assistant_message_sse_response("Enjoy your new git repo!")?,
     ];
     let server = create_mock_chat_completions_server(responses).await;
-    create_config_toml(&codex_home, &server.uri()).expect("write config");
+    create_config_toml(&codex_home, &server.uri())?;
 
     // Start MCP server and initialize.
-    let mut mcp = McpProcess::new(&codex_home).await.expect("spawn mcp");
-    timeout(DEFAULT_READ_TIMEOUT, mcp.initialize())
-        .await
-        .expect("init timeout")
-        .expect("init error");
+    let mut mcp = McpProcess::new(&codex_home).await?;
+    timeout(DEFAULT_READ_TIMEOUT, mcp.initialize()).await??;
 
     // 1) newConversation
     let new_conv_id = mcp
@@ -82,17 +77,13 @@ async fn test_codex_jsonrpc_conversation_flow() {
             cwd: Some(working_directory.to_string_lossy().into_owned()),
             ..Default::default()
         })
-        .await
-        .expect("send newConversation");
+        .await?;
     let new_conv_resp: JSONRPCResponse = timeout(
         DEFAULT_READ_TIMEOUT,
         mcp.read_stream_until_response_message(RequestId::Integer(new_conv_id)),
     )
-    .await
-    .expect("newConversation timeout")
-    .expect("newConversation resp");
-    let new_conv_resp = to_response::<NewConversationResponse>(new_conv_resp)
-        .expect("deserialize newConversation response");
+    .await??;
+    let new_conv_resp = to_response::<NewConversationResponse>(new_conv_resp)?;
     let NewConversationResponse {
         conversation_id,
         model,
@@ -107,18 +98,14 @@ async fn test_codex_jsonrpc_conversation_flow() {
             conversation_id,
             experimental_raw_events: false,
         })
-        .await
-        .expect("send addConversationListener");
+        .await?;
     let add_listener_resp: JSONRPCResponse = timeout(
         DEFAULT_READ_TIMEOUT,
         mcp.read_stream_until_response_message(RequestId::Integer(add_listener_id)),
     )
-    .await
-    .expect("addConversationListener timeout")
-    .expect("addConversationListener resp");
+    .await??;
     let AddConversationSubscriptionResponse { subscription_id } =
-        to_response::<AddConversationSubscriptionResponse>(add_listener_resp)
-            .expect("deserialize addConversationListener response");
+        to_response::<AddConversationSubscriptionResponse>(add_listener_resp)?;
 
     // 3) sendUserMessage (should trigger notifications; we only validate an OK response)
     let send_user_id = mcp
@@ -128,17 +115,13 @@ async fn test_codex_jsonrpc_conversation_flow() {
                 text: "text".to_string(),
             }],
         })
-        .await
-        .expect("send sendUserMessage");
+        .await?;
     let send_user_resp: JSONRPCResponse = timeout(
         DEFAULT_READ_TIMEOUT,
         mcp.read_stream_until_response_message(RequestId::Integer(send_user_id)),
     )
-    .await
-    .expect("sendUserMessage timeout")
-    .expect("sendUserMessage resp");
-    let SendUserMessageResponse {} = to_response::<SendUserMessageResponse>(send_user_resp)
-        .expect("deserialize sendUserMessage response");
+    .await??;
+    let SendUserMessageResponse {} = to_response::<SendUserMessageResponse>(send_user_resp)?;
 
     // Verify the task_finished notification is received.
     // Note this also ensures that the final request to the server was made.
@@ -146,9 +129,7 @@ async fn test_codex_jsonrpc_conversation_flow() {
         DEFAULT_READ_TIMEOUT,
         mcp.read_stream_until_notification_message("codex/event/task_complete"),
     )
-    .await
-    .expect("task_finished_notification timeout")
-    .expect("task_finished_notification resp");
+    .await??;
     let serde_json::Value::Object(map) = task_finished_notification
         .params
         .expect("notification should have params")
@@ -166,33 +147,31 @@ async fn test_codex_jsonrpc_conversation_flow() {
         .send_remove_conversation_listener_request(RemoveConversationListenerParams {
             subscription_id,
         })
-        .await
-        .expect("send removeConversationListener");
+        .await?;
     let remove_listener_resp: JSONRPCResponse = timeout(
         DEFAULT_READ_TIMEOUT,
         mcp.read_stream_until_response_message(RequestId::Integer(remove_listener_id)),
     )
-    .await
-    .expect("removeConversationListener timeout")
-    .expect("removeConversationListener resp");
-    let RemoveConversationSubscriptionResponse {} =
-        to_response(remove_listener_resp).expect("deserialize removeConversationListener response");
+    .await??;
+    let RemoveConversationSubscriptionResponse {} = to_response(remove_listener_resp)?;
+
+    Ok(())
 }
 
 #[tokio::test(flavor = "multi_thread", worker_threads = 4)]
-async fn test_send_user_turn_changes_approval_policy_behavior() {
+async fn test_send_user_turn_changes_approval_policy_behavior() -> Result<()> {
     if env::var(CODEX_SANDBOX_NETWORK_DISABLED_ENV_VAR).is_ok() {
         println!(
             "Skipping test because it cannot execute when network is disabled in a Codex sandbox."
         );
-        return;
+        return Ok(());
     }
 
-    let tmp = TempDir::new().expect("tmp dir");
+    let tmp = TempDir::new()?;
     let codex_home = tmp.path().join("codex_home");
-    std::fs::create_dir(&codex_home).expect("create codex home dir");
+    std::fs::create_dir(&codex_home)?;
     let working_directory = tmp.path().join("workdir");
-    std::fs::create_dir(&working_directory).expect("create working directory");
+    std::fs::create_dir(&working_directory)?;
 
     // Mock server will request a python shell call for the first and second turn, then finish.
     let responses = vec![
@@ -205,10 +184,8 @@ async fn test_send_user_turn_changes_approval_policy_behavior() {
             Some(&working_directory),
             Some(5000),
             "call1",
-        )
-        .expect("create first shell sse response"),
-        create_final_assistant_message_sse_response("done 1")
-            .expect("create final assistant message 1"),
+        )?,
+        create_final_assistant_message_sse_response("done 1")?,
         create_shell_sse_response(
             vec![
                 "python3".to_string(),
@@ -218,20 +195,15 @@ async fn test_send_user_turn_changes_approval_policy_behavior() {
             Some(&working_directory),
             Some(5000),
             "call2",
-        )
-        .expect("create second shell sse response"),
-        create_final_assistant_message_sse_response("done 2")
-            .expect("create final assistant message 2"),
+        )?,
+        create_final_assistant_message_sse_response("done 2")?,
     ];
     let server = create_mock_chat_completions_server(responses).await;
-    create_config_toml(&codex_home, &server.uri()).expect("write config");
+    create_config_toml(&codex_home, &server.uri())?;
 
     // Start MCP server and initialize.
-    let mut mcp = McpProcess::new(&codex_home).await.expect("spawn mcp");
-    timeout(DEFAULT_READ_TIMEOUT, mcp.initialize())
-        .await
-        .expect("init timeout")
-        .expect("init error");
+    let mut mcp = McpProcess::new(&codex_home).await?;
+    timeout(DEFAULT_READ_TIMEOUT, mcp.initialize()).await??;
 
     // 1) Start conversation with approval_policy=untrusted
     let new_conv_id = mcp
@@ -239,19 +211,15 @@ async fn test_send_user_turn_changes_approval_policy_behavior() {
             cwd: Some(working_directory.to_string_lossy().into_owned()),
             ..Default::default()
         })
-        .await
-        .expect("send newConversation");
+        .await?;
     let new_conv_resp: JSONRPCResponse = timeout(
         DEFAULT_READ_TIMEOUT,
         mcp.read_stream_until_response_message(RequestId::Integer(new_conv_id)),
     )
-    .await
-    .expect("newConversation timeout")
-    .expect("newConversation resp");
+    .await??;
     let NewConversationResponse {
         conversation_id, ..
-    } = to_response::<NewConversationResponse>(new_conv_resp)
-        .expect("deserialize newConversation response");
+    } = to_response::<NewConversationResponse>(new_conv_resp)?;
 
     // 2) addConversationListener
     let add_listener_id = mcp
@@ -259,19 +227,14 @@ async fn test_send_user_turn_changes_approval_policy_behavior() {
             conversation_id,
             experimental_raw_events: false,
         })
-        .await
-        .expect("send addConversationListener");
-    let _: AddConversationSubscriptionResponse =
-        to_response::<AddConversationSubscriptionResponse>(
-            timeout(
-                DEFAULT_READ_TIMEOUT,
-                mcp.read_stream_until_response_message(RequestId::Integer(add_listener_id)),
-            )
-            .await
-            .expect("addConversationListener timeout")
-            .expect("addConversationListener resp"),
+        .await?;
+    let _: AddConversationSubscriptionResponse = to_response::<AddConversationSubscriptionResponse>(
+        timeout(
+            DEFAULT_READ_TIMEOUT,
+            mcp.read_stream_until_response_message(RequestId::Integer(add_listener_id)),
         )
-        .expect("deserialize addConversationListener response");
+        .await??,
+    )?;
 
     // 3) sendUserMessage triggers a shell call; approval policy is Untrusted so we should get an elicitation
     let send_user_id = mcp
@@ -281,27 +244,21 @@ async fn test_send_user_turn_changes_approval_policy_behavior() {
                 text: "run python".to_string(),
             }],
         })
-        .await
-        .expect("send sendUserMessage");
+        .await?;
     let _send_user_resp: SendUserMessageResponse = to_response::<SendUserMessageResponse>(
         timeout(
             DEFAULT_READ_TIMEOUT,
             mcp.read_stream_until_response_message(RequestId::Integer(send_user_id)),
         )
-        .await
-        .expect("sendUserMessage timeout")
-        .expect("sendUserMessage resp"),
-    )
-    .expect("deserialize sendUserMessage response");
+        .await??,
+    )?;
 
     // Expect an ExecCommandApproval request (elicitation)
     let request = timeout(
         DEFAULT_READ_TIMEOUT,
         mcp.read_stream_until_request_message(),
     )
-    .await
-    .expect("waiting for exec approval request timeout")
-    .expect("exec approval request");
+    .await??;
     let ServerRequest::ExecCommandApproval { request_id, params } = request else {
         panic!("expected ExecCommandApproval request, got: {request:?}");
     };
@@ -330,17 +287,14 @@ async fn test_send_user_turn_changes_approval_policy_behavior() {
         request_id,
         serde_json::json!({ "decision": codex_core::protocol::ReviewDecision::Approved }),
     )
-    .await
-    .expect("send approval response");
+    .await?;
 
     // Wait for first TaskComplete
     let _ = timeout(
         DEFAULT_READ_TIMEOUT,
         mcp.read_stream_until_notification_message("codex/event/task_complete"),
     )
-    .await
-    .expect("task_complete 1 timeout")
-    .expect("task_complete 1 notification");
+    .await??;
 
     // 4) sendUserTurn with approval_policy=never should run without elicitation
     let send_turn_id = mcp
@@ -356,19 +310,15 @@ async fn test_send_user_turn_changes_approval_policy_behavior() {
             effort: Some(ReasoningEffort::Medium),
             summary: ReasoningSummary::Auto,
         })
-        .await
-        .expect("send sendUserTurn");
+        .await?;
     // Acknowledge sendUserTurn
     let _send_turn_resp: SendUserTurnResponse = to_response::<SendUserTurnResponse>(
         timeout(
             DEFAULT_READ_TIMEOUT,
             mcp.read_stream_until_response_message(RequestId::Integer(send_turn_id)),
         )
-        .await
-        .expect("sendUserTurn timeout")
-        .expect("sendUserTurn resp"),
-    )
-    .expect("deserialize sendUserTurn response");
+        .await??,
+    )?;
 
     // Ensure we do NOT receive an ExecCommandApproval request before the task completes.
     // If any Request is seen while waiting for task_complete, the helper will error and the test fails.
@@ -376,31 +326,31 @@ async fn test_send_user_turn_changes_approval_policy_behavior() {
         DEFAULT_READ_TIMEOUT,
         mcp.read_stream_until_notification_message("codex/event/task_complete"),
     )
-    .await
-    .expect("task_complete 2 timeout")
-    .expect("task_complete 2 notification");
+    .await??;
+
+    Ok(())
 }
 
 // Helper: minimal config.toml pointing at mock provider.
 
 #[tokio::test(flavor = "multi_thread", worker_threads = 4)]
-async fn test_send_user_turn_updates_sandbox_and_cwd_between_turns() {
+async fn test_send_user_turn_updates_sandbox_and_cwd_between_turns() -> Result<()> {
     if env::var(CODEX_SANDBOX_NETWORK_DISABLED_ENV_VAR).is_ok() {
         println!(
             "Skipping test because it cannot execute when network is disabled in a Codex sandbox."
         );
-        return;
+        return Ok(());
     }
 
-    let tmp = TempDir::new().expect("tmp dir");
+    let tmp = TempDir::new()?;
     let codex_home = tmp.path().join("codex_home");
-    std::fs::create_dir(&codex_home).expect("create codex home dir");
+    std::fs::create_dir(&codex_home)?;
     let workspace_root = tmp.path().join("workspace");
-    std::fs::create_dir(&workspace_root).expect("create workspace root");
+    std::fs::create_dir(&workspace_root)?;
     let first_cwd = workspace_root.join("turn1");
     let second_cwd = workspace_root.join("turn2");
-    std::fs::create_dir(&first_cwd).expect("create first cwd");
-    std::fs::create_dir(&second_cwd).expect("create second cwd");
+    std::fs::create_dir(&first_cwd)?;
+    std::fs::create_dir(&second_cwd)?;
 
     let responses = vec![
         create_shell_sse_response(
@@ -412,10 +362,8 @@ async fn test_send_user_turn_updates_sandbox_and_cwd_between_turns() {
             None,
             Some(5000),
             "call-first",
-        )
-        .expect("create first shell response"),
-        create_final_assistant_message_sse_response("done first")
-            .expect("create first final assistant message"),
+        )?,
+        create_final_assistant_message_sse_response("done first")?,
         create_shell_sse_response(
             vec![
                 "bash".to_string(),
@@ -425,21 +373,14 @@ async fn test_send_user_turn_updates_sandbox_and_cwd_between_turns() {
             None,
             Some(5000),
             "call-second",
-        )
-        .expect("create second shell response"),
-        create_final_assistant_message_sse_response("done second")
-            .expect("create second final assistant message"),
+        )?,
+        create_final_assistant_message_sse_response("done second")?,
     ];
     let server = create_mock_chat_completions_server(responses).await;
-    create_config_toml(&codex_home, &server.uri()).expect("write config");
+    create_config_toml(&codex_home, &server.uri())?;
 
-    let mut mcp = McpProcess::new(&codex_home)
-        .await
-        .expect("spawn mcp process");
-    timeout(DEFAULT_READ_TIMEOUT, mcp.initialize())
-        .await
-        .expect("init timeout")
-        .expect("init failed");
+    let mut mcp = McpProcess::new(&codex_home).await?;
+    timeout(DEFAULT_READ_TIMEOUT, mcp.initialize()).await??;
 
     let new_conv_id = mcp
         .send_new_conversation_request(NewConversationParams {
@@ -448,36 +389,29 @@ async fn test_send_user_turn_updates_sandbox_and_cwd_between_turns() {
             sandbox: Some(SandboxMode::WorkspaceWrite),
             ..Default::default()
         })
-        .await
-        .expect("send newConversation");
+        .await?;
     let new_conv_resp: JSONRPCResponse = timeout(
         DEFAULT_READ_TIMEOUT,
         mcp.read_stream_until_response_message(RequestId::Integer(new_conv_id)),
     )
-    .await
-    .expect("newConversation timeout")
-    .expect("newConversation resp");
+    .await??;
     let NewConversationResponse {
         conversation_id,
         model,
         ..
-    } = to_response::<NewConversationResponse>(new_conv_resp)
-        .expect("deserialize newConversation response");
+    } = to_response::<NewConversationResponse>(new_conv_resp)?;
 
     let add_listener_id = mcp
         .send_add_conversation_listener_request(AddConversationListenerParams {
             conversation_id,
             experimental_raw_events: false,
         })
-        .await
-        .expect("send addConversationListener");
+        .await?;
     timeout(
         DEFAULT_READ_TIMEOUT,
         mcp.read_stream_until_response_message(RequestId::Integer(add_listener_id)),
     )
-    .await
-    .expect("addConversationListener timeout")
-    .expect("addConversationListener resp");
+    .await??;
 
     let first_turn_id = mcp
         .send_send_user_turn_request(SendUserTurnParams {
@@ -497,22 +431,17 @@ async fn test_send_user_turn_updates_sandbox_and_cwd_between_turns() {
             effort: Some(ReasoningEffort::Medium),
             summary: ReasoningSummary::Auto,
         })
-        .await
-        .expect("send first sendUserTurn");
+        .await?;
     timeout(
         DEFAULT_READ_TIMEOUT,
         mcp.read_stream_until_response_message(RequestId::Integer(first_turn_id)),
     )
-    .await
-    .expect("sendUserTurn 1 timeout")
-    .expect("sendUserTurn 1 resp");
+    .await??;
     timeout(
         DEFAULT_READ_TIMEOUT,
         mcp.read_stream_until_notification_message("codex/event/task_complete"),
     )
-    .await
-    .expect("task_complete 1 timeout")
-    .expect("task_complete 1 notification");
+    .await??;
 
     let second_turn_id = mcp
         .send_send_user_turn_request(SendUserTurnParams {
@@ -527,23 +456,18 @@ async fn test_send_user_turn_updates_sandbox_and_cwd_between_turns() {
             effort: Some(ReasoningEffort::Medium),
             summary: ReasoningSummary::Auto,
         })
-        .await
-        .expect("send second sendUserTurn");
+        .await?;
     timeout(
         DEFAULT_READ_TIMEOUT,
         mcp.read_stream_until_response_message(RequestId::Integer(second_turn_id)),
     )
-    .await
-    .expect("sendUserTurn 2 timeout")
-    .expect("sendUserTurn 2 resp");
+    .await??;
 
     let exec_begin_notification = timeout(
         DEFAULT_READ_TIMEOUT,
         mcp.read_stream_until_notification_message("codex/event/exec_command_begin"),
     )
-    .await
-    .expect("exec_command_begin timeout")
-    .expect("exec_command_begin notification");
+    .await??;
     let params = exec_begin_notification
         .params
         .clone()
@@ -571,9 +495,9 @@ async fn test_send_user_turn_updates_sandbox_and_cwd_between_turns() {
         DEFAULT_READ_TIMEOUT,
         mcp.read_stream_until_notification_message("codex/event/task_complete"),
     )
-    .await
-    .expect("task_complete 2 timeout")
-    .expect("task_complete 2 notification");
+    .await??;
+
+    Ok(())
 }
 
 fn create_config_toml(codex_home: &Path, server_uri: &str) -> std::io::Result<()> {
diff --git a/codex-rs/app-server/tests/suite/config.rs b/codex-rs/app-server/tests/suite/config.rs
index 2809d68ca4..227d30bbe3 100644
--- a/codex-rs/app-server/tests/suite/config.rs
+++ b/codex-rs/app-server/tests/suite/config.rs
@@ -1,6 +1,4 @@
-use std::collections::HashMap;
-use std::path::Path;
-
+use anyhow::Result;
 use app_test_support::McpProcess;
 use app_test_support::to_response;
 use codex_app_server_protocol::GetUserSavedConfigResponse;
@@ -17,6 +15,8 @@ use codex_protocol::config_types::ReasoningSummary;
 use codex_protocol::config_types::SandboxMode;
 use codex_protocol::config_types::Verbosity;
 use pretty_assertions::assert_eq;
+use std::collections::HashMap;
+use std::path::Path;
 use tempfile::TempDir;
 use tokio::time::timeout;
 
@@ -60,31 +60,21 @@ chatgpt_base_url = "https://api.chatgpt.com"
 }
 
 #[tokio::test(flavor = "multi_thread", worker_threads = 4)]
-async fn get_config_toml_parses_all_fields() {
-    let codex_home = TempDir::new().unwrap_or_else(|e| panic!("create tempdir: {e}"));
-    create_config_toml(codex_home.path()).expect("write config.toml");
+async fn get_config_toml_parses_all_fields() -> Result<()> {
+    let codex_home = TempDir::new()?;
+    create_config_toml(codex_home.path())?;
 
-    let mut mcp = McpProcess::new(codex_home.path())
-        .await
-        .expect("spawn mcp process");
-    timeout(DEFAULT_READ_TIMEOUT, mcp.initialize())
-        .await
-        .expect("init timeout")
-        .expect("init failed");
+    let mut mcp = McpProcess::new(codex_home.path()).await?;
+    timeout(DEFAULT_READ_TIMEOUT, mcp.initialize()).await??;
 
-    let request_id = mcp
-        .send_get_user_saved_config_request()
-        .await
-        .expect("send getUserSavedConfig");
+    let request_id = mcp.send_get_user_saved_config_request().await?;
     let resp: JSONRPCResponse = timeout(
         DEFAULT_READ_TIMEOUT,
         mcp.read_stream_until_response_message(RequestId::Integer(request_id)),
     )
-    .await
-    .expect("getUserSavedConfig timeout")
-    .expect("getUserSavedConfig response");
+    .await??;
 
-    let config: GetUserSavedConfigResponse = to_response(resp).expect("deserialize config");
+    let config: GetUserSavedConfigResponse = to_response(resp)?;
     let expected = GetUserSavedConfigResponse {
         config: UserSavedConfig {
             approval_policy: Some(AskForApproval::OnRequest),
@@ -122,33 +112,24 @@ async fn get_config_toml_parses_all_fields() {
     };
 
     assert_eq!(config, expected);
+    Ok(())
 }
 
 #[tokio::test(flavor = "multi_thread", worker_threads = 2)]
-async fn get_config_toml_empty() {
-    let codex_home = TempDir::new().unwrap_or_else(|e| panic!("create tempdir: {e}"));
+async fn get_config_toml_empty() -> Result<()> {
+    let codex_home = TempDir::new()?;
 
-    let mut mcp = McpProcess::new(codex_home.path())
-        .await
-        .expect("spawn mcp process");
-    timeout(DEFAULT_READ_TIMEOUT, mcp.initialize())
-        .await
-        .expect("init timeout")
-        .expect("init failed");
+    let mut mcp = McpProcess::new(codex_home.path()).await?;
+    timeout(DEFAULT_READ_TIMEOUT, mcp.initialize()).await??;
 
-    let request_id = mcp
-        .send_get_user_saved_config_request()
-        .await
-        .expect("send getUserSavedConfig");
+    let request_id = mcp.send_get_user_saved_config_request().await?;
     let resp: JSONRPCResponse = timeout(
         DEFAULT_READ_TIMEOUT,
         mcp.read_stream_until_response_message(RequestId::Integer(request_id)),
     )
-    .await
-    .expect("getUserSavedConfig timeout")
-    .expect("getUserSavedConfig response");
+    .await??;
 
-    let config: GetUserSavedConfigResponse = to_response(resp).expect("deserialize config");
+    let config: GetUserSavedConfigResponse = to_response(resp)?;
     let expected = GetUserSavedConfigResponse {
         config: UserSavedConfig {
             approval_policy: None,
@@ -167,4 +148,5 @@ async fn get_config_toml_empty() {
     };
 
     assert_eq!(config, expected);
+    Ok(())
 }
diff --git a/codex-rs/app-server/tests/suite/create_conversation.rs b/codex-rs/app-server/tests/suite/create_conversation.rs
index f507baa30d..7788b8f382 100644
--- a/codex-rs/app-server/tests/suite/create_conversation.rs
+++ b/codex-rs/app-server/tests/suite/create_conversation.rs
@@ -1,5 +1,4 @@
-use std::path::Path;
-
+use anyhow::Result;
 use app_test_support::McpProcess;
 use app_test_support::create_final_assistant_message_sse_response;
 use app_test_support::create_mock_chat_completions_server;
@@ -15,31 +14,25 @@ use codex_app_server_protocol::SendUserMessageParams;
 use codex_app_server_protocol::SendUserMessageResponse;
 use pretty_assertions::assert_eq;
 use serde_json::json;
+use std::path::Path;
 use tempfile::TempDir;
 use tokio::time::timeout;
 
 const DEFAULT_READ_TIMEOUT: std::time::Duration = std::time::Duration::from_secs(10);
 
 #[tokio::test(flavor = "multi_thread", worker_threads = 2)]
-async fn test_conversation_create_and_send_message_ok() {
+async fn test_conversation_create_and_send_message_ok() -> Result<()> {
     // Mock server – we won't strictly rely on it, but provide one to satisfy any model wiring.
-    let responses = vec![
-        create_final_assistant_message_sse_response("Done").expect("build mock assistant message"),
-    ];
+    let responses = vec![create_final_assistant_message_sse_response("Done")?];
     let server = create_mock_chat_completions_server(responses).await;
 
     // Temporary Codex home with config pointing at the mock server.
-    let codex_home = TempDir::new().expect("create temp dir");
-    create_config_toml(codex_home.path(), &server.uri()).expect("write config.toml");
+    let codex_home = TempDir::new()?;
+    create_config_toml(codex_home.path(), &server.uri())?;
 
     // Start MCP server process and initialize.
-    let mut mcp = McpProcess::new(codex_home.path())
-        .await
-        .expect("spawn mcp process");
-    timeout(DEFAULT_READ_TIMEOUT, mcp.initialize())
-        .await
-        .expect("init timeout")
-        .expect("init failed");
+    let mut mcp = McpProcess::new(codex_home.path()).await?;
+    timeout(DEFAULT_READ_TIMEOUT, mcp.initialize()).await??;
 
     // Create a conversation via the new JSON-RPC API.
     let new_conv_id = mcp
@@ -47,22 +40,18 @@ async fn test_conversation_create_and_send_message_ok() {
             model: Some("o3".to_string()),
             ..Default::default()
         })
-        .await
-        .expect("send newConversation");
+        .await?;
     let new_conv_resp: JSONRPCResponse = timeout(
         DEFAULT_READ_TIMEOUT,
         mcp.read_stream_until_response_message(RequestId::Integer(new_conv_id)),
     )
-    .await
-    .expect("newConversation timeout")
-    .expect("newConversation resp");
+    .await??;
     let NewConversationResponse {
         conversation_id,
         model,
         reasoning_effort: _,
         rollout_path: _,
-    } = to_response::<NewConversationResponse>(new_conv_resp)
-        .expect("deserialize newConversation response");
+    } = to_response::<NewConversationResponse>(new_conv_resp)?;
     assert_eq!(model, "o3");
 
     // Add a listener so we receive notifications for this conversation (not strictly required for this test).
@@ -71,19 +60,15 @@ async fn test_conversation_create_and_send_message_ok() {
             conversation_id,
             experimental_raw_events: false,
         })
-        .await
-        .expect("send addConversationListener");
+        .await?;
     let _sub: AddConversationSubscriptionResponse =
         to_response::<AddConversationSubscriptionResponse>(
             timeout(
                 DEFAULT_READ_TIMEOUT,
                 mcp.read_stream_until_response_message(RequestId::Integer(add_listener_id)),
             )
-            .await
-            .expect("addConversationListener timeout")
-            .expect("addConversationListener resp"),
-        )
-        .expect("deserialize addConversationListener response");
+            .await??,
+        )?;
 
     // Now send a user message via the wire API and expect an OK (empty object) result.
     let send_id = mcp
@@ -93,36 +78,32 @@ async fn test_conversation_create_and_send_message_ok() {
                 text: "Hello".to_string(),
             }],
         })
-        .await
-        .expect("send sendUserMessage");
+        .await?;
     let send_resp: JSONRPCResponse = timeout(
         DEFAULT_READ_TIMEOUT,
         mcp.read_stream_until_response_message(RequestId::Integer(send_id)),
     )
-    .await
-    .expect("sendUserMessage timeout")
-    .expect("sendUserMessage resp");
-    let _ok: SendUserMessageResponse = to_response::<SendUserMessageResponse>(send_resp)
-        .expect("deserialize sendUserMessage response");
+    .await??;
+    let _ok: SendUserMessageResponse = to_response::<SendUserMessageResponse>(send_resp)?;
 
     // avoid race condition by waiting for the mock server to receive the chat.completions request
     let deadline = std::time::Instant::now() + DEFAULT_READ_TIMEOUT;
-    loop {
+    let requests = loop {
         let requests = server.received_requests().await.unwrap_or_default();
         if !requests.is_empty() {
-            break;
+            break requests;
         }
         if std::time::Instant::now() >= deadline {
             panic!("mock server did not receive the chat.completions request in time");
         }
         tokio::time::sleep(std::time::Duration::from_millis(10)).await;
-    }
+    };
 
     // Verify the outbound request body matches expectations for Chat Completions.
-    let request = &server.received_requests().await.unwrap()[0];
-    let body = request
-        .body_json::<serde_json::Value>()
-        .expect("parse request body as JSON");
+    let request = requests
+        .first()
+        .expect("mock server should have received at least one request");
+    let body = request.body_json::<serde_json::Value>()?;
     assert_eq!(body["model"], json!("o3"));
     assert!(body["stream"].as_bool().unwrap_or(false));
     let messages = body["messages"]
@@ -133,6 +114,7 @@ async fn test_conversation_create_and_send_message_ok() {
     assert_eq!(last["content"], json!("Hello"));
 
     drop(server);
+    Ok(())
 }
 
 // Helper to create a config.toml pointing at the mock model server.
diff --git a/codex-rs/app-server/tests/suite/fuzzy_file_search.rs b/codex-rs/app-server/tests/suite/fuzzy_file_search.rs
index a2bc974ad0..9c95e3de34 100644
--- a/codex-rs/app-server/tests/suite/fuzzy_file_search.rs
+++ b/codex-rs/app-server/tests/suite/fuzzy_file_search.rs
@@ -1,5 +1,5 @@
-use anyhow::Context;
 use anyhow::Result;
+use anyhow::anyhow;
 use app_test_support::McpProcess;
 use codex_app_server_protocol::JSONRPCResponse;
 use codex_app_server_protocol::RequestId;
@@ -13,48 +13,39 @@ const DEFAULT_READ_TIMEOUT: std::time::Duration = std::time::Duration::from_secs
 #[tokio::test(flavor = "multi_thread", worker_threads = 2)]
 async fn test_fuzzy_file_search_sorts_and_includes_indices() -> Result<()> {
     // Prepare a temporary Codex home and a separate root with test files.
-    let codex_home = TempDir::new().context("create temp codex home")?;
-    let root = TempDir::new().context("create temp search root")?;
+    let codex_home = TempDir::new()?;
+    let root = TempDir::new()?;
 
     // Create files designed to have deterministic ordering for query "abe".
-    std::fs::write(root.path().join("abc"), "x").context("write file abc")?;
-    std::fs::write(root.path().join("abcde"), "x").context("write file abcde")?;
-    std::fs::write(root.path().join("abexy"), "x").context("write file abexy")?;
-    std::fs::write(root.path().join("zzz.txt"), "x").context("write file zzz")?;
+    std::fs::write(root.path().join("abc"), "x")?;
+    std::fs::write(root.path().join("abcde"), "x")?;
+    std::fs::write(root.path().join("abexy"), "x")?;
+    std::fs::write(root.path().join("zzz.txt"), "x")?;
     let sub_dir = root.path().join("sub");
-    std::fs::create_dir_all(&sub_dir).context("create sub dir")?;
+    std::fs::create_dir_all(&sub_dir)?;
     let sub_abce_path = sub_dir.join("abce");
-    std::fs::write(&sub_abce_path, "x").context("write file sub/abce")?;
+    std::fs::write(&sub_abce_path, "x")?;
     let sub_abce_rel = sub_abce_path
-        .strip_prefix(root.path())
-        .context("strip root prefix from sub/abce")?
+        .strip_prefix(root.path())?
         .to_string_lossy()
         .to_string();
 
     // Start MCP server and initialize.
-    let mut mcp = McpProcess::new(codex_home.path())
-        .await
-        .context("spawn mcp")?;
-    timeout(DEFAULT_READ_TIMEOUT, mcp.initialize())
-        .await
-        .context("init timeout")?
-        .context("init failed")?;
+    let mut mcp = McpProcess::new(codex_home.path()).await?;
+    timeout(DEFAULT_READ_TIMEOUT, mcp.initialize()).await??;
 
     let root_path = root.path().to_string_lossy().to_string();
     // Send fuzzyFileSearch request.
     let request_id = mcp
         .send_fuzzy_file_search_request("abe", vec![root_path.clone()], None)
-        .await
-        .context("send fuzzyFileSearch")?;
+        .await?;
 
     // Read response and verify shape and ordering.
     let resp: JSONRPCResponse = timeout(
         DEFAULT_READ_TIMEOUT,
         mcp.read_stream_until_response_message(RequestId::Integer(request_id)),
     )
-    .await
-    .context("fuzzyFileSearch timeout")?
-    .context("fuzzyFileSearch resp")?;
+    .await??;
 
     let value = resp.result;
     // The path separator on Windows affects the score.
@@ -94,24 +85,18 @@ async fn test_fuzzy_file_search_sorts_and_includes_indices() -> Result<()> {
 
 #[tokio::test(flavor = "multi_thread", worker_threads = 2)]
 async fn test_fuzzy_file_search_accepts_cancellation_token() -> Result<()> {
-    let codex_home = TempDir::new().context("create temp codex home")?;
-    let root = TempDir::new().context("create temp search root")?;
+    let codex_home = TempDir::new()?;
+    let root = TempDir::new()?;
 
-    std::fs::write(root.path().join("alpha.txt"), "contents").context("write alpha")?;
+    std::fs::write(root.path().join("alpha.txt"), "contents")?;
 
-    let mut mcp = McpProcess::new(codex_home.path())
-        .await
-        .context("spawn mcp")?;
-    timeout(DEFAULT_READ_TIMEOUT, mcp.initialize())
-        .await
-        .context("init timeout")?
-        .context("init failed")?;
+    let mut mcp = McpProcess::new(codex_home.path()).await?;
+    timeout(DEFAULT_READ_TIMEOUT, mcp.initialize()).await??;
 
     let root_path = root.path().to_string_lossy().to_string();
     let request_id = mcp
         .send_fuzzy_file_search_request("alp", vec![root_path.clone()], None)
-        .await
-        .context("send fuzzyFileSearch")?;
+        .await?;
 
     let request_id_2 = mcp
         .send_fuzzy_file_search_request(
@@ -119,23 +104,20 @@ async fn test_fuzzy_file_search_accepts_cancellation_token() -> Result<()> {
             vec![root_path.clone()],
             Some(request_id.to_string()),
         )
-        .await
-        .context("send fuzzyFileSearch")?;
+        .await?;
 
     let resp: JSONRPCResponse = timeout(
         DEFAULT_READ_TIMEOUT,
         mcp.read_stream_until_response_message(RequestId::Integer(request_id_2)),
     )
-    .await
-    .context("fuzzyFileSearch timeout")?
-    .context("fuzzyFileSearch resp")?;
+    .await??;
 
     let files = resp
         .result
         .get("files")
-        .context("files key missing")?
+        .ok_or_else(|| anyhow!("files key missing"))?
         .as_array()
-        .context("files not array")?
+        .ok_or_else(|| anyhow!("files not array"))?
         .clone();
 
     assert_eq!(files.len(), 1);
diff --git a/codex-rs/app-server/tests/suite/list_resume.rs b/codex-rs/app-server/tests/suite/list_resume.rs
index 85416f60bd..44578e04fa 100644
--- a/codex-rs/app-server/tests/suite/list_resume.rs
+++ b/codex-rs/app-server/tests/suite/list_resume.rs
@@ -1,6 +1,4 @@
-use std::fs;
-use std::path::Path;
-
+use anyhow::Result;
 use app_test_support::McpProcess;
 use app_test_support::to_response;
 use codex_app_server_protocol::JSONRPCNotification;
@@ -15,6 +13,8 @@ use codex_app_server_protocol::ServerNotification;
 use codex_app_server_protocol::SessionConfiguredNotification;
 use pretty_assertions::assert_eq;
 use serde_json::json;
+use std::fs;
+use std::path::Path;
 use tempfile::TempDir;
 use tokio::time::timeout;
 use uuid::Uuid;
@@ -22,38 +22,33 @@ use uuid::Uuid;
 const DEFAULT_READ_TIMEOUT: std::time::Duration = std::time::Duration::from_secs(10);
 
 #[tokio::test(flavor = "multi_thread", worker_threads = 2)]
-async fn test_list_and_resume_conversations() {
+async fn test_list_and_resume_conversations() -> Result<()> {
     // Prepare a temporary CODEX_HOME with a few fake rollout files.
-    let codex_home = TempDir::new().expect("create temp dir");
+    let codex_home = TempDir::new()?;
     create_fake_rollout(
         codex_home.path(),
         "2025-01-02T12-00-00",
         "2025-01-02T12:00:00Z",
         "Hello A",
         Some("openai"),
-    );
+    )?;
     create_fake_rollout(
         codex_home.path(),
         "2025-01-01T13-00-00",
         "2025-01-01T13:00:00Z",
         "Hello B",
         Some("openai"),
-    );
+    )?;
     create_fake_rollout(
         codex_home.path(),
         "2025-01-01T12-00-00",
         "2025-01-01T12:00:00Z",
         "Hello C",
         None,
-    );
+    )?;
 
-    let mut mcp = McpProcess::new(codex_home.path())
-        .await
-        .expect("spawn mcp process");
-    timeout(DEFAULT_READ_TIMEOUT, mcp.initialize())
-        .await
-        .expect("init timeout")
-        .expect("init failed");
+    let mut mcp = McpProcess::new(codex_home.path()).await?;
+    timeout(DEFAULT_READ_TIMEOUT, mcp.initialize()).await??;
 
     // Request first page with size 2
     let req_id = mcp
@@ -62,17 +57,14 @@ async fn test_list_and_resume_conversations() {
             cursor: None,
             model_providers: None,
         })
-        .await
-        .expect("send listConversations");
+        .await?;
     let resp: JSONRPCResponse = timeout(
         DEFAULT_READ_TIMEOUT,
         mcp.read_stream_until_response_message(RequestId::Integer(req_id)),
     )
-    .await
-    .expect("listConversations timeout")
-    .expect("listConversations resp");
+    .await??;
     let ListConversationsResponse { items, next_cursor } =
-        to_response::<ListConversationsResponse>(resp).expect("deserialize response");
+        to_response::<ListConversationsResponse>(resp)?;
 
     assert_eq!(items.len(), 2);
     // Newest first; preview text should match
@@ -90,20 +82,17 @@ async fn test_list_and_resume_conversations() {
             cursor: next_cursor,
             model_providers: None,
         })
-        .await
-        .expect("send listConversations page 2");
+        .await?;
     let resp2: JSONRPCResponse = timeout(
         DEFAULT_READ_TIMEOUT,
         mcp.read_stream_until_response_message(RequestId::Integer(req_id2)),
     )
-    .await
-    .expect("listConversations page 2 timeout")
-    .expect("listConversations page 2 resp");
+    .await??;
     let ListConversationsResponse {
         items: items2,
         next_cursor: next2,
         ..
-    } = to_response::<ListConversationsResponse>(resp2).expect("deserialize response");
+    } = to_response::<ListConversationsResponse>(resp2)?;
     assert_eq!(items2.len(), 1);
     assert_eq!(items2[0].preview, "Hello C");
     assert_eq!(items2[0].model_provider, "openai");
@@ -116,7 +105,7 @@ async fn test_list_and_resume_conversations() {
         "2025-01-01T11:30:00Z",
         "Hello TP",
         Some("test-provider"),
-    );
+    )?;
 
     // Filtering by model provider should return only matching sessions.
     let filter_req_id = mcp
@@ -125,19 +114,16 @@ async fn test_list_and_resume_conversations() {
             cursor: None,
             model_providers: Some(vec!["test-provider".to_string()]),
         })
-        .await
-        .expect("send listConversations filtered");
+        .await?;
     let filter_resp: JSONRPCResponse = timeout(
         DEFAULT_READ_TIMEOUT,
         mcp.read_stream_until_response_message(RequestId::Integer(filter_req_id)),
     )
-    .await
-    .expect("listConversations filtered timeout")
-    .expect("listConversations filtered resp");
+    .await??;
     let ListConversationsResponse {
         items: filtered_items,
         next_cursor: filtered_next,
-    } = to_response::<ListConversationsResponse>(filter_resp).expect("deserialize filtered");
+    } = to_response::<ListConversationsResponse>(filter_resp)?;
     assert_eq!(filtered_items.len(), 1);
     assert_eq!(filtered_next, None);
     assert_eq!(filtered_items[0].preview, "Hello TP");
@@ -150,20 +136,16 @@ async fn test_list_and_resume_conversations() {
             cursor: None,
             model_providers: Some(Vec::new()),
         })
-        .await
-        .expect("send listConversations unfiltered");
+        .await?;
     let unfiltered_resp: JSONRPCResponse = timeout(
         DEFAULT_READ_TIMEOUT,
         mcp.read_stream_until_response_message(RequestId::Integer(unfiltered_req_id)),
     )
-    .await
-    .expect("listConversations unfiltered timeout")
-    .expect("listConversations unfiltered resp");
+    .await??;
     let ListConversationsResponse {
         items: unfiltered_items,
         next_cursor: unfiltered_next,
-    } = to_response::<ListConversationsResponse>(unfiltered_resp)
-        .expect("deserialize unfiltered response");
+    } = to_response::<ListConversationsResponse>(unfiltered_resp)?;
     assert_eq!(unfiltered_items.len(), 4);
     assert!(unfiltered_next.is_none());
 
@@ -173,19 +155,16 @@ async fn test_list_and_resume_conversations() {
             cursor: None,
             model_providers: Some(vec!["other".to_string()]),
         })
-        .await
-        .expect("send listConversations filtered empty");
+        .await?;
     let empty_resp: JSONRPCResponse = timeout(
         DEFAULT_READ_TIMEOUT,
         mcp.read_stream_until_response_message(RequestId::Integer(empty_req_id)),
     )
-    .await
-    .expect("listConversations filtered empty timeout")
-    .expect("listConversations filtered empty resp");
+    .await??;
     let ListConversationsResponse {
         items: empty_items,
         next_cursor: empty_next,
-    } = to_response::<ListConversationsResponse>(empty_resp).expect("deserialize filtered empty");
+    } = to_response::<ListConversationsResponse>(empty_resp)?;
     assert!(empty_items.is_empty());
     assert!(empty_next.is_none());
 
@@ -198,20 +177,15 @@ async fn test_list_and_resume_conversations() {
                 ..Default::default()
             }),
         })
-        .await
-        .expect("send resumeConversation");
+        .await?;
 
     // Expect a codex/event notification with msg.type == sessionConfigured
     let notification: JSONRPCNotification = timeout(
         DEFAULT_READ_TIMEOUT,
         mcp.read_stream_until_notification_message("sessionConfigured"),
     )
-    .await
-    .expect("sessionConfigured notification timeout")
-    .expect("sessionConfigured notification");
-    let session_configured: ServerNotification = notification
-        .try_into()
-        .expect("deserialize sessionConfigured notification");
+    .await??;
+    let session_configured: ServerNotification = notification.try_into()?;
     // Basic shape assertion: ensure event type is sessionConfigured
     let ServerNotification::SessionConfigured(SessionConfiguredNotification {
         model,
@@ -229,15 +203,14 @@ async fn test_list_and_resume_conversations() {
         DEFAULT_READ_TIMEOUT,
         mcp.read_stream_until_response_message(RequestId::Integer(resume_req_id)),
     )
-    .await
-    .expect("resumeConversation timeout")
-    .expect("resumeConversation resp");
+    .await??;
     let ResumeConversationResponse {
         conversation_id, ..
-    } = to_response::<ResumeConversationResponse>(resume_resp)
-        .expect("deserialize resumeConversation response");
+    } = to_response::<ResumeConversationResponse>(resume_resp)?;
     // conversation id should be a valid UUID
     assert!(!conversation_id.to_string().is_empty());
+
+    Ok(())
 }
 
 fn create_fake_rollout(
@@ -246,14 +219,14 @@ fn create_fake_rollout(
     meta_rfc3339: &str,
     preview: &str,
     model_provider: Option<&str>,
-) {
+) -> Result<()> {
     let uuid = Uuid::new_v4();
     // sessions/YYYY/MM/DD/ derived from filename_ts (YYYY-MM-DDThh-mm-ss)
     let year = &filename_ts[0..4];
     let month = &filename_ts[5..7];
     let day = &filename_ts[8..10];
     let dir = codex_home.join("sessions").join(year).join(month).join(day);
-    fs::create_dir_all(&dir).unwrap_or_else(|e| panic!("create sessions dir: {e}"));
+    fs::create_dir_all(&dir)?;
 
     let file_path = dir.join(format!("rollout-{filename_ts}-{uuid}.jsonl"));
     let mut lines = Vec::new();
@@ -303,6 +276,6 @@ fn create_fake_rollout(
         })
         .to_string(),
     );
-    fs::write(file_path, lines.join("\n") + "\n")
-        .unwrap_or_else(|e| panic!("write rollout file: {e}"));
+    fs::write(file_path, lines.join("\n") + "\n")?;
+    Ok(())
 }
diff --git a/codex-rs/app-server/tests/suite/login.rs b/codex-rs/app-server/tests/suite/login.rs
index d4d6374bc4..c5470c3ec4 100644
--- a/codex-rs/app-server/tests/suite/login.rs
+++ b/codex-rs/app-server/tests/suite/login.rs
@@ -1,6 +1,4 @@
-use std::path::Path;
-use std::time::Duration;
-
+use anyhow::Result;
 use app_test_support::McpProcess;
 use app_test_support::to_response;
 use codex_app_server_protocol::CancelLoginChatGptParams;
@@ -12,8 +10,11 @@ use codex_app_server_protocol::JSONRPCResponse;
 use codex_app_server_protocol::LoginChatGptResponse;
 use codex_app_server_protocol::LogoutChatGptResponse;
 use codex_app_server_protocol::RequestId;
+use codex_core::auth::AuthCredentialsStoreMode;
 use codex_login::login_with_api_key;
 use serial_test::serial;
+use std::path::Path;
+use std::time::Duration;
 use tempfile::TempDir;
 use tokio::time::timeout;
 
@@ -42,32 +43,26 @@ stream_max_retries = 0
 }
 
 #[tokio::test(flavor = "multi_thread", worker_threads = 2)]
-async fn logout_chatgpt_removes_auth() {
-    let codex_home = TempDir::new().unwrap_or_else(|e| panic!("create tempdir: {e}"));
-    create_config_toml(codex_home.path()).expect("write config.toml");
-    login_with_api_key(codex_home.path(), "sk-test-key").expect("seed api key");
+async fn logout_chatgpt_removes_auth() -> Result<()> {
+    let codex_home = TempDir::new()?;
+    create_config_toml(codex_home.path())?;
+    login_with_api_key(
+        codex_home.path(),
+        "sk-test-key",
+        AuthCredentialsStoreMode::File,
+    )?;
     assert!(codex_home.path().join("auth.json").exists());
 
-    let mut mcp = McpProcess::new_with_env(codex_home.path(), &[("OPENAI_API_KEY", None)])
-        .await
-        .expect("spawn mcp process");
-    timeout(DEFAULT_READ_TIMEOUT, mcp.initialize())
-        .await
-        .expect("init timeout")
-        .expect("init failed");
+    let mut mcp = McpProcess::new_with_env(codex_home.path(), &[("OPENAI_API_KEY", None)]).await?;
+    timeout(DEFAULT_READ_TIMEOUT, mcp.initialize()).await??;
 
-    let id = mcp
-        .send_logout_chat_gpt_request()
-        .await
-        .expect("send logoutChatGpt");
+    let id = mcp.send_logout_chat_gpt_request().await?;
     let resp: JSONRPCResponse = timeout(
         DEFAULT_READ_TIMEOUT,
         mcp.read_stream_until_response_message(RequestId::Integer(id)),
     )
-    .await
-    .expect("logoutChatGpt timeout")
-    .expect("logoutChatGpt response");
-    let _ok: LogoutChatGptResponse = to_response(resp).expect("deserialize logout response");
+    .await??;
+    let _ok: LogoutChatGptResponse = to_response(resp)?;
 
     assert!(
         !codex_home.path().join("auth.json").exists(),
@@ -80,63 +75,47 @@ async fn logout_chatgpt_removes_auth() {
             include_token: Some(true),
             refresh_token: Some(false),
         })
-        .await
-        .expect("send getAuthStatus");
+        .await?;
     let status_resp: JSONRPCResponse = timeout(
         DEFAULT_READ_TIMEOUT,
         mcp.read_stream_until_response_message(RequestId::Integer(status_id)),
     )
-    .await
-    .expect("getAuthStatus timeout")
-    .expect("getAuthStatus response");
-    let status: GetAuthStatusResponse = to_response(status_resp).expect("deserialize status");
+    .await??;
+    let status: GetAuthStatusResponse = to_response(status_resp)?;
     assert_eq!(status.auth_method, None);
     assert_eq!(status.auth_token, None);
+    Ok(())
 }
 
 #[tokio::test(flavor = "multi_thread", worker_threads = 2)]
 // Serialize tests that launch the login server since it binds to a fixed port.
 #[serial(login_port)]
-async fn login_and_cancel_chatgpt() {
-    let codex_home = TempDir::new().unwrap_or_else(|e| panic!("create tempdir: {e}"));
-    create_config_toml(codex_home.path()).unwrap_or_else(|err| panic!("write config.toml: {err}"));
+async fn login_and_cancel_chatgpt() -> Result<()> {
+    let codex_home = TempDir::new()?;
+    create_config_toml(codex_home.path())?;
 
-    let mut mcp = McpProcess::new(codex_home.path())
-        .await
-        .expect("spawn mcp process");
-    timeout(DEFAULT_READ_TIMEOUT, mcp.initialize())
-        .await
-        .expect("init timeout")
-        .expect("init failed");
+    let mut mcp = McpProcess::new(codex_home.path()).await?;
+    timeout(DEFAULT_READ_TIMEOUT, mcp.initialize()).await??;
 
-    let login_id = mcp
-        .send_login_chat_gpt_request()
-        .await
-        .expect("send loginChatGpt");
+    let login_id = mcp.send_login_chat_gpt_request().await?;
     let login_resp: JSONRPCResponse = timeout(
         DEFAULT_READ_TIMEOUT,
         mcp.read_stream_until_response_message(RequestId::Integer(login_id)),
     )
-    .await
-    .expect("loginChatGpt timeout")
-    .expect("loginChatGpt response");
-    let login: LoginChatGptResponse = to_response(login_resp).expect("deserialize login resp");
+    .await??;
+    let login: LoginChatGptResponse = to_response(login_resp)?;
 
     let cancel_id = mcp
         .send_cancel_login_chat_gpt_request(CancelLoginChatGptParams {
             login_id: login.login_id,
         })
-        .await
-        .expect("send cancelLoginChatGpt");
+        .await?;
     let cancel_resp: JSONRPCResponse = timeout(
         DEFAULT_READ_TIMEOUT,
         mcp.read_stream_until_response_message(RequestId::Integer(cancel_id)),
     )
-    .await
-    .expect("cancelLoginChatGpt timeout")
-    .expect("cancelLoginChatGpt response");
-    let _ok: CancelLoginChatGptResponse =
-        to_response(cancel_resp).expect("deserialize cancel response");
+    .await??;
+    let _ok: CancelLoginChatGptResponse = to_response(cancel_resp)?;
 
     // Optionally observe the completion notification; do not fail if it races.
     let maybe_note = timeout(
@@ -147,6 +126,7 @@ async fn login_and_cancel_chatgpt() {
     if maybe_note.is_err() {
         eprintln!("warning: did not observe login_chat_gpt_complete notification after cancel");
     }
+    Ok(())
 }
 
 fn create_config_toml_forced_login(codex_home: &Path, forced_method: &str) -> std::io::Result<()> {
@@ -179,68 +159,48 @@ forced_chatgpt_workspace_id = "{workspace_id}"
 }
 
 #[tokio::test(flavor = "multi_thread", worker_threads = 2)]
-async fn login_chatgpt_rejected_when_forced_api() {
-    let codex_home = TempDir::new().unwrap_or_else(|e| panic!("create tempdir: {e}"));
-    create_config_toml_forced_login(codex_home.path(), "api")
-        .unwrap_or_else(|err| panic!("write config.toml: {err}"));
+async fn login_chatgpt_rejected_when_forced_api() -> Result<()> {
+    let codex_home = TempDir::new()?;
+    create_config_toml_forced_login(codex_home.path(), "api")?;
 
-    let mut mcp = McpProcess::new(codex_home.path())
-        .await
-        .expect("spawn mcp process");
-    timeout(DEFAULT_READ_TIMEOUT, mcp.initialize())
-        .await
-        .expect("init timeout")
-        .expect("init failed");
+    let mut mcp = McpProcess::new(codex_home.path()).await?;
+    timeout(DEFAULT_READ_TIMEOUT, mcp.initialize()).await??;
 
-    let request_id = mcp
-        .send_login_chat_gpt_request()
-        .await
-        .expect("send loginChatGpt");
+    let request_id = mcp.send_login_chat_gpt_request().await?;
     let err: JSONRPCError = timeout(
         DEFAULT_READ_TIMEOUT,
         mcp.read_stream_until_error_message(RequestId::Integer(request_id)),
     )
-    .await
-    .expect("loginChatGpt error timeout")
-    .expect("loginChatGpt error");
+    .await??;
 
     assert_eq!(
         err.error.message,
         "ChatGPT login is disabled. Use API key login instead."
     );
+    Ok(())
 }
 
 #[tokio::test(flavor = "multi_thread", worker_threads = 2)]
 // Serialize tests that launch the login server since it binds to a fixed port.
 #[serial(login_port)]
-async fn login_chatgpt_includes_forced_workspace_query_param() {
-    let codex_home = TempDir::new().unwrap_or_else(|e| panic!("create tempdir: {e}"));
-    create_config_toml_forced_workspace(codex_home.path(), "ws-forced")
-        .unwrap_or_else(|err| panic!("write config.toml: {err}"));
+async fn login_chatgpt_includes_forced_workspace_query_param() -> Result<()> {
+    let codex_home = TempDir::new()?;
+    create_config_toml_forced_workspace(codex_home.path(), "ws-forced")?;
 
-    let mut mcp = McpProcess::new(codex_home.path())
-        .await
-        .expect("spawn mcp process");
-    timeout(DEFAULT_READ_TIMEOUT, mcp.initialize())
-        .await
-        .expect("init timeout")
-        .expect("init failed");
+    let mut mcp = McpProcess::new(codex_home.path()).await?;
+    timeout(DEFAULT_READ_TIMEOUT, mcp.initialize()).await??;
 
-    let request_id = mcp
-        .send_login_chat_gpt_request()
-        .await
-        .expect("send loginChatGpt");
+    let request_id = mcp.send_login_chat_gpt_request().await?;
     let resp: JSONRPCResponse = timeout(
         DEFAULT_READ_TIMEOUT,
         mcp.read_stream_until_response_message(RequestId::Integer(request_id)),
     )
-    .await
-    .expect("loginChatGpt timeout")
-    .expect("loginChatGpt response");
+    .await??;
 
-    let login: LoginChatGptResponse = to_response(resp).expect("deserialize login resp");
+    let login: LoginChatGptResponse = to_response(resp)?;
     assert!(
         login.auth_url.contains("allowed_workspace_id=ws-forced"),
         "auth URL should include forced workspace"
     );
+    Ok(())
 }
diff --git a/codex-rs/app-server/tests/suite/rate_limits.rs b/codex-rs/app-server/tests/suite/rate_limits.rs
index 302fd60349..16cbf153e9 100644
--- a/codex-rs/app-server/tests/suite/rate_limits.rs
+++ b/codex-rs/app-server/tests/suite/rate_limits.rs
@@ -1,4 +1,3 @@
-use anyhow::Context;
 use anyhow::Result;
 use app_test_support::ChatGptAuthFixture;
 use app_test_support::McpProcess;
@@ -9,6 +8,7 @@ use codex_app_server_protocol::JSONRPCError;
 use codex_app_server_protocol::JSONRPCResponse;
 use codex_app_server_protocol::LoginApiKeyParams;
 use codex_app_server_protocol::RequestId;
+use codex_core::auth::AuthCredentialsStoreMode;
 use codex_protocol::protocol::RateLimitSnapshot;
 use codex_protocol::protocol::RateLimitWindow;
 use pretty_assertions::assert_eq;
@@ -28,28 +28,18 @@ const INVALID_REQUEST_ERROR_CODE: i64 = -32600;
 
 #[tokio::test(flavor = "multi_thread", worker_threads = 2)]
 async fn get_account_rate_limits_requires_auth() -> Result<()> {
-    let codex_home = TempDir::new().context("create codex home tempdir")?;
+    let codex_home = TempDir::new()?;
 
-    let mut mcp = McpProcess::new_with_env(codex_home.path(), &[("OPENAI_API_KEY", None)])
-        .await
-        .context("spawn mcp process")?;
-    timeout(DEFAULT_READ_TIMEOUT, mcp.initialize())
-        .await
-        .context("initialize timeout")?
-        .context("initialize request")?;
+    let mut mcp = McpProcess::new_with_env(codex_home.path(), &[("OPENAI_API_KEY", None)]).await?;
+    timeout(DEFAULT_READ_TIMEOUT, mcp.initialize()).await??;
 
-    let request_id = mcp
-        .send_get_account_rate_limits_request()
-        .await
-        .context("send account/rateLimits/read")?;
+    let request_id = mcp.send_get_account_rate_limits_request().await?;
 
     let error: JSONRPCError = timeout(
         DEFAULT_READ_TIMEOUT,
         mcp.read_stream_until_error_message(RequestId::Integer(request_id)),
     )
-    .await
-    .context("account/rateLimits/read timeout")?
-    .context("account/rateLimits/read error")?;
+    .await??;
 
     assert_eq!(error.id, RequestId::Integer(request_id));
     assert_eq!(error.error.code, INVALID_REQUEST_ERROR_CODE);
@@ -63,30 +53,20 @@ async fn get_account_rate_limits_requires_auth() -> Result<()> {
 
 #[tokio::test(flavor = "multi_thread", worker_threads = 2)]
 async fn get_account_rate_limits_requires_chatgpt_auth() -> Result<()> {
-    let codex_home = TempDir::new().context("create codex home tempdir")?;
+    let codex_home = TempDir::new()?;
 
-    let mut mcp = McpProcess::new(codex_home.path())
-        .await
-        .context("spawn mcp process")?;
-    timeout(DEFAULT_READ_TIMEOUT, mcp.initialize())
-        .await
-        .context("initialize timeout")?
-        .context("initialize request")?;
+    let mut mcp = McpProcess::new(codex_home.path()).await?;
+    timeout(DEFAULT_READ_TIMEOUT, mcp.initialize()).await??;
 
     login_with_api_key(&mut mcp, "sk-test-key").await?;
 
-    let request_id = mcp
-        .send_get_account_rate_limits_request()
-        .await
-        .context("send account/rateLimits/read")?;
+    let request_id = mcp.send_get_account_rate_limits_request().await?;
 
     let error: JSONRPCError = timeout(
         DEFAULT_READ_TIMEOUT,
         mcp.read_stream_until_error_message(RequestId::Integer(request_id)),
     )
-    .await
-    .context("account/rateLimits/read timeout")?
-    .context("account/rateLimits/read error")?;
+    .await??;
 
     assert_eq!(error.id, RequestId::Integer(request_id));
     assert_eq!(error.error.code, INVALID_REQUEST_ERROR_CODE);
@@ -100,18 +80,18 @@ async fn get_account_rate_limits_requires_chatgpt_auth() -> Result<()> {
 
 #[tokio::test(flavor = "multi_thread", worker_threads = 2)]
 async fn get_account_rate_limits_returns_snapshot() -> Result<()> {
-    let codex_home = TempDir::new().context("create codex home tempdir")?;
+    let codex_home = TempDir::new()?;
     write_chatgpt_auth(
         codex_home.path(),
         ChatGptAuthFixture::new("chatgpt-token")
             .account_id("account-123")
             .plan_type("pro"),
-    )
-    .context("write chatgpt auth")?;
+        AuthCredentialsStoreMode::File,
+    )?;
 
     let server = MockServer::start().await;
     let server_url = server.uri();
-    write_chatgpt_base_url(codex_home.path(), &server_url).context("write chatgpt base url")?;
+    write_chatgpt_base_url(codex_home.path(), &server_url)?;
 
     let primary_reset_timestamp = chrono::DateTime::parse_from_rfc3339("2025-01-01T00:02:00Z")
         .expect("parse primary reset timestamp")
@@ -147,29 +127,18 @@ async fn get_account_rate_limits_returns_snapshot() -> Result<()> {
         .mount(&server)
         .await;
 
-    let mut mcp = McpProcess::new_with_env(codex_home.path(), &[("OPENAI_API_KEY", None)])
-        .await
-        .context("spawn mcp process")?;
-    timeout(DEFAULT_READ_TIMEOUT, mcp.initialize())
-        .await
-        .context("initialize timeout")?
-        .context("initialize request")?;
+    let mut mcp = McpProcess::new_with_env(codex_home.path(), &[("OPENAI_API_KEY", None)]).await?;
+    timeout(DEFAULT_READ_TIMEOUT, mcp.initialize()).await??;
 
-    let request_id = mcp
-        .send_get_account_rate_limits_request()
-        .await
-        .context("send account/rateLimits/read")?;
+    let request_id = mcp.send_get_account_rate_limits_request().await?;
 
     let response: JSONRPCResponse = timeout(
         DEFAULT_READ_TIMEOUT,
         mcp.read_stream_until_response_message(RequestId::Integer(request_id)),
     )
-    .await
-    .context("account/rateLimits/read timeout")?
-    .context("account/rateLimits/read response")?;
+    .await??;
 
-    let received: GetAccountRateLimitsResponse =
-        to_response(response).context("deserialize rate limit response")?;
+    let received: GetAccountRateLimitsResponse = to_response(response)?;
 
     let expected = GetAccountRateLimitsResponse {
         rate_limits: RateLimitSnapshot {
@@ -195,16 +164,13 @@ async fn login_with_api_key(mcp: &mut McpProcess, api_key: &str) -> Result<()> {
         .send_login_api_key_request(LoginApiKeyParams {
             api_key: api_key.to_string(),
         })
-        .await
-        .context("send loginApiKey")?;
+        .await?;
 
     timeout(
         DEFAULT_READ_TIMEOUT,
         mcp.read_stream_until_response_message(RequestId::Integer(request_id)),
     )
-    .await
-    .context("loginApiKey timeout")?
-    .context("loginApiKey response")?;
+    .await??;
 
     Ok(())
 }
diff --git a/codex-rs/app-server/tests/suite/send_message.rs b/codex-rs/app-server/tests/suite/send_message.rs
index b3f04c3393..8d4e508c38 100644
--- a/codex-rs/app-server/tests/suite/send_message.rs
+++ b/codex-rs/app-server/tests/suite/send_message.rs
@@ -1,5 +1,4 @@
-use std::path::Path;
-
+use anyhow::Result;
 use app_test_support::McpProcess;
 use app_test_support::create_final_assistant_message_sse_response;
 use app_test_support::create_mock_chat_completions_server;
@@ -18,49 +17,42 @@ use codex_protocol::ConversationId;
 use codex_protocol::models::ContentItem;
 use codex_protocol::models::ResponseItem;
 use pretty_assertions::assert_eq;
+use std::path::Path;
 use tempfile::TempDir;
 use tokio::time::timeout;
 
 const DEFAULT_READ_TIMEOUT: std::time::Duration = std::time::Duration::from_secs(10);
 
 #[tokio::test]
-async fn test_send_message_success() {
+async fn test_send_message_success() -> Result<()> {
     // Spin up a mock completions server that immediately ends the Codex turn.
     // Two Codex turns hit the mock model (session start + send-user-message). Provide two SSE responses.
     let responses = vec![
-        create_final_assistant_message_sse_response("Done").expect("build mock assistant message"),
-        create_final_assistant_message_sse_response("Done").expect("build mock assistant message"),
+        create_final_assistant_message_sse_response("Done")?,
+        create_final_assistant_message_sse_response("Done")?,
     ];
     let server = create_mock_chat_completions_server(responses).await;
 
     // Create a temporary Codex home with config pointing at the mock server.
-    let codex_home = TempDir::new().expect("create temp dir");
-    create_config_toml(codex_home.path(), &server.uri()).expect("write config.toml");
+    let codex_home = TempDir::new()?;
+    create_config_toml(codex_home.path(), &server.uri())?;
 
     // Start MCP server process and initialize.
-    let mut mcp = McpProcess::new(codex_home.path())
-        .await
-        .expect("spawn mcp process");
-    timeout(DEFAULT_READ_TIMEOUT, mcp.initialize())
-        .await
-        .expect("init timed out")
-        .expect("init failed");
+    let mut mcp = McpProcess::new(codex_home.path()).await?;
+    timeout(DEFAULT_READ_TIMEOUT, mcp.initialize()).await??;
 
     // Start a conversation using the new wire API.
     let new_conv_id = mcp
         .send_new_conversation_request(NewConversationParams::default())
-        .await
-        .expect("send newConversation");
+        .await?;
     let new_conv_resp: JSONRPCResponse = timeout(
         DEFAULT_READ_TIMEOUT,
         mcp.read_stream_until_response_message(RequestId::Integer(new_conv_id)),
     )
-    .await
-    .expect("newConversation timeout")
-    .expect("newConversation resp");
+    .await??;
     let NewConversationResponse {
         conversation_id, ..
-    } = to_response::<_>(new_conv_resp).expect("deserialize newConversation response");
+    } = to_response::<_>(new_conv_resp)?;
 
     // 2) addConversationListener
     let add_listener_id = mcp
@@ -68,25 +60,27 @@ async fn test_send_message_success() {
             conversation_id,
             experimental_raw_events: false,
         })
-        .await
-        .expect("send addConversationListener");
+        .await?;
     let add_listener_resp: JSONRPCResponse = timeout(
         DEFAULT_READ_TIMEOUT,
         mcp.read_stream_until_response_message(RequestId::Integer(add_listener_id)),
     )
-    .await
-    .expect("addConversationListener timeout")
-    .expect("addConversationListener resp");
+    .await??;
     let AddConversationSubscriptionResponse { subscription_id: _ } =
-        to_response::<_>(add_listener_resp).expect("deserialize addConversationListener response");
+        to_response::<_>(add_listener_resp)?;
 
     // Now exercise sendUserMessage twice.
-    send_message("Hello", conversation_id, &mut mcp).await;
-    send_message("Hello again", conversation_id, &mut mcp).await;
+    send_message("Hello", conversation_id, &mut mcp).await?;
+    send_message("Hello again", conversation_id, &mut mcp).await?;
+    Ok(())
 }
 
 #[expect(clippy::expect_used)]
-async fn send_message(message: &str, conversation_id: ConversationId, mcp: &mut McpProcess) {
+async fn send_message(
+    message: &str,
+    conversation_id: ConversationId,
+    mcp: &mut McpProcess,
+) -> Result<()> {
     // Now exercise sendUserMessage.
     let send_id = mcp
         .send_send_user_message_request(SendUserMessageParams {
@@ -95,19 +89,15 @@ async fn send_message(message: &str, conversation_id: ConversationId, mcp: &mut
                 text: message.to_string(),
             }],
         })
-        .await
-        .expect("send sendUserMessage");
+        .await?;
 
     let response: JSONRPCResponse = timeout(
         DEFAULT_READ_TIMEOUT,
         mcp.read_stream_until_response_message(RequestId::Integer(send_id)),
     )
-    .await
-    .expect("sendUserMessage response timeout")
-    .expect("sendUserMessage response error");
+    .await??;
 
-    let _ok: SendUserMessageResponse = to_response::<SendUserMessageResponse>(response)
-        .expect("deserialize sendUserMessage response");
+    let _ok: SendUserMessageResponse = to_response::<SendUserMessageResponse>(response)?;
 
     // Verify the task_finished notification is received.
     // Note this also ensures that the final request to the server was made.
@@ -115,9 +105,7 @@ async fn send_message(message: &str, conversation_id: ConversationId, mcp: &mut
         DEFAULT_READ_TIMEOUT,
         mcp.read_stream_until_notification_message("codex/event/task_complete"),
     )
-    .await
-    .expect("task_finished_notification timeout")
-    .expect("task_finished_notification resp");
+    .await??;
     let serde_json::Value::Object(map) = task_finished_notification
         .params
         .expect("notification should have params")
@@ -139,57 +127,45 @@ async fn send_message(message: &str, conversation_id: ConversationId, mcp: &mut
         raw_attempt.is_err(),
         "unexpected raw item notification when not opted in"
     );
+    Ok(())
 }
 
 #[tokio::test]
-async fn test_send_message_raw_notifications_opt_in() {
-    let responses = vec![
-        create_final_assistant_message_sse_response("Done").expect("build mock assistant message"),
-    ];
+async fn test_send_message_raw_notifications_opt_in() -> Result<()> {
+    let responses = vec![create_final_assistant_message_sse_response("Done")?];
     let server = create_mock_chat_completions_server(responses).await;
 
-    let codex_home = TempDir::new().expect("create temp dir");
-    create_config_toml(codex_home.path(), &server.uri()).expect("write config.toml");
+    let codex_home = TempDir::new()?;
+    create_config_toml(codex_home.path(), &server.uri())?;
 
-    let mut mcp = McpProcess::new(codex_home.path())
-        .await
-        .expect("spawn mcp process");
-    timeout(DEFAULT_READ_TIMEOUT, mcp.initialize())
-        .await
-        .expect("init timed out")
-        .expect("init failed");
+    let mut mcp = McpProcess::new(codex_home.path()).await?;
+    timeout(DEFAULT_READ_TIMEOUT, mcp.initialize()).await??;
 
     let new_conv_id = mcp
         .send_new_conversation_request(NewConversationParams::default())
-        .await
-        .expect("send newConversation");
+        .await?;
     let new_conv_resp: JSONRPCResponse = timeout(
         DEFAULT_READ_TIMEOUT,
         mcp.read_stream_until_response_message(RequestId::Integer(new_conv_id)),
     )
-    .await
-    .expect("newConversation timeout")
-    .expect("newConversation resp");
+    .await??;
     let NewConversationResponse {
         conversation_id, ..
-    } = to_response::<_>(new_conv_resp).expect("deserialize newConversation response");
+    } = to_response::<_>(new_conv_resp)?;
 
     let add_listener_id = mcp
         .send_add_conversation_listener_request(AddConversationListenerParams {
             conversation_id,
             experimental_raw_events: true,
         })
-        .await
-        .expect("send addConversationListener");
+        .await?;
     let add_listener_resp: JSONRPCResponse = timeout(
         DEFAULT_READ_TIMEOUT,
         mcp.read_stream_until_response_message(RequestId::Integer(add_listener_id)),
     )
-    .await
-    .expect("addConversationListener timeout")
-    .expect("addConversationListener resp");
+    .await??;
     let AddConversationSubscriptionResponse { subscription_id: _ } =
-        to_response::<_>(add_listener_resp).expect("deserialize addConversationListener response");
+        to_response::<_>(add_listener_resp)?;
 
     let send_id = mcp
         .send_send_user_message_request(SendUserMessageParams {
@@ -198,8 +174,7 @@ async fn test_send_message_raw_notifications_opt_in() {
                 text: "Hello".to_string(),
             }],
         })
-        .await
-        .expect("send sendUserMessage");
+        .await?;
 
     let instructions = read_raw_response_item(&mut mcp, conversation_id).await;
     assert_instructions_message(&instructions);
@@ -211,11 +186,8 @@ async fn test_send_message_raw_notifications_opt_in() {
         DEFAULT_READ_TIMEOUT,
         mcp.read_stream_until_response_message(RequestId::Integer(send_id)),
     )
-    .await
-    .expect("sendUserMessage response timeout")
-    .expect("sendUserMessage response error");
-    let _ok: SendUserMessageResponse = to_response::<SendUserMessageResponse>(response)
-        .expect("deserialize sendUserMessage response");
+    .await??;
+    let _ok: SendUserMessageResponse = to_response::<SendUserMessageResponse>(response)?;
 
     let user_message = read_raw_response_item(&mut mcp, conversation_id).await;
     assert_user_message(&user_message, "Hello");
@@ -228,17 +200,16 @@ async fn test_send_message_raw_notifications_opt_in() {
         mcp.read_stream_until_notification_message("codex/event/task_complete"),
     )
     .await;
+
+    Ok(())
 }
 
 #[tokio::test]
-async fn test_send_message_session_not_found() {
+async fn test_send_message_session_not_found() -> Result<()> {
     // Start MCP without creating a Codex session
-    let codex_home = TempDir::new().expect("tempdir");
-    let mut mcp = McpProcess::new(codex_home.path()).await.expect("spawn");
-    timeout(DEFAULT_READ_TIMEOUT, mcp.initialize())
-        .await
-        .expect("timeout")
-        .expect("init");
+    let codex_home = TempDir::new()?;
+    let mut mcp = McpProcess::new(codex_home.path()).await?;
+    timeout(DEFAULT_READ_TIMEOUT, mcp.initialize()).await??;
 
     let unknown = ConversationId::new();
     let req_id = mcp
@@ -248,18 +219,16 @@ async fn test_send_message_session_not_found() {
                 text: "ping".to_string(),
             }],
         })
-        .await
-        .expect("send sendUserMessage");
+        .await?;
 
     // Expect an error response for unknown conversation.
     let err = timeout(
         DEFAULT_READ_TIMEOUT,
         mcp.read_stream_until_error_message(RequestId::Integer(req_id)),
     )
-    .await
-    .expect("timeout")
-    .expect("error");
+    .await??;
     assert_eq!(err.id, RequestId::Integer(req_id));
+    Ok(())
 }
 
 // ---------------------------------------------------------------------------
diff --git a/codex-rs/app-server/tests/suite/set_default_model.rs b/codex-rs/app-server/tests/suite/set_default_model.rs
index 6769faa9dc..0c2aa22999 100644
--- a/codex-rs/app-server/tests/suite/set_default_model.rs
+++ b/codex-rs/app-server/tests/suite/set_default_model.rs
@@ -1,5 +1,4 @@
-use std::path::Path;
-
+use anyhow::Result;
 use app_test_support::McpProcess;
 use app_test_support::to_response;
 use codex_app_server_protocol::JSONRPCResponse;
@@ -8,50 +7,38 @@ use codex_app_server_protocol::SetDefaultModelParams;
 use codex_app_server_protocol::SetDefaultModelResponse;
 use codex_core::config::ConfigToml;
 use pretty_assertions::assert_eq;
+use std::path::Path;
 use tempfile::TempDir;
 use tokio::time::timeout;
 
 const DEFAULT_READ_TIMEOUT: std::time::Duration = std::time::Duration::from_secs(10);
 
 #[tokio::test(flavor = "multi_thread", worker_threads = 2)]
-async fn set_default_model_persists_overrides() {
-    let codex_home = TempDir::new().expect("create tempdir");
-    create_config_toml(codex_home.path()).expect("write config.toml");
+async fn set_default_model_persists_overrides() -> Result<()> {
+    let codex_home = TempDir::new()?;
+    create_config_toml(codex_home.path())?;
 
-    let mut mcp = McpProcess::new(codex_home.path())
-        .await
-        .expect("spawn mcp process");
-    timeout(DEFAULT_READ_TIMEOUT, mcp.initialize())
-        .await
-        .expect("init timeout")
-        .expect("init failed");
+    let mut mcp = McpProcess::new(codex_home.path()).await?;
+    timeout(DEFAULT_READ_TIMEOUT, mcp.initialize()).await??;
 
     let params = SetDefaultModelParams {
         model: Some("gpt-4.1".to_string()),
         reasoning_effort: None,
     };
 
-    let request_id = mcp
-        .send_set_default_model_request(params)
-        .await
-        .expect("send setDefaultModel");
+    let request_id = mcp.send_set_default_model_request(params).await?;
 
     let resp: JSONRPCResponse = timeout(
         DEFAULT_READ_TIMEOUT,
         mcp.read_stream_until_response_message(RequestId::Integer(request_id)),
     )
-    .await
-    .expect("setDefaultModel timeout")
-    .expect("setDefaultModel response");
+    .await??;
 
-    let _: SetDefaultModelResponse =
-        to_response(resp).expect("deserialize setDefaultModel response");
+    let _: SetDefaultModelResponse = to_response(resp)?;
 
     let config_path = codex_home.path().join("config.toml");
-    let config_contents = tokio::fs::read_to_string(&config_path)
-        .await
-        .expect("read config.toml");
-    let config_toml: ConfigToml = toml::from_str(&config_contents).expect("parse config.toml");
+    let config_contents = tokio::fs::read_to_string(&config_path).await?;
+    let config_toml: ConfigToml = toml::from_str(&config_contents)?;
 
     assert_eq!(
         ConfigToml {
@@ -61,6 +48,7 @@ async fn set_default_model_persists_overrides() {
         },
         config_toml,
     );
+    Ok(())
 }
 
 // Helper to create a config.toml; mirrors create_conversation.rs
diff --git a/codex-rs/app-server/tests/suite/user_agent.rs b/codex-rs/app-server/tests/suite/user_agent.rs
index 95a0b1a3e0..52ba6e56aa 100644
--- a/codex-rs/app-server/tests/suite/user_agent.rs
+++ b/codex-rs/app-server/tests/suite/user_agent.rs
@@ -1,3 +1,4 @@
+use anyhow::Result;
 use app_test_support::McpProcess;
 use app_test_support::to_response;
 use codex_app_server_protocol::GetUserAgentResponse;
@@ -10,28 +11,18 @@ use tokio::time::timeout;
 const DEFAULT_READ_TIMEOUT: std::time::Duration = std::time::Duration::from_secs(10);
 
 #[tokio::test(flavor = "multi_thread", worker_threads = 2)]
-async fn get_user_agent_returns_current_codex_user_agent() {
-    let codex_home = TempDir::new().unwrap_or_else(|err| panic!("create tempdir: {err}"));
+async fn get_user_agent_returns_current_codex_user_agent() -> Result<()> {
+    let codex_home = TempDir::new()?;
 
-    let mut mcp = McpProcess::new(codex_home.path())
-        .await
-        .expect("spawn mcp process");
-    timeout(DEFAULT_READ_TIMEOUT, mcp.initialize())
-        .await
-        .expect("initialize timeout")
-        .expect("initialize request");
+    let mut mcp = McpProcess::new(codex_home.path()).await?;
+    timeout(DEFAULT_READ_TIMEOUT, mcp.initialize()).await??;
 
-    let request_id = mcp
-        .send_get_user_agent_request()
-        .await
-        .expect("send getUserAgent");
+    let request_id = mcp.send_get_user_agent_request().await?;
     let response: JSONRPCResponse = timeout(
         DEFAULT_READ_TIMEOUT,
         mcp.read_stream_until_response_message(RequestId::Integer(request_id)),
     )
-    .await
-    .expect("getUserAgent timeout")
-    .expect("getUserAgent response");
+    .await??;
 
     let os_info = os_info::get();
     let user_agent = format!(
@@ -42,9 +33,9 @@ async fn get_user_agent_returns_current_codex_user_agent() {
         codex_core::terminal::user_agent()
     );
 
-    let received: GetUserAgentResponse =
-        to_response(response).expect("deserialize getUserAgent response");
+    let received: GetUserAgentResponse = to_response(response)?;
     let expected = GetUserAgentResponse { user_agent };
 
     assert_eq!(received, expected);
+    Ok(())
 }
diff --git a/codex-rs/app-server/tests/suite/user_info.rs b/codex-rs/app-server/tests/suite/user_info.rs
index b730b77fed..6a44f1a3ab 100644
--- a/codex-rs/app-server/tests/suite/user_info.rs
+++ b/codex-rs/app-server/tests/suite/user_info.rs
@@ -1,5 +1,4 @@
-use std::time::Duration;
-
+use anyhow::Result;
 use app_test_support::ChatGptAuthFixture;
 use app_test_support::McpProcess;
 use app_test_support::to_response;
@@ -7,45 +6,41 @@ use app_test_support::write_chatgpt_auth;
 use codex_app_server_protocol::JSONRPCResponse;
 use codex_app_server_protocol::RequestId;
 use codex_app_server_protocol::UserInfoResponse;
+use codex_core::auth::AuthCredentialsStoreMode;
 use pretty_assertions::assert_eq;
+use std::time::Duration;
 use tempfile::TempDir;
 use tokio::time::timeout;
 
 const DEFAULT_READ_TIMEOUT: Duration = Duration::from_secs(10);
 
 #[tokio::test(flavor = "multi_thread", worker_threads = 2)]
-async fn user_info_returns_email_from_auth_json() {
-    let codex_home = TempDir::new().expect("create tempdir");
+async fn user_info_returns_email_from_auth_json() -> Result<()> {
+    let codex_home = TempDir::new()?;
 
     write_chatgpt_auth(
         codex_home.path(),
         ChatGptAuthFixture::new("access")
             .refresh_token("refresh")
             .email("user@example.com"),
-    )
-    .expect("write chatgpt auth");
+        AuthCredentialsStoreMode::File,
+    )?;
 
-    let mut mcp = McpProcess::new(codex_home.path())
-        .await
-        .expect("spawn mcp process");
-    timeout(DEFAULT_READ_TIMEOUT, mcp.initialize())
-        .await
-        .expect("initialize timeout")
-        .expect("initialize request");
+    let mut mcp = McpProcess::new(codex_home.path()).await?;
+    timeout(DEFAULT_READ_TIMEOUT, mcp.initialize()).await??;
 
-    let request_id = mcp.send_user_info_request().await.expect("send userInfo");
+    let request_id = mcp.send_user_info_request().await?;
     let response: JSONRPCResponse = timeout(
         DEFAULT_READ_TIMEOUT,
         mcp.read_stream_until_response_message(RequestId::Integer(request_id)),
     )
-    .await
-    .expect("userInfo timeout")
-    .expect("userInfo response");
+    .await??;
 
-    let received: UserInfoResponse = to_response(response).expect("deserialize userInfo response");
+    let received: UserInfoResponse = to_response(response)?;
     let expected = UserInfoResponse {
         alleged_user_email: Some("user@example.com".to_string()),
     };
 
     assert_eq!(received, expected);
+    Ok(())
 }
diff --git a/codex-rs/chatgpt/src/apply_command.rs b/codex-rs/chatgpt/src/apply_command.rs
index ea7c69d572..e1289c213c 100644
--- a/codex-rs/chatgpt/src/apply_command.rs
+++ b/codex-rs/chatgpt/src/apply_command.rs
@@ -32,7 +32,8 @@ pub async fn run_apply_command(
     )
     .await?;
 
-    init_chatgpt_token_from_auth(&config.codex_home).await?;
+    init_chatgpt_token_from_auth(&config.codex_home, config.cli_auth_credentials_store_mode)
+        .await?;
 
     let task_response = get_task(&config, apply_cli.task_id).await?;
     apply_diff_from_task(task_response, cwd).await
diff --git a/codex-rs/chatgpt/src/chatgpt_client.rs b/codex-rs/chatgpt/src/chatgpt_client.rs
index d450f7c59e..7528631981 100644
--- a/codex-rs/chatgpt/src/chatgpt_client.rs
+++ b/codex-rs/chatgpt/src/chatgpt_client.rs
@@ -13,7 +13,8 @@ pub(crate) async fn chatgpt_get_request<T: DeserializeOwned>(
     path: String,
 ) -> anyhow::Result<T> {
     let chatgpt_base_url = &config.chatgpt_base_url;
-    init_chatgpt_token_from_auth(&config.codex_home).await?;
+    init_chatgpt_token_from_auth(&config.codex_home, config.cli_auth_credentials_store_mode)
+        .await?;
 
     // Make direct HTTP request to ChatGPT backend API with the token
     let client = create_client();
diff --git a/codex-rs/chatgpt/src/chatgpt_token.rs b/codex-rs/chatgpt/src/chatgpt_token.rs
index 70c6940f76..e8879ad21c 100644
--- a/codex-rs/chatgpt/src/chatgpt_token.rs
+++ b/codex-rs/chatgpt/src/chatgpt_token.rs
@@ -3,6 +3,7 @@ use std::path::Path;
 use std::sync::LazyLock;
 use std::sync::RwLock;
 
+use codex_core::auth::AuthCredentialsStoreMode;
 use codex_core::token_data::TokenData;
 
 static CHATGPT_TOKEN: LazyLock<RwLock<Option<TokenData>>> = LazyLock::new(|| RwLock::new(None));
@@ -18,8 +19,11 @@ pub fn set_chatgpt_token_data(value: TokenData) {
 }
 
 /// Initialize the ChatGPT token from auth.json file
-pub async fn init_chatgpt_token_from_auth(codex_home: &Path) -> std::io::Result<()> {
-    let auth = CodexAuth::from_auth_storage(codex_home)?;
+pub async fn init_chatgpt_token_from_auth(
+    codex_home: &Path,
+    auth_credentials_store_mode: AuthCredentialsStoreMode,
+) -> std::io::Result<()> {
+    let auth = CodexAuth::from_auth_storage(codex_home, auth_credentials_store_mode)?;
     if let Some(auth) = auth {
         let token_data = auth.get_token_data().await?;
         set_chatgpt_token_data(token_data);
diff --git a/codex-rs/cli/src/login.rs b/codex-rs/cli/src/login.rs
index bbb58f226a..6681ab20c8 100644
--- a/codex-rs/cli/src/login.rs
+++ b/codex-rs/cli/src/login.rs
@@ -1,6 +1,7 @@
 use codex_app_server_protocol::AuthMode;
 use codex_common::CliConfigOverrides;
 use codex_core::CodexAuth;
+use codex_core::auth::AuthCredentialsStoreMode;
 use codex_core::auth::CLIENT_ID;
 use codex_core::auth::login_with_api_key;
 use codex_core::auth::logout;
@@ -17,11 +18,13 @@ use std::path::PathBuf;
 pub async fn login_with_chatgpt(
     codex_home: PathBuf,
     forced_chatgpt_workspace_id: Option<String>,
+    cli_auth_credentials_store_mode: AuthCredentialsStoreMode,
 ) -> std::io::Result<()> {
     let opts = ServerOptions::new(
         codex_home,
         CLIENT_ID.to_string(),
         forced_chatgpt_workspace_id,
+        cli_auth_credentials_store_mode,
     );
     let server = run_login_server(opts)?;
 
@@ -43,7 +46,13 @@ pub async fn run_login_with_chatgpt(cli_config_overrides: CliConfigOverrides) ->
 
     let forced_chatgpt_workspace_id = config.forced_chatgpt_workspace_id.clone();
 
-    match login_with_chatgpt(config.codex_home, forced_chatgpt_workspace_id).await {
+    match login_with_chatgpt(
+        config.codex_home,
+        forced_chatgpt_workspace_id,
+        config.cli_auth_credentials_store_mode,
+    )
+    .await
+    {
         Ok(_) => {
             eprintln!("Successfully logged in");
             std::process::exit(0);
@@ -66,7 +75,11 @@ pub async fn run_login_with_api_key(
         std::process::exit(1);
     }
 
-    match login_with_api_key(&config.codex_home, &api_key) {
+    match login_with_api_key(
+        &config.codex_home,
+        &api_key,
+        config.cli_auth_credentials_store_mode,
+    ) {
         Ok(_) => {
             eprintln!("Successfully logged in");
             std::process::exit(0);
@@ -121,6 +134,7 @@ pub async fn run_login_with_device_code(
         config.codex_home,
         client_id.unwrap_or(CLIENT_ID.to_string()),
         forced_chatgpt_workspace_id,
+        config.cli_auth_credentials_store_mode,
     );
     if let Some(iss) = issuer_base_url {
         opts.issuer = iss;
@@ -140,7 +154,7 @@ pub async fn run_login_with_device_code(
 pub async fn run_login_status(cli_config_overrides: CliConfigOverrides) -> ! {
     let config = load_config_or_exit(cli_config_overrides).await;
 
-    match CodexAuth::from_auth_storage(&config.codex_home) {
+    match CodexAuth::from_auth_storage(&config.codex_home, config.cli_auth_credentials_store_mode) {
         Ok(Some(auth)) => match auth.mode {
             AuthMode::ApiKey => match auth.get_token().await {
                 Ok(api_key) => {
@@ -171,7 +185,7 @@ pub async fn run_login_status(cli_config_overrides: CliConfigOverrides) -> ! {
 pub async fn run_logout(cli_config_overrides: CliConfigOverrides) -> ! {
     let config = load_config_or_exit(cli_config_overrides).await;
 
-    match logout(&config.codex_home) {
+    match logout(&config.codex_home, config.cli_auth_credentials_store_mode) {
         Ok(true) => {
             eprintln!("Successfully logged out");
             std::process::exit(0);
diff --git a/codex-rs/cloud-tasks/src/lib.rs b/codex-rs/cloud-tasks/src/lib.rs
index 6087cbea5b..a9bf0a8e19 100644
--- a/codex-rs/cloud-tasks/src/lib.rs
+++ b/codex-rs/cloud-tasks/src/lib.rs
@@ -58,7 +58,16 @@ async fn init_backend(user_agent_suffix: &str) -> anyhow::Result<BackendContext>
 
     let auth = match codex_core::config::find_codex_home()
         .ok()
-        .map(|home| codex_login::AuthManager::new(home, false))
+        .map(|home| {
+            let store_mode = codex_core::config::Config::load_from_base_config_with_overrides(
+                codex_core::config::ConfigToml::default(),
+                codex_core::config::ConfigOverrides::default(),
+                home.clone(),
+            )
+            .map(|cfg| cfg.cli_auth_credentials_store_mode)
+            .unwrap_or_default();
+            codex_login::AuthManager::new(home, false, store_mode)
+        })
         .and_then(|am| am.auth())
     {
         Some(auth) => auth,
diff --git a/codex-rs/cloud-tasks/src/util.rs b/codex-rs/cloud-tasks/src/util.rs
index 5d160e54fa..ff17265f3d 100644
--- a/codex-rs/cloud-tasks/src/util.rs
+++ b/codex-rs/cloud-tasks/src/util.rs
@@ -70,7 +70,14 @@ pub async fn build_chatgpt_headers() -> HeaderMap {
         HeaderValue::from_str(&ua).unwrap_or(HeaderValue::from_static("codex-cli")),
     );
     if let Ok(home) = codex_core::config::find_codex_home() {
-        let am = codex_login::AuthManager::new(home, false);
+        let store_mode = codex_core::config::Config::load_from_base_config_with_overrides(
+            codex_core::config::ConfigToml::default(),
+            codex_core::config::ConfigOverrides::default(),
+            home.clone(),
+        )
+        .map(|cfg| cfg.cli_auth_credentials_store_mode)
+        .unwrap_or_default();
+        let am = codex_login::AuthManager::new(home, false, store_mode);
         if let Some(auth) = am.auth()
             && let Ok(tok) = auth.get_token().await
             && !tok.is_empty()
diff --git a/codex-rs/common/src/approval_presets.rs b/codex-rs/common/src/approval_presets.rs
index 6dc0cff0ee..6c3bf395ad 100644
--- a/codex-rs/common/src/approval_presets.rs
+++ b/codex-rs/common/src/approval_presets.rs
@@ -24,21 +24,21 @@ pub fn builtin_approval_presets() -> Vec<ApprovalPreset> {
         ApprovalPreset {
             id: "read-only",
             label: "Read Only",
-            description: "Codex can read files and answer questions. Codex requires approval to make edits, run commands, or access network",
+            description: "Codex can read files and answer questions. Codex requires approval to make edits, run commands, or access network.",
             approval: AskForApproval::OnRequest,
             sandbox: SandboxPolicy::ReadOnly,
         },
         ApprovalPreset {
             id: "auto",
             label: "Auto",
-            description: "Codex can read files, make edits, and run commands in the workspace. Codex requires approval to work outside the workspace or access network",
+            description: "Codex can read files, make edits, and run commands in the workspace. Codex requires approval to work outside the workspace or access network.",
             approval: AskForApproval::OnRequest,
             sandbox: SandboxPolicy::new_workspace_write_policy(),
         },
         ApprovalPreset {
             id: "full-access",
             label: "Full Access",
-            description: "Codex can read files, make edits, and run commands with network access, without approval. Exercise caution",
+            description: "Codex can read files, make edits, and run commands with network access, without approval. Exercise caution.",
             approval: AskForApproval::Never,
             sandbox: SandboxPolicy::DangerFullAccess,
         },
diff --git a/codex-rs/core/Cargo.toml b/codex-rs/core/Cargo.toml
index 1962411b6e..3fc6ccd093 100644
--- a/codex-rs/core/Cargo.toml
+++ b/codex-rs/core/Cargo.toml
@@ -39,7 +39,12 @@ eventsource-stream = { workspace = true }
 futures = { workspace = true }
 http = { workspace = true }
 indexmap = { workspace = true }
-keyring = { workspace = true }
+keyring = { workspace = true, features = [
+    "apple-native",
+    "crypto-rust",
+    "linux-native-async-persistent",
+    "windows-native",
+] }
 libc = { workspace = true }
 mcp-types = { workspace = true }
 os_info = { workspace = true }
diff --git a/codex-rs/core/src/auth.rs b/codex-rs/core/src/auth.rs
index 49f9b77815..85fb172192 100644
--- a/codex-rs/core/src/auth.rs
+++ b/codex-rs/core/src/auth.rs
@@ -79,8 +79,11 @@ impl CodexAuth {
     }
 
     /// Loads the available auth information from auth storage.
-    pub fn from_auth_storage(codex_home: &Path) -> std::io::Result<Option<CodexAuth>> {
-        load_auth(codex_home, false)
+    pub fn from_auth_storage(
+        codex_home: &Path,
+        auth_credentials_store_mode: AuthCredentialsStoreMode,
+    ) -> std::io::Result<Option<CodexAuth>> {
+        load_auth(codex_home, false, auth_credentials_store_mode)
     }
 
     pub async fn get_token_data(&self) -> Result<TokenData, std::io::Error> {
@@ -217,36 +220,55 @@ pub fn read_codex_api_key_from_env() -> Option<String> {
 
 /// Delete the auth.json file inside `codex_home` if it exists. Returns `Ok(true)`
 /// if a file was removed, `Ok(false)` if no auth file was present.
-pub fn logout(codex_home: &Path) -> std::io::Result<bool> {
-    let storage = create_auth_storage(codex_home.to_path_buf(), AuthCredentialsStoreMode::File);
+pub fn logout(
+    codex_home: &Path,
+    auth_credentials_store_mode: AuthCredentialsStoreMode,
+) -> std::io::Result<bool> {
+    let storage = create_auth_storage(codex_home.to_path_buf(), auth_credentials_store_mode);
     storage.delete()
 }
 
 /// Writes an `auth.json` that contains only the API key.
-pub fn login_with_api_key(codex_home: &Path, api_key: &str) -> std::io::Result<()> {
+pub fn login_with_api_key(
+    codex_home: &Path,
+    api_key: &str,
+    auth_credentials_store_mode: AuthCredentialsStoreMode,
+) -> std::io::Result<()> {
     let auth_dot_json = AuthDotJson {
         openai_api_key: Some(api_key.to_string()),
         tokens: None,
         last_refresh: None,
     };
-    save_auth(codex_home, &auth_dot_json)
+    save_auth(codex_home, &auth_dot_json, auth_credentials_store_mode)
 }
 
 /// Persist the provided auth payload using the specified backend.
-pub fn save_auth(codex_home: &Path, auth: &AuthDotJson) -> std::io::Result<()> {
-    let storage = create_auth_storage(codex_home.to_path_buf(), AuthCredentialsStoreMode::File);
+pub fn save_auth(
+    codex_home: &Path,
+    auth: &AuthDotJson,
+    auth_credentials_store_mode: AuthCredentialsStoreMode,
+) -> std::io::Result<()> {
+    let storage = create_auth_storage(codex_home.to_path_buf(), auth_credentials_store_mode);
     storage.save(auth)
 }
 
 /// Load CLI auth data using the configured credential store backend.
 /// Returns `None` when no credentials are stored.
-pub fn load_auth_dot_json(codex_home: &Path) -> std::io::Result<Option<AuthDotJson>> {
-    let storage = create_auth_storage(codex_home.to_path_buf(), AuthCredentialsStoreMode::File);
+pub fn load_auth_dot_json(
+    codex_home: &Path,
+    auth_credentials_store_mode: AuthCredentialsStoreMode,
+) -> std::io::Result<Option<AuthDotJson>> {
+    let storage = create_auth_storage(codex_home.to_path_buf(), auth_credentials_store_mode);
     storage.load()
 }
 
 pub async fn enforce_login_restrictions(config: &Config) -> std::io::Result<()> {
-    let Some(auth) = load_auth(&config.codex_home, true)? else {
+    let Some(auth) = load_auth(
+        &config.codex_home,
+        true,
+        config.cli_auth_credentials_store_mode,
+    )?
+    else {
         return Ok(());
     };
 
@@ -265,7 +287,11 @@ pub async fn enforce_login_restrictions(config: &Config) -> std::io::Result<()>
         };
 
         if let Some(message) = method_violation {
-            return logout_with_message(&config.codex_home, message);
+            return logout_with_message(
+                &config.codex_home,
+                message,
+                config.cli_auth_credentials_store_mode,
+            );
         }
     }
 
@@ -282,6 +308,7 @@ pub async fn enforce_login_restrictions(config: &Config) -> std::io::Result<()>
                     format!(
                         "Failed to load ChatGPT credentials while enforcing workspace restrictions: {err}. Logging out."
                     ),
+                    config.cli_auth_credentials_store_mode,
                 );
             }
         };
@@ -297,15 +324,23 @@ pub async fn enforce_login_restrictions(config: &Config) -> std::io::Result<()>
                     "Login is restricted to workspace {expected_account_id}, but current credentials lack a workspace identifier. Logging out."
                 ),
             };
-            return logout_with_message(&config.codex_home, message);
+            return logout_with_message(
+                &config.codex_home,
+                message,
+                config.cli_auth_credentials_store_mode,
+            );
         }
     }
 
     Ok(())
 }
 
-fn logout_with_message(codex_home: &Path, message: String) -> std::io::Result<()> {
-    match logout(codex_home) {
+fn logout_with_message(
+    codex_home: &Path,
+    message: String,
+    auth_credentials_store_mode: AuthCredentialsStoreMode,
+) -> std::io::Result<()> {
+    match logout(codex_home, auth_credentials_store_mode) {
         Ok(_) => Err(std::io::Error::other(message)),
         Err(err) => Err(std::io::Error::other(format!(
             "{message}. Failed to remove auth.json: {err}"
@@ -316,6 +351,7 @@ fn logout_with_message(codex_home: &Path, message: String) -> std::io::Result<()
 fn load_auth(
     codex_home: &Path,
     enable_codex_api_key_env: bool,
+    auth_credentials_store_mode: AuthCredentialsStoreMode,
 ) -> std::io::Result<Option<CodexAuth>> {
     if enable_codex_api_key_env && let Some(api_key) = read_codex_api_key_from_env() {
         let client = crate::default_client::create_client();
@@ -325,7 +361,7 @@ fn load_auth(
         )));
     }
 
-    let storage = create_auth_storage(codex_home.to_path_buf(), AuthCredentialsStoreMode::File);
+    let storage = create_auth_storage(codex_home.to_path_buf(), auth_credentials_store_mode);
 
     let client = crate::default_client::create_client();
     let auth_dot_json = match storage.load()? {
@@ -512,7 +548,8 @@ mod tests {
         )
         .unwrap();
 
-        super::login_with_api_key(dir.path(), "sk-new").expect("login_with_api_key should succeed");
+        super::login_with_api_key(dir.path(), "sk-new", AuthCredentialsStoreMode::File)
+            .expect("login_with_api_key should succeed");
 
         let storage = FileAuthStorage::new(dir.path().to_path_buf());
         let auth = storage
@@ -525,7 +562,8 @@ mod tests {
     #[test]
     fn missing_auth_json_returns_none() {
         let dir = tempdir().unwrap();
-        let auth = CodexAuth::from_auth_storage(dir.path()).expect("call should succeed");
+        let auth = CodexAuth::from_auth_storage(dir.path(), AuthCredentialsStoreMode::File)
+            .expect("call should succeed");
         assert_eq!(auth, None);
     }
 
@@ -549,7 +587,9 @@ mod tests {
             auth_dot_json,
             storage: _,
             ..
-        } = super::load_auth(codex_home.path(), false).unwrap().unwrap();
+        } = super::load_auth(codex_home.path(), false, AuthCredentialsStoreMode::File)
+            .unwrap()
+            .unwrap();
         assert_eq!(None, api_key);
         assert_eq!(AuthMode::ChatGPT, mode);
 
@@ -590,7 +630,9 @@ mod tests {
         )
         .unwrap();
 
-        let auth = super::load_auth(dir.path(), false).unwrap().unwrap();
+        let auth = super::load_auth(dir.path(), false, AuthCredentialsStoreMode::File)
+            .unwrap()
+            .unwrap();
         assert_eq!(auth.mode, AuthMode::ApiKey);
         assert_eq!(auth.api_key, Some("sk-test-key".to_string()));
 
@@ -605,10 +647,10 @@ mod tests {
             tokens: None,
             last_refresh: None,
         };
-        super::save_auth(dir.path(), &auth_dot_json)?;
+        super::save_auth(dir.path(), &auth_dot_json, AuthCredentialsStoreMode::File)?;
         let auth_file = get_auth_file(dir.path());
         assert!(auth_file.exists());
-        assert!(logout(dir.path())?);
+        assert!(logout(dir.path(), AuthCredentialsStoreMode::File)?);
         assert!(!auth_file.exists());
         Ok(())
     }
@@ -717,7 +759,8 @@ mod tests {
     #[tokio::test]
     async fn enforce_login_restrictions_logs_out_for_method_mismatch() {
         let codex_home = tempdir().unwrap();
-        login_with_api_key(codex_home.path(), "sk-test").expect("seed api key");
+        login_with_api_key(codex_home.path(), "sk-test", AuthCredentialsStoreMode::File)
+            .expect("seed api key");
 
         let config = build_config(codex_home.path(), Some(ForcedLoginMethod::Chatgpt), None);
 
@@ -786,7 +829,8 @@ mod tests {
     async fn enforce_login_restrictions_allows_api_key_if_login_method_not_set_but_forced_chatgpt_workspace_id_is_set()
      {
         let codex_home = tempdir().unwrap();
-        login_with_api_key(codex_home.path(), "sk-test").expect("seed api key");
+        login_with_api_key(codex_home.path(), "sk-test", AuthCredentialsStoreMode::File)
+            .expect("seed api key");
 
         let config = build_config(codex_home.path(), None, Some("org_mine".to_string()));
 
@@ -830,6 +874,7 @@ pub struct AuthManager {
     codex_home: PathBuf,
     inner: RwLock<CachedAuth>,
     enable_codex_api_key_env: bool,
+    auth_credentials_store_mode: AuthCredentialsStoreMode,
 }
 
 impl AuthManager {
@@ -837,14 +882,23 @@ impl AuthManager {
     /// preferred auth method. Errors loading auth are swallowed; `auth()` will
     /// simply return `None` in that case so callers can treat it as an
     /// unauthenticated state.
-    pub fn new(codex_home: PathBuf, enable_codex_api_key_env: bool) -> Self {
-        let auth = load_auth(&codex_home, enable_codex_api_key_env)
-            .ok()
-            .flatten();
+    pub fn new(
+        codex_home: PathBuf,
+        enable_codex_api_key_env: bool,
+        auth_credentials_store_mode: AuthCredentialsStoreMode,
+    ) -> Self {
+        let auth = load_auth(
+            &codex_home,
+            enable_codex_api_key_env,
+            auth_credentials_store_mode,
+        )
+        .ok()
+        .flatten();
         Self {
             codex_home,
             inner: RwLock::new(CachedAuth { auth }),
             enable_codex_api_key_env,
+            auth_credentials_store_mode,
         }
     }
 
@@ -855,6 +909,7 @@ impl AuthManager {
             codex_home: PathBuf::new(),
             inner: RwLock::new(cached),
             enable_codex_api_key_env: false,
+            auth_credentials_store_mode: AuthCredentialsStoreMode::File,
         })
     }
 
@@ -866,9 +921,13 @@ impl AuthManager {
     /// Force a reload of the auth information from auth.json. Returns
     /// whether the auth value changed.
     pub fn reload(&self) -> bool {
-        let new_auth = load_auth(&self.codex_home, self.enable_codex_api_key_env)
-            .ok()
-            .flatten();
+        let new_auth = load_auth(
+            &self.codex_home,
+            self.enable_codex_api_key_env,
+            self.auth_credentials_store_mode,
+        )
+        .ok()
+        .flatten();
         if let Ok(mut guard) = self.inner.write() {
             let changed = !AuthManager::auths_equal(&guard.auth, &new_auth);
             guard.auth = new_auth;
@@ -887,8 +946,16 @@ impl AuthManager {
     }
 
     /// Convenience constructor returning an `Arc` wrapper.
-    pub fn shared(codex_home: PathBuf, enable_codex_api_key_env: bool) -> Arc<Self> {
-        Arc::new(Self::new(codex_home, enable_codex_api_key_env))
+    pub fn shared(
+        codex_home: PathBuf,
+        enable_codex_api_key_env: bool,
+        auth_credentials_store_mode: AuthCredentialsStoreMode,
+    ) -> Arc<Self> {
+        Arc::new(Self::new(
+            codex_home,
+            enable_codex_api_key_env,
+            auth_credentials_store_mode,
+        ))
     }
 
     /// Attempt to refresh the current auth token (if any). On success, reload
@@ -916,7 +983,7 @@ impl AuthManager {
     /// reloads the in‑memory auth cache so callers immediately observe the
     /// unauthenticated state.
     pub fn logout(&self) -> std::io::Result<bool> {
-        let removed = super::auth::logout(&self.codex_home)?;
+        let removed = super::auth::logout(&self.codex_home, self.auth_credentials_store_mode)?;
         // Always reload to clear any cached auth (even if file absent).
         self.reload();
         Ok(removed)
diff --git a/codex-rs/core/src/chat_completions.rs b/codex-rs/core/src/chat_completions.rs
index 76cbdf9f0a..1fb2230d79 100644
--- a/codex-rs/core/src/chat_completions.rs
+++ b/codex-rs/core/src/chat_completions.rs
@@ -17,6 +17,7 @@ use crate::util::backoff;
 use bytes::Bytes;
 use codex_otel::otel_event_manager::OtelEventManager;
 use codex_protocol::models::ContentItem;
+use codex_protocol::models::FunctionCallOutputContentItem;
 use codex_protocol::models::ReasoningItemContent;
 use codex_protocol::models::ResponseItem;
 use eventsource_stream::Eventsource;
@@ -159,16 +160,26 @@ pub(crate) async fn stream_chat_completions(
     for (idx, item) in input.iter().enumerate() {
         match item {
             ResponseItem::Message { role, content, .. } => {
+                // Build content either as a plain string (typical for assistant text)
+                // or as an array of content items when images are present (user/tool multimodal).
                 let mut text = String::new();
+                let mut items: Vec<serde_json::Value> = Vec::new();
+                let mut saw_image = false;
+
                 for c in content {
                     match c {
                         ContentItem::InputText { text: t }
                         | ContentItem::OutputText { text: t } => {
                             text.push_str(t);
+                            items.push(json!({"type":"text","text": t}));
+                        }
+                        ContentItem::InputImage { image_url } => {
+                            saw_image = true;
+                            items.push(json!({"type":"image_url","image_url": {"url": image_url}}));
                         }
-                        _ => {}
                     }
                 }
+
                 // Skip exact-duplicate assistant messages.
                 if role == "assistant" {
                     if let Some(prev) = &last_assistant_text
@@ -179,7 +190,17 @@ pub(crate) async fn stream_chat_completions(
                     last_assistant_text = Some(text.clone());
                 }
 
-                let mut msg = json!({"role": role, "content": text});
+                // For assistant messages, always send a plain string for compatibility.
+                // For user messages, if an image is present, send an array of content items.
+                let content_value = if role == "assistant" {
+                    json!(text)
+                } else if saw_image {
+                    json!(items)
+                } else {
+                    json!(text)
+                };
+
+                let mut msg = json!({"role": role, "content": content_value});
                 if role == "assistant"
                     && let Some(reasoning) = reasoning_by_anchor_index.get(&idx)
                     && let Some(obj) = msg.as_object_mut()
@@ -238,10 +259,29 @@ pub(crate) async fn stream_chat_completions(
                 messages.push(msg);
             }
             ResponseItem::FunctionCallOutput { call_id, output } => {
+                // Prefer structured content items when available (e.g., images)
+                // otherwise fall back to the legacy plain-string content.
+                let content_value = if let Some(items) = &output.content_items {
+                    let mapped: Vec<serde_json::Value> = items
+                        .iter()
+                        .map(|it| match it {
+                            FunctionCallOutputContentItem::InputText { text } => {
+                                json!({"type":"text","text": text})
+                            }
+                            FunctionCallOutputContentItem::InputImage { image_url } => {
+                                json!({"type":"image_url","image_url": {"url": image_url}})
+                            }
+                        })
+                        .collect();
+                    json!(mapped)
+                } else {
+                    json!(output.content)
+                };
+
                 messages.push(json!({
                     "role": "tool",
                     "tool_call_id": call_id,
-                    "content": output.content,
+                    "content": content_value,
                 }));
             }
             ResponseItem::CustomToolCall {
diff --git a/codex-rs/core/src/codex.rs b/codex-rs/core/src/codex.rs
index f811211c4f..47d80c5cb4 100644
--- a/codex-rs/core/src/codex.rs
+++ b/codex-rs/core/src/codex.rs
@@ -82,7 +82,6 @@ use crate::protocol::ErrorEvent;
 use crate::protocol::Event;
 use crate::protocol::EventMsg;
 use crate::protocol::ExecApprovalRequestEvent;
-use crate::protocol::ListCustomPromptsResponseEvent;
 use crate::protocol::Op;
 use crate::protocol::RateLimitSnapshot;
 use crate::protocol::ReviewDecision;
@@ -103,13 +102,10 @@ use crate::state::ActiveTurn;
 use crate::state::SessionServices;
 use crate::state::SessionState;
 use crate::state::TaskKind;
-use crate::tasks::CompactTask;
 use crate::tasks::GhostSnapshotTask;
-use crate::tasks::RegularTask;
 use crate::tasks::ReviewTask;
 use crate::tasks::SessionTask;
 use crate::tasks::SessionTaskContext;
-use crate::tasks::UndoTask;
 use crate::tools::ToolRouter;
 use crate::tools::context::SharedTurnDiffTracker;
 use crate::tools::parallel::ToolCallRuntime;
@@ -125,7 +121,6 @@ use codex_async_utils::OrCancelExt;
 use codex_otel::otel_event_manager::OtelEventManager;
 use codex_protocol::config_types::ReasoningEffort as ReasoningEffortConfig;
 use codex_protocol::config_types::ReasoningSummary as ReasoningSummaryConfig;
-use codex_protocol::custom_prompts::CustomPrompt;
 use codex_protocol::models::ContentItem;
 use codex_protocol::models::FunctionCallOutputPayload;
 use codex_protocol::models::ResponseInputItem;
@@ -1243,9 +1238,9 @@ async fn submission_loop(sess: Arc<Session>, config: Arc<Config>, rx_sub: Receiv
     // To break out of this loop, send Op::Shutdown.
     while let Ok(sub) = rx_sub.recv().await {
         debug!(?sub, "Submission");
-        match sub.op {
+        match sub.op.clone() {
             Op::Interrupt => {
-                sess.interrupt_task().await;
+                handlers::interrupt(&sess).await;
             }
             Op::OverrideTurnContext {
                 cwd,
@@ -1255,246 +1250,336 @@ async fn submission_loop(sess: Arc<Session>, config: Arc<Config>, rx_sub: Receiv
                 effort,
                 summary,
             } => {
-                let updates = SessionSettingsUpdate {
-                    cwd,
-                    approval_policy,
-                    sandbox_policy,
-                    model,
-                    reasoning_effort: effort,
-                    reasoning_summary: summary,
-                    ..Default::default()
-                };
-                sess.update_settings(updates).await;
-            }
-
-            Op::UserInput { .. } | Op::UserTurn { .. } => {
-                let (items, updates) = match sub.op {
-                    Op::UserTurn {
+                handlers::override_turn_context(
+                    &sess,
+                    SessionSettingsUpdate {
                         cwd,
                         approval_policy,
                         sandbox_policy,
                         model,
-                        effort,
-                        summary,
-                        final_output_json_schema,
-                        items,
-                    } => (
-                        items,
-                        SessionSettingsUpdate {
-                            cwd: Some(cwd),
-                            approval_policy: Some(approval_policy),
-                            sandbox_policy: Some(sandbox_policy),
-                            model: Some(model),
-                            reasoning_effort: Some(effort),
-                            reasoning_summary: Some(summary),
-                            final_output_json_schema: Some(final_output_json_schema),
-                        },
-                    ),
-                    Op::UserInput { items } => (items, SessionSettingsUpdate::default()),
-                    _ => unreachable!(),
-                };
-                let current_context = sess.new_turn_with_sub_id(sub.id.clone(), updates).await;
-                current_context
-                    .client
-                    .get_otel_event_manager()
-                    .user_prompt(&items);
-                // attempt to inject input into current task
-                if let Err(items) = sess.inject_input(items).await {
-                    if let Some(env_item) = sess
-                        .build_environment_update_item(previous_context.as_ref(), &current_context)
-                    {
-                        sess.record_conversation_items(
-                            &current_context,
-                            std::slice::from_ref(&env_item),
-                        )
-                        .await;
-                    }
-
-                    sess.spawn_task(Arc::clone(&current_context), items, RegularTask)
-                        .await;
-                    previous_context = Some(current_context);
-                }
-            }
-            Op::ExecApproval { id, decision } => match decision {
-                ReviewDecision::Abort => {
-                    sess.interrupt_task().await;
-                }
-                other => sess.notify_approval(&id, other).await,
-            },
-            Op::PatchApproval { id, decision } => match decision {
-                ReviewDecision::Abort => {
-                    sess.interrupt_task().await;
-                }
-                other => sess.notify_approval(&id, other).await,
-            },
-            Op::AddToHistory { text } => {
-                let id = sess.conversation_id;
-                let config = config.clone();
-                tokio::spawn(async move {
-                    if let Err(e) = crate::message_history::append_entry(&text, &id, &config).await
-                    {
-                        warn!("failed to append to message history: {e}");
-                    }
-                });
-            }
-
-            Op::GetHistoryEntryRequest { offset, log_id } => {
-                let config = config.clone();
-                let sess_clone = sess.clone();
-                let sub_id = sub.id.clone();
-
-                tokio::spawn(async move {
-                    // Run lookup in blocking thread because it does file IO + locking.
-                    let entry_opt = tokio::task::spawn_blocking(move || {
-                        crate::message_history::lookup(log_id, offset, &config)
-                    })
-                    .await
-                    .unwrap_or(None);
-
-                    let event = Event {
-                        id: sub_id,
-                        msg: EventMsg::GetHistoryEntryResponse(
-                            crate::protocol::GetHistoryEntryResponseEvent {
-                                offset,
-                                log_id,
-                                entry: entry_opt.map(|e| {
-                                    codex_protocol::message_history::HistoryEntry {
-                                        conversation_id: e.session_id,
-                                        ts: e.ts,
-                                        text: e.text,
-                                    }
-                                }),
-                            },
-                        ),
-                    };
-
-                    sess_clone.send_event_raw(event).await;
-                });
-            }
-            Op::ListMcpTools => {
-                let sub_id = sub.id.clone();
-
-                // This is a cheap lookup from the connection manager's cache.
-                let tools = sess.services.mcp_connection_manager.list_all_tools();
-                let (auth_status_entries, resources, resource_templates) = tokio::join!(
-                    compute_auth_statuses(
-                        config.mcp_servers.iter(),
-                        config.mcp_oauth_credentials_store_mode,
-                    ),
-                    sess.services.mcp_connection_manager.list_all_resources(),
-                    sess.services
-                        .mcp_connection_manager
-                        .list_all_resource_templates()
-                );
-                let auth_statuses = auth_status_entries
-                    .iter()
-                    .map(|(name, entry)| (name.clone(), entry.auth_status))
-                    .collect();
-                let event = Event {
-                    id: sub_id,
-                    msg: EventMsg::McpListToolsResponse(
-                        crate::protocol::McpListToolsResponseEvent {
-                            tools,
-                            resources,
-                            resource_templates,
-                            auth_statuses,
-                        },
-                    ),
-                };
-                sess.send_event_raw(event).await;
-            }
-            Op::ListCustomPrompts => {
-                let sub_id = sub.id.clone();
-
-                let custom_prompts: Vec<CustomPrompt> =
-                    if let Some(dir) = crate::custom_prompts::default_prompts_dir() {
-                        crate::custom_prompts::discover_prompts_in(&dir).await
-                    } else {
-                        Vec::new()
-                    };
-
-                let event = Event {
-                    id: sub_id,
-                    msg: EventMsg::ListCustomPromptsResponse(ListCustomPromptsResponseEvent {
-                        custom_prompts,
-                    }),
-                };
-                sess.send_event_raw(event).await;
-            }
-            Op::Undo => {
-                let turn_context = sess
-                    .new_turn_with_sub_id(sub.id.clone(), SessionSettingsUpdate::default())
-                    .await;
-                sess.spawn_task(turn_context, Vec::new(), UndoTask::new())
-                    .await;
-            }
-            Op::Compact => {
-                let turn_context = sess
-                    .new_turn_with_sub_id(sub.id.clone(), SessionSettingsUpdate::default())
-                    .await;
-                // Attempt to inject input into current task
-                if let Err(items) = sess
-                    .inject_input(vec![UserInput::Text {
-                        text: compact::SUMMARIZATION_PROMPT.to_string(),
-                    }])
-                    .await
-                {
-                    sess.spawn_task(Arc::clone(&turn_context), items, CompactTask)
-                        .await;
-                }
-            }
-            Op::Shutdown => {
-                sess.abort_all_tasks(TurnAbortReason::Interrupted).await;
-                info!("Shutting down Codex instance");
-
-                // Gracefully flush and shutdown rollout recorder on session end so tests
-                // that inspect the rollout file do not race with the background writer.
-                let recorder_opt = {
-                    let mut guard = sess.services.rollout.lock().await;
-                    guard.take()
-                };
-                if let Some(rec) = recorder_opt
-                    && let Err(e) = rec.shutdown().await
-                {
-                    warn!("failed to shutdown rollout recorder: {e}");
-                    let event = Event {
-                        id: sub.id.clone(),
-                        msg: EventMsg::Error(ErrorEvent {
-                            message: "Failed to shutdown rollout recorder".to_string(),
-                        }),
-                    };
-                    sess.send_event_raw(event).await;
-                }
-
-                let event = Event {
-                    id: sub.id.clone(),
-                    msg: EventMsg::ShutdownComplete,
-                };
-                sess.send_event_raw(event).await;
-                break;
-            }
-
-            Op::Review { review_request } => {
-                let turn_context = sess
-                    .new_turn_with_sub_id(sub.id.clone(), SessionSettingsUpdate::default())
-                    .await;
-                spawn_review_thread(
-                    sess.clone(),
-                    config.clone(),
-                    turn_context.clone(),
-                    sub.id,
-                    review_request,
+                        reasoning_effort: effort,
+                        reasoning_summary: summary,
+                        ..Default::default()
+                    },
                 )
                 .await;
             }
-            _ => {
-                // Ignore unknown ops; enum is non_exhaustive to allow extensions.
+            Op::UserInput { .. } | Op::UserTurn { .. } => {
+                handlers::user_input_or_turn(&sess, sub.id.clone(), sub.op, &mut previous_context)
+                    .await;
             }
+            Op::ExecApproval { id, decision } => {
+                handlers::exec_approval(&sess, id, decision).await;
+            }
+            Op::PatchApproval { id, decision } => {
+                handlers::patch_approval(&sess, id, decision).await;
+            }
+            Op::AddToHistory { text } => {
+                handlers::add_to_history(&sess, &config, text).await;
+            }
+            Op::GetHistoryEntryRequest { offset, log_id } => {
+                handlers::get_history_entry_request(&sess, &config, sub.id.clone(), offset, log_id)
+                    .await;
+            }
+            Op::ListMcpTools => {
+                handlers::list_mcp_tools(&sess, &config, sub.id.clone()).await;
+            }
+            Op::ListCustomPrompts => {
+                handlers::list_custom_prompts(&sess, sub.id.clone()).await;
+            }
+            Op::Undo => {
+                handlers::undo(&sess, sub.id.clone()).await;
+            }
+            Op::Compact => {
+                handlers::compact(&sess, sub.id.clone()).await;
+            }
+            Op::Shutdown => {
+                if handlers::shutdown(&sess, sub.id.clone()).await {
+                    break;
+                }
+            }
+            Op::Review { review_request } => {
+                handlers::review(&sess, &config, sub.id.clone(), review_request).await;
+            }
+            _ => {} // Ignore unknown ops; enum is non_exhaustive to allow extensions.
         }
     }
     debug!("Agent loop exited");
 }
 
+/// Operation handlers
+mod handlers {
+    use crate::codex::Session;
+    use crate::codex::SessionSettingsUpdate;
+    use crate::codex::TurnContext;
+    use crate::codex::compact;
+    use crate::codex::spawn_review_thread;
+    use crate::config::Config;
+    use crate::mcp::auth::compute_auth_statuses;
+    use crate::tasks::CompactTask;
+    use crate::tasks::RegularTask;
+    use crate::tasks::UndoTask;
+    use codex_protocol::custom_prompts::CustomPrompt;
+    use codex_protocol::protocol::ErrorEvent;
+    use codex_protocol::protocol::Event;
+    use codex_protocol::protocol::EventMsg;
+    use codex_protocol::protocol::ListCustomPromptsResponseEvent;
+    use codex_protocol::protocol::Op;
+    use codex_protocol::protocol::ReviewDecision;
+    use codex_protocol::protocol::ReviewRequest;
+    use codex_protocol::protocol::TurnAbortReason;
+    use codex_protocol::user_input::UserInput;
+    use std::sync::Arc;
+    use tracing::info;
+    use tracing::warn;
+
+    pub async fn interrupt(sess: &Arc<Session>) {
+        sess.interrupt_task().await;
+    }
+
+    pub async fn override_turn_context(sess: &Session, updates: SessionSettingsUpdate) {
+        sess.update_settings(updates).await;
+    }
+
+    pub async fn user_input_or_turn(
+        sess: &Arc<Session>,
+        sub_id: String,
+        op: Op,
+        previous_context: &mut Option<Arc<TurnContext>>,
+    ) {
+        let (items, updates) = match op {
+            Op::UserTurn {
+                cwd,
+                approval_policy,
+                sandbox_policy,
+                model,
+                effort,
+                summary,
+                final_output_json_schema,
+                items,
+            } => (
+                items,
+                SessionSettingsUpdate {
+                    cwd: Some(cwd),
+                    approval_policy: Some(approval_policy),
+                    sandbox_policy: Some(sandbox_policy),
+                    model: Some(model),
+                    reasoning_effort: Some(effort),
+                    reasoning_summary: Some(summary),
+                    final_output_json_schema: Some(final_output_json_schema),
+                },
+            ),
+            Op::UserInput { items } => (items, SessionSettingsUpdate::default()),
+            _ => unreachable!(),
+        };
+
+        let current_context = sess.new_turn_with_sub_id(sub_id, updates).await;
+        current_context
+            .client
+            .get_otel_event_manager()
+            .user_prompt(&items);
+
+        // Attempt to inject input into current task
+        if let Err(items) = sess.inject_input(items).await {
+            if let Some(env_item) =
+                sess.build_environment_update_item(previous_context.as_ref(), &current_context)
+            {
+                sess.record_conversation_items(&current_context, std::slice::from_ref(&env_item))
+                    .await;
+            }
+
+            sess.spawn_task(Arc::clone(&current_context), items, RegularTask)
+                .await;
+            *previous_context = Some(current_context);
+        }
+    }
+
+    pub async fn exec_approval(sess: &Arc<Session>, id: String, decision: ReviewDecision) {
+        match decision {
+            ReviewDecision::Abort => {
+                sess.interrupt_task().await;
+            }
+            other => sess.notify_approval(&id, other).await,
+        }
+    }
+
+    pub async fn patch_approval(sess: &Arc<Session>, id: String, decision: ReviewDecision) {
+        match decision {
+            ReviewDecision::Abort => {
+                sess.interrupt_task().await;
+            }
+            other => sess.notify_approval(&id, other).await,
+        }
+    }
+
+    pub async fn add_to_history(sess: &Arc<Session>, config: &Arc<Config>, text: String) {
+        let id = sess.conversation_id;
+        let config = Arc::clone(config);
+        tokio::spawn(async move {
+            if let Err(e) = crate::message_history::append_entry(&text, &id, &config).await {
+                warn!("failed to append to message history: {e}");
+            }
+        });
+    }
+
+    pub async fn get_history_entry_request(
+        sess: &Arc<Session>,
+        config: &Arc<Config>,
+        sub_id: String,
+        offset: usize,
+        log_id: u64,
+    ) {
+        let config = Arc::clone(config);
+        let sess_clone = Arc::clone(sess);
+
+        tokio::spawn(async move {
+            // Run lookup in blocking thread because it does file IO + locking.
+            let entry_opt = tokio::task::spawn_blocking(move || {
+                crate::message_history::lookup(log_id, offset, &config)
+            })
+            .await
+            .unwrap_or(None);
+
+            let event = Event {
+                id: sub_id,
+                msg: EventMsg::GetHistoryEntryResponse(
+                    crate::protocol::GetHistoryEntryResponseEvent {
+                        offset,
+                        log_id,
+                        entry: entry_opt.map(|e| codex_protocol::message_history::HistoryEntry {
+                            conversation_id: e.session_id,
+                            ts: e.ts,
+                            text: e.text,
+                        }),
+                    },
+                ),
+            };
+
+            sess_clone.send_event_raw(event).await;
+        });
+    }
+
+    pub async fn list_mcp_tools(sess: &Session, config: &Arc<Config>, sub_id: String) {
+        // This is a cheap lookup from the connection manager's cache.
+        let tools = sess.services.mcp_connection_manager.list_all_tools();
+        let (auth_status_entries, resources, resource_templates) = tokio::join!(
+            compute_auth_statuses(
+                config.mcp_servers.iter(),
+                config.mcp_oauth_credentials_store_mode,
+            ),
+            sess.services.mcp_connection_manager.list_all_resources(),
+            sess.services
+                .mcp_connection_manager
+                .list_all_resource_templates()
+        );
+        let auth_statuses = auth_status_entries
+            .iter()
+            .map(|(name, entry)| (name.clone(), entry.auth_status))
+            .collect();
+        let event = Event {
+            id: sub_id,
+            msg: EventMsg::McpListToolsResponse(crate::protocol::McpListToolsResponseEvent {
+                tools,
+                resources,
+                resource_templates,
+                auth_statuses,
+            }),
+        };
+        sess.send_event_raw(event).await;
+    }
+
+    pub async fn list_custom_prompts(sess: &Session, sub_id: String) {
+        let custom_prompts: Vec<CustomPrompt> =
+            if let Some(dir) = crate::custom_prompts::default_prompts_dir() {
+                crate::custom_prompts::discover_prompts_in(&dir).await
+            } else {
+                Vec::new()
+            };
+
+        let event = Event {
+            id: sub_id,
+            msg: EventMsg::ListCustomPromptsResponse(ListCustomPromptsResponseEvent {
+                custom_prompts,
+            }),
+        };
+        sess.send_event_raw(event).await;
+    }
+
+    pub async fn undo(sess: &Arc<Session>, sub_id: String) {
+        let turn_context = sess
+            .new_turn_with_sub_id(sub_id, SessionSettingsUpdate::default())
+            .await;
+        sess.spawn_task(turn_context, Vec::new(), UndoTask::new())
+            .await;
+    }
+
+    pub async fn compact(sess: &Arc<Session>, sub_id: String) {
+        let turn_context = sess
+            .new_turn_with_sub_id(sub_id, SessionSettingsUpdate::default())
+            .await;
+        // Attempt to inject input into current task
+        if let Err(items) = sess
+            .inject_input(vec![UserInput::Text {
+                text: compact::SUMMARIZATION_PROMPT.to_string(),
+            }])
+            .await
+        {
+            sess.spawn_task(Arc::clone(&turn_context), items, CompactTask)
+                .await;
+        }
+    }
+
+    pub async fn shutdown(sess: &Arc<Session>, sub_id: String) -> bool {
+        sess.abort_all_tasks(TurnAbortReason::Interrupted).await;
+        info!("Shutting down Codex instance");
+
+        // Gracefully flush and shutdown rollout recorder on session end so tests
+        // that inspect the rollout file do not race with the background writer.
+        let recorder_opt = {
+            let mut guard = sess.services.rollout.lock().await;
+            guard.take()
+        };
+        if let Some(rec) = recorder_opt
+            && let Err(e) = rec.shutdown().await
+        {
+            warn!("failed to shutdown rollout recorder: {e}");
+            let event = Event {
+                id: sub_id.clone(),
+                msg: EventMsg::Error(ErrorEvent {
+                    message: "Failed to shutdown rollout recorder".to_string(),
+                }),
+            };
+            sess.send_event_raw(event).await;
+        }
+
+        let event = Event {
+            id: sub_id,
+            msg: EventMsg::ShutdownComplete,
+        };
+        sess.send_event_raw(event).await;
+        true
+    }
+
+    pub async fn review(
+        sess: &Arc<Session>,
+        config: &Arc<Config>,
+        sub_id: String,
+        review_request: ReviewRequest,
+    ) {
+        let turn_context = sess
+            .new_turn_with_sub_id(sub_id.clone(), SessionSettingsUpdate::default())
+            .await;
+        spawn_review_thread(
+            Arc::clone(sess),
+            Arc::clone(config),
+            turn_context.clone(),
+            sub_id,
+            review_request,
+        )
+        .await;
+    }
+}
+
 /// Spawn a review thread using the given prompt.
 async fn spawn_review_thread(
     sess: Arc<Session>,
@@ -2047,7 +2132,7 @@ async fn try_run_turn(
                             call_id: String::new(),
                             output: FunctionCallOutputPayload {
                                 content: msg.to_string(),
-                                success: None,
+                                ..Default::default()
                             },
                         };
                         add_completed(ProcessedResponseItem {
@@ -2061,7 +2146,7 @@ async fn try_run_turn(
                             call_id: String::new(),
                             output: FunctionCallOutputPayload {
                                 content: message,
-                                success: None,
+                                ..Default::default()
                             },
                         };
                         add_completed(ProcessedResponseItem {
@@ -2199,41 +2284,6 @@ pub(super) fn get_last_assistant_message_from_turn(responses: &[ResponseItem]) -
         }
     })
 }
-pub(crate) fn convert_call_tool_result_to_function_call_output_payload(
-    call_tool_result: &CallToolResult,
-) -> FunctionCallOutputPayload {
-    let CallToolResult {
-        content,
-        is_error,
-        structured_content,
-    } = call_tool_result;
-
-    // In terms of what to send back to the model, we prefer structured_content,
-    // if available, and fallback to content, otherwise.
-    let mut is_success = is_error != &Some(true);
-    let content = if let Some(structured_content) = structured_content
-        && structured_content != &serde_json::Value::Null
-        && let Ok(serialized_structured_content) = serde_json::to_string(&structured_content)
-    {
-        serialized_structured_content
-    } else {
-        match serde_json::to_string(&content) {
-            Ok(serialized_content) => serialized_content,
-            Err(err) => {
-                // If we could not serialize either content or structured_content to
-                // JSON, flag this as an error.
-                is_success = false;
-                err.to_string()
-            }
-        }
-    };
-
-    FunctionCallOutputPayload {
-        content,
-        success: Some(is_success),
-    }
-}
-
 /// Emits an ExitedReviewMode Event with optional ReviewOutput,
 /// and records a developer message with the review output.
 pub(crate) async fn exit_review_mode(
@@ -2366,10 +2416,6 @@ mod tests {
     use crate::state::TaskKind;
     use crate::tasks::SessionTask;
     use crate::tasks::SessionTaskContext;
-    use crate::tools::MODEL_FORMAT_HEAD_LINES;
-    use crate::tools::MODEL_FORMAT_MAX_BYTES;
-    use crate::tools::MODEL_FORMAT_MAX_LINES;
-    use crate::tools::MODEL_FORMAT_TAIL_LINES;
     use crate::tools::ToolRouter;
     use crate::tools::context::ToolInvocation;
     use crate::tools::context::ToolOutput;
@@ -2443,7 +2489,7 @@ mod tests {
             })),
         };
 
-        let got = convert_call_tool_result_to_function_call_output_payload(&ctr);
+        let got = FunctionCallOutputPayload::from(&ctr);
         let expected = FunctionCallOutputPayload {
             content: serde_json::to_string(&json!({
                 "ok": true,
@@ -2451,100 +2497,12 @@ mod tests {
             }))
             .unwrap(),
             success: Some(true),
+            ..Default::default()
         };
 
         assert_eq!(expected, got);
     }
 
-    #[test]
-    fn model_truncation_head_tail_by_lines() {
-        // Build 400 short lines so line-count limit, not byte budget, triggers truncation
-        let lines: Vec<String> = (1..=400).map(|i| format!("line{i}")).collect();
-        let full = lines.join("\n");
-
-        let exec = ExecToolCallOutput {
-            exit_code: 0,
-            stdout: StreamOutput::new(String::new()),
-            stderr: StreamOutput::new(String::new()),
-            aggregated_output: StreamOutput::new(full),
-            duration: StdDuration::from_secs(1),
-            timed_out: false,
-        };
-
-        let out = format_exec_output_str(&exec);
-
-        // Strip truncation header if present for subsequent assertions
-        let body = out
-            .strip_prefix("Total output lines: ")
-            .and_then(|rest| rest.split_once("\n\n").map(|x| x.1))
-            .unwrap_or(out.as_str());
-
-        // Expect elision marker with correct counts
-        let omitted = 400 - MODEL_FORMAT_MAX_LINES; // 144
-        let marker = format!("\n[... omitted {omitted} of 400 lines ...]\n\n");
-        assert!(out.contains(&marker), "missing marker: {out}");
-
-        // Validate head and tail
-        let parts: Vec<&str> = body.split(&marker).collect();
-        assert_eq!(parts.len(), 2, "expected one marker split");
-        let head = parts[0];
-        let tail = parts[1];
-
-        let expected_head: String = (1..=MODEL_FORMAT_HEAD_LINES)
-            .map(|i| format!("line{i}"))
-            .collect::<Vec<_>>()
-            .join("\n");
-        assert!(head.starts_with(&expected_head), "head mismatch");
-
-        let expected_tail: String = ((400 - MODEL_FORMAT_TAIL_LINES + 1)..=400)
-            .map(|i| format!("line{i}"))
-            .collect::<Vec<_>>()
-            .join("\n");
-        assert!(tail.ends_with(&expected_tail), "tail mismatch");
-    }
-
-    #[test]
-    fn model_truncation_respects_byte_budget() {
-        // Construct a large output (about 100kB) so byte budget dominates
-        let big_line = "x".repeat(100);
-        let full = std::iter::repeat_n(big_line, 1000)
-            .collect::<Vec<_>>()
-            .join("\n");
-
-        let exec = ExecToolCallOutput {
-            exit_code: 0,
-            stdout: StreamOutput::new(String::new()),
-            stderr: StreamOutput::new(String::new()),
-            aggregated_output: StreamOutput::new(full.clone()),
-            duration: StdDuration::from_secs(1),
-            timed_out: false,
-        };
-
-        let out = format_exec_output_str(&exec);
-        // Keep strict budget on the truncated body (excluding header)
-        let body = out
-            .strip_prefix("Total output lines: ")
-            .and_then(|rest| rest.split_once("\n\n").map(|x| x.1))
-            .unwrap_or(out.as_str());
-        assert!(body.len() <= MODEL_FORMAT_MAX_BYTES, "exceeds byte budget");
-        assert!(out.contains("omitted"), "should contain elision marker");
-
-        // Ensure head and tail are drawn from the original
-        assert!(full.starts_with(body.chars().take(8).collect::<String>().as_str()));
-        assert!(
-            full.ends_with(
-                body.chars()
-                    .rev()
-                    .take(8)
-                    .collect::<String>()
-                    .chars()
-                    .rev()
-                    .collect::<String>()
-                    .as_str()
-            )
-        );
-    }
-
     #[test]
     fn includes_timed_out_message() {
         let exec = ExecToolCallOutput {
@@ -2572,11 +2530,12 @@ mod tests {
             structured_content: Some(serde_json::Value::Null),
         };
 
-        let got = convert_call_tool_result_to_function_call_output_payload(&ctr);
+        let got = FunctionCallOutputPayload::from(&ctr);
         let expected = FunctionCallOutputPayload {
             content: serde_json::to_string(&vec![text_block("hello"), text_block("world")])
                 .unwrap(),
             success: Some(true),
+            ..Default::default()
         };
 
         assert_eq!(expected, got);
@@ -2590,10 +2549,11 @@ mod tests {
             structured_content: Some(json!({ "message": "bad" })),
         };
 
-        let got = convert_call_tool_result_to_function_call_output_payload(&ctr);
+        let got = FunctionCallOutputPayload::from(&ctr);
         let expected = FunctionCallOutputPayload {
             content: serde_json::to_string(&json!({ "message": "bad" })).unwrap(),
             success: Some(false),
+            ..Default::default()
         };
 
         assert_eq!(expected, got);
@@ -2607,10 +2567,11 @@ mod tests {
             structured_content: None,
         };
 
-        let got = convert_call_tool_result_to_function_call_output_payload(&ctr);
+        let got = FunctionCallOutputPayload::from(&ctr);
         let expected = FunctionCallOutputPayload {
             content: serde_json::to_string(&vec![text_block("alpha")]).unwrap(),
             success: Some(true),
+            ..Default::default()
         };
 
         assert_eq!(expected, got);
@@ -2649,7 +2610,11 @@ mod tests {
         let config = Arc::new(config);
         let conversation_id = ConversationId::default();
         let otel_event_manager = otel_event_manager(conversation_id, config.as_ref());
-        let auth_manager = AuthManager::shared(config.cwd.clone(), false);
+        let auth_manager = AuthManager::shared(
+            config.cwd.clone(),
+            false,
+            config.cli_auth_credentials_store_mode,
+        );
 
         let session_configuration = SessionConfiguration {
             provider: config.model_provider.clone(),
@@ -2718,7 +2683,11 @@ mod tests {
         let config = Arc::new(config);
         let conversation_id = ConversationId::default();
         let otel_event_manager = otel_event_manager(conversation_id, config.as_ref());
-        let auth_manager = AuthManager::shared(config.cwd.clone(), false);
+        let auth_manager = AuthManager::shared(
+            config.cwd.clone(),
+            false,
+            config.cli_auth_credentials_store_mode,
+        );
 
         let session_configuration = SessionConfiguration {
             provider: config.model_provider.clone(),
diff --git a/codex-rs/core/src/config.rs b/codex-rs/core/src/config.rs
index c5406b1d27..f671e8dd36 100644
--- a/codex-rs/core/src/config.rs
+++ b/codex-rs/core/src/config.rs
@@ -1,3 +1,4 @@
+use crate::auth::AuthCredentialsStoreMode;
 use crate::config_loader::LoadedConfigLayers;
 pub use crate::config_loader::load_config_as_toml;
 use crate::config_loader::load_config_layers_with_overrides;
@@ -108,6 +109,10 @@ pub struct Config {
     /// for either of approval_policy or sandbox_mode.
     pub did_user_set_custom_approval_policy_or_sandbox_mode: bool,
 
+    /// On Windows, indicates that a previously configured workspace-write sandbox
+    /// was coerced to read-only because native auto mode is unsupported.
+    pub forced_auto_mode_downgraded_on_windows: bool,
+
     pub shell_environment_policy: ShellEnvironmentPolicy,
 
     /// When `true`, `AgentReasoning` events emitted by the backend will be
@@ -156,6 +161,12 @@ pub struct Config {
     /// resolved against this path.
     pub cwd: PathBuf,
 
+    /// Preferred store for CLI auth credentials.
+    /// file (default): Use a file in the Codex home directory.
+    /// keyring: Use an OS-specific keyring service.
+    /// auto: Use the OS-specific keyring service if available, otherwise use a file.
+    pub cli_auth_credentials_store_mode: AuthCredentialsStoreMode,
+
     /// Definition for MCP servers that Codex can reach out to for tool calls.
     pub mcp_servers: HashMap<String, McpServerConfig>,
 
@@ -869,6 +880,13 @@ pub struct ConfigToml {
     #[serde(default)]
     pub forced_login_method: Option<ForcedLoginMethod>,
 
+    /// Preferred backend for storing CLI auth credentials.
+    /// file (default): Use a file in the Codex home directory.
+    /// keyring: Use an OS-specific keyring service.
+    /// auto: Use the keyring if available, otherwise use a file.
+    #[serde(default)]
+    pub cli_auth_credentials_store: Option<AuthCredentialsStoreMode>,
+
     /// Definition for MCP servers that Codex can reach out to for tool calls.
     #[serde(default)]
     pub mcp_servers: HashMap<String, McpServerConfig>,
@@ -1022,6 +1040,12 @@ impl From<ToolsToml> for Tools {
     }
 }
 
+#[derive(Debug, PartialEq, Eq)]
+pub struct SandboxPolicyResolution {
+    pub policy: SandboxPolicy,
+    pub forced_auto_mode_downgraded_on_windows: bool,
+}
+
 impl ConfigToml {
     /// Derive the effective sandbox policy from the configuration.
     fn derive_sandbox_policy(
@@ -1029,7 +1053,7 @@ impl ConfigToml {
         sandbox_mode_override: Option<SandboxMode>,
         profile_sandbox_mode: Option<SandboxMode>,
         resolved_cwd: &Path,
-    ) -> SandboxPolicy {
+    ) -> SandboxPolicyResolution {
         let resolved_sandbox_mode = sandbox_mode_override
             .or(profile_sandbox_mode)
             .or(self.sandbox_mode)
@@ -1044,7 +1068,7 @@ impl ConfigToml {
                 })
             })
             .unwrap_or_default();
-        match resolved_sandbox_mode {
+        let mut sandbox_policy = match resolved_sandbox_mode {
             SandboxMode::ReadOnly => SandboxPolicy::new_read_only_policy(),
             SandboxMode::WorkspaceWrite => match self.sandbox_workspace_write.as_ref() {
                 Some(SandboxWorkspaceWrite {
@@ -1061,6 +1085,17 @@ impl ConfigToml {
                 None => SandboxPolicy::new_workspace_write_policy(),
             },
             SandboxMode::DangerFullAccess => SandboxPolicy::DangerFullAccess,
+        };
+        let mut forced_auto_mode_downgraded_on_windows = false;
+        if cfg!(target_os = "windows")
+            && matches!(resolved_sandbox_mode, SandboxMode::WorkspaceWrite)
+        {
+            sandbox_policy = SandboxPolicy::new_read_only_policy();
+            forced_auto_mode_downgraded_on_windows = true;
+        }
+        SandboxPolicyResolution {
+            policy: sandbox_policy,
+            forced_auto_mode_downgraded_on_windows,
         }
     }
 
@@ -1221,8 +1256,10 @@ impl Config {
             .get_active_project(&resolved_cwd)
             .unwrap_or(ProjectConfig { trust_level: None });
 
-        let mut sandbox_policy =
-            cfg.derive_sandbox_policy(sandbox_mode, config_profile.sandbox_mode, &resolved_cwd);
+        let SandboxPolicyResolution {
+            policy: mut sandbox_policy,
+            forced_auto_mode_downgraded_on_windows,
+        } = cfg.derive_sandbox_policy(sandbox_mode, config_profile.sandbox_mode, &resolved_cwd);
         if let SandboxPolicy::WorkspaceWrite { writable_roots, .. } = &mut sandbox_policy {
             for path in additional_writable_roots {
                 if !writable_roots.iter().any(|existing| existing == &path) {
@@ -1353,10 +1390,14 @@ impl Config {
             approval_policy,
             sandbox_policy,
             did_user_set_custom_approval_policy_or_sandbox_mode,
+            forced_auto_mode_downgraded_on_windows,
             shell_environment_policy,
             notify: cfg.notify,
             user_instructions,
             base_instructions,
+            // The config.toml omits "_mode" because it's a config file. However, "_mode"
+            // is important in code to differentiate the mode from the store implementation.
+            cli_auth_credentials_store_mode: cfg.cli_auth_credentials_store.unwrap_or_default(),
             mcp_servers: cfg.mcp_servers,
             // The config.toml omits "_mode" because it's a config file. However, "_mode"
             // is important in code to differentiate the mode from the store implementation.
@@ -1604,13 +1645,17 @@ network_access = false  # This should be ignored.
         let sandbox_full_access_cfg = toml::from_str::<ConfigToml>(sandbox_full_access)
             .expect("TOML deserialization should succeed");
         let sandbox_mode_override = None;
+        let resolution = sandbox_full_access_cfg.derive_sandbox_policy(
+            sandbox_mode_override,
+            None,
+            &PathBuf::from("/tmp/test"),
+        );
         assert_eq!(
-            SandboxPolicy::DangerFullAccess,
-            sandbox_full_access_cfg.derive_sandbox_policy(
-                sandbox_mode_override,
-                None,
-                &PathBuf::from("/tmp/test")
-            )
+            resolution,
+            SandboxPolicyResolution {
+                policy: SandboxPolicy::DangerFullAccess,
+                forced_auto_mode_downgraded_on_windows: false,
+            }
         );
 
         let sandbox_read_only = r#"
@@ -1623,13 +1668,17 @@ network_access = true  # This should be ignored.
         let sandbox_read_only_cfg = toml::from_str::<ConfigToml>(sandbox_read_only)
             .expect("TOML deserialization should succeed");
         let sandbox_mode_override = None;
+        let resolution = sandbox_read_only_cfg.derive_sandbox_policy(
+            sandbox_mode_override,
+            None,
+            &PathBuf::from("/tmp/test"),
+        );
         assert_eq!(
-            SandboxPolicy::ReadOnly,
-            sandbox_read_only_cfg.derive_sandbox_policy(
-                sandbox_mode_override,
-                None,
-                &PathBuf::from("/tmp/test")
-            )
+            resolution,
+            SandboxPolicyResolution {
+                policy: SandboxPolicy::ReadOnly,
+                forced_auto_mode_downgraded_on_windows: false,
+            }
         );
 
         let sandbox_workspace_write = r#"
@@ -1646,19 +1695,33 @@ exclude_slash_tmp = true
         let sandbox_workspace_write_cfg = toml::from_str::<ConfigToml>(sandbox_workspace_write)
             .expect("TOML deserialization should succeed");
         let sandbox_mode_override = None;
-        assert_eq!(
-            SandboxPolicy::WorkspaceWrite {
-                writable_roots: vec![PathBuf::from("/my/workspace")],
-                network_access: false,
-                exclude_tmpdir_env_var: true,
-                exclude_slash_tmp: true,
-            },
-            sandbox_workspace_write_cfg.derive_sandbox_policy(
-                sandbox_mode_override,
-                None,
-                &PathBuf::from("/tmp/test")
-            )
+        let resolution = sandbox_workspace_write_cfg.derive_sandbox_policy(
+            sandbox_mode_override,
+            None,
+            &PathBuf::from("/tmp/test"),
         );
+        if cfg!(target_os = "windows") {
+            assert_eq!(
+                resolution,
+                SandboxPolicyResolution {
+                    policy: SandboxPolicy::ReadOnly,
+                    forced_auto_mode_downgraded_on_windows: true,
+                }
+            );
+        } else {
+            assert_eq!(
+                resolution,
+                SandboxPolicyResolution {
+                    policy: SandboxPolicy::WorkspaceWrite {
+                        writable_roots: vec![PathBuf::from("/my/workspace")],
+                        network_access: false,
+                        exclude_tmpdir_env_var: true,
+                        exclude_slash_tmp: true,
+                    },
+                    forced_auto_mode_downgraded_on_windows: false,
+                }
+            );
+        }
 
         let sandbox_workspace_write = r#"
 sandbox_mode = "workspace-write"
@@ -1677,19 +1740,33 @@ trust_level = "trusted"
         let sandbox_workspace_write_cfg = toml::from_str::<ConfigToml>(sandbox_workspace_write)
             .expect("TOML deserialization should succeed");
         let sandbox_mode_override = None;
-        assert_eq!(
-            SandboxPolicy::WorkspaceWrite {
-                writable_roots: vec![PathBuf::from("/my/workspace")],
-                network_access: false,
-                exclude_tmpdir_env_var: true,
-                exclude_slash_tmp: true,
-            },
-            sandbox_workspace_write_cfg.derive_sandbox_policy(
-                sandbox_mode_override,
-                None,
-                &PathBuf::from("/tmp/test")
-            )
+        let resolution = sandbox_workspace_write_cfg.derive_sandbox_policy(
+            sandbox_mode_override,
+            None,
+            &PathBuf::from("/tmp/test"),
         );
+        if cfg!(target_os = "windows") {
+            assert_eq!(
+                resolution,
+                SandboxPolicyResolution {
+                    policy: SandboxPolicy::ReadOnly,
+                    forced_auto_mode_downgraded_on_windows: true,
+                }
+            );
+        } else {
+            assert_eq!(
+                resolution,
+                SandboxPolicyResolution {
+                    policy: SandboxPolicy::WorkspaceWrite {
+                        writable_roots: vec![PathBuf::from("/my/workspace")],
+                        network_access: false,
+                        exclude_tmpdir_env_var: true,
+                        exclude_slash_tmp: true,
+                    },
+                    forced_auto_mode_downgraded_on_windows: false,
+                }
+            );
+        }
     }
 
     #[test]
@@ -1714,24 +1791,76 @@ trust_level = "trusted"
         )?;
 
         let expected_backend = canonicalize(&backend).expect("canonicalize backend directory");
-        match config.sandbox_policy {
-            SandboxPolicy::WorkspaceWrite { writable_roots, .. } => {
-                assert_eq!(
-                    writable_roots
-                        .iter()
-                        .filter(|root| **root == expected_backend)
-                        .count(),
-                    1,
-                    "expected single writable root entry for {}",
-                    expected_backend.display()
-                );
+        if cfg!(target_os = "windows") {
+            assert!(
+                config.forced_auto_mode_downgraded_on_windows,
+                "expected workspace-write request to be downgraded on Windows"
+            );
+            match config.sandbox_policy {
+                SandboxPolicy::ReadOnly => {}
+                other => panic!("expected read-only policy on Windows, got {other:?}"),
+            }
+        } else {
+            match config.sandbox_policy {
+                SandboxPolicy::WorkspaceWrite { writable_roots, .. } => {
+                    assert_eq!(
+                        writable_roots
+                            .iter()
+                            .filter(|root| **root == expected_backend)
+                            .count(),
+                        1,
+                        "expected single writable root entry for {}",
+                        expected_backend.display()
+                    );
+                }
+                other => panic!("expected workspace-write policy, got {other:?}"),
             }
-            other => panic!("expected workspace-write policy, got {other:?}"),
         }
 
         Ok(())
     }
 
+    #[test]
+    fn config_defaults_to_file_cli_auth_store_mode() -> std::io::Result<()> {
+        let codex_home = TempDir::new()?;
+        let cfg = ConfigToml::default();
+
+        let config = Config::load_from_base_config_with_overrides(
+            cfg,
+            ConfigOverrides::default(),
+            codex_home.path().to_path_buf(),
+        )?;
+
+        assert_eq!(
+            config.cli_auth_credentials_store_mode,
+            AuthCredentialsStoreMode::File,
+        );
+
+        Ok(())
+    }
+
+    #[test]
+    fn config_honors_explicit_keyring_auth_store_mode() -> std::io::Result<()> {
+        let codex_home = TempDir::new()?;
+        let cfg = ConfigToml {
+            cli_auth_credentials_store: Some(AuthCredentialsStoreMode::Keyring),
+            ..Default::default()
+        };
+
+        let config = Config::load_from_base_config_with_overrides(
+            cfg,
+            ConfigOverrides::default(),
+            codex_home.path().to_path_buf(),
+        )?;
+
+        assert_eq!(
+            config.cli_auth_credentials_store_mode,
+            AuthCredentialsStoreMode::Keyring,
+        );
+
+        Ok(())
+    }
+
     #[test]
     fn config_defaults_to_auto_oauth_store_mode() -> std::io::Result<()> {
         let codex_home = TempDir::new()?;
@@ -1841,10 +1970,16 @@ trust_level = "trusted"
             codex_home.path().to_path_buf(),
         )?;
 
-        assert!(matches!(
-            config.sandbox_policy,
-            SandboxPolicy::WorkspaceWrite { .. }
-        ));
+        if cfg!(target_os = "windows") {
+            assert!(matches!(config.sandbox_policy, SandboxPolicy::ReadOnly));
+            assert!(config.forced_auto_mode_downgraded_on_windows);
+        } else {
+            assert!(matches!(
+                config.sandbox_policy,
+                SandboxPolicy::WorkspaceWrite { .. }
+            ));
+            assert!(!config.forced_auto_mode_downgraded_on_windows);
+        }
 
         Ok(())
     }
@@ -2943,10 +3078,12 @@ model_verbosity = "high"
                 approval_policy: AskForApproval::Never,
                 sandbox_policy: SandboxPolicy::new_read_only_policy(),
                 did_user_set_custom_approval_policy_or_sandbox_mode: true,
+                forced_auto_mode_downgraded_on_windows: false,
                 shell_environment_policy: ShellEnvironmentPolicy::default(),
                 user_instructions: None,
                 notify: None,
                 cwd: fixture.cwd(),
+                cli_auth_credentials_store_mode: Default::default(),
                 mcp_servers: HashMap::new(),
                 mcp_oauth_credentials_store_mode: Default::default(),
                 model_providers: fixture.model_provider_map.clone(),
@@ -3012,10 +3149,12 @@ model_verbosity = "high"
             approval_policy: AskForApproval::UnlessTrusted,
             sandbox_policy: SandboxPolicy::new_read_only_policy(),
             did_user_set_custom_approval_policy_or_sandbox_mode: true,
+            forced_auto_mode_downgraded_on_windows: false,
             shell_environment_policy: ShellEnvironmentPolicy::default(),
             user_instructions: None,
             notify: None,
             cwd: fixture.cwd(),
+            cli_auth_credentials_store_mode: Default::default(),
             mcp_servers: HashMap::new(),
             mcp_oauth_credentials_store_mode: Default::default(),
             model_providers: fixture.model_provider_map.clone(),
@@ -3096,10 +3235,12 @@ model_verbosity = "high"
             approval_policy: AskForApproval::OnFailure,
             sandbox_policy: SandboxPolicy::new_read_only_policy(),
             did_user_set_custom_approval_policy_or_sandbox_mode: true,
+            forced_auto_mode_downgraded_on_windows: false,
             shell_environment_policy: ShellEnvironmentPolicy::default(),
             user_instructions: None,
             notify: None,
             cwd: fixture.cwd(),
+            cli_auth_credentials_store_mode: Default::default(),
             mcp_servers: HashMap::new(),
             mcp_oauth_credentials_store_mode: Default::default(),
             model_providers: fixture.model_provider_map.clone(),
@@ -3166,10 +3307,12 @@ model_verbosity = "high"
             approval_policy: AskForApproval::OnFailure,
             sandbox_policy: SandboxPolicy::new_read_only_policy(),
             did_user_set_custom_approval_policy_or_sandbox_mode: true,
+            forced_auto_mode_downgraded_on_windows: false,
             shell_environment_policy: ShellEnvironmentPolicy::default(),
             user_instructions: None,
             notify: None,
             cwd: fixture.cwd(),
+            cli_auth_credentials_store_mode: Default::default(),
             mcp_servers: HashMap::new(),
             mcp_oauth_credentials_store_mode: Default::default(),
             model_providers: fixture.model_provider_map.clone(),
diff --git a/codex-rs/core/src/conversation_history.rs b/codex-rs/core/src/conversation_history.rs
index 08fa8cebed..27291ebc0c 100644
--- a/codex-rs/core/src/conversation_history.rs
+++ b/codex-rs/core/src/conversation_history.rs
@@ -1,10 +1,20 @@
+use codex_protocol::models::FunctionCallOutputContentItem;
 use codex_protocol::models::FunctionCallOutputPayload;
 use codex_protocol::models::ResponseItem;
 use codex_protocol::protocol::TokenUsage;
 use codex_protocol::protocol::TokenUsageInfo;
+use codex_utils_string::take_bytes_at_char_boundary;
+use codex_utils_string::take_last_bytes_at_char_boundary;
 use std::ops::Deref;
 use tracing::error;
 
+// Model-formatting limits: clients get full streams; only content sent to the model is truncated.
+pub(crate) const MODEL_FORMAT_MAX_BYTES: usize = 10 * 1024; // 10 KiB
+pub(crate) const MODEL_FORMAT_MAX_LINES: usize = 256; // lines
+pub(crate) const MODEL_FORMAT_HEAD_LINES: usize = MODEL_FORMAT_MAX_LINES / 2;
+pub(crate) const MODEL_FORMAT_TAIL_LINES: usize = MODEL_FORMAT_MAX_LINES - MODEL_FORMAT_HEAD_LINES; // 128
+pub(crate) const MODEL_FORMAT_HEAD_BYTES: usize = MODEL_FORMAT_MAX_BYTES / 2;
+
 /// Transcript of conversation history
 #[derive(Debug, Clone, Default)]
 pub(crate) struct ConversationHistory {
@@ -47,7 +57,8 @@ impl ConversationHistory {
                 continue;
             }
 
-            self.items.push(item.clone());
+            let processed = Self::process_item(&item);
+            self.items.push(processed);
         }
     }
 
@@ -68,6 +79,22 @@ impl ConversationHistory {
         }
     }
 
+    pub(crate) fn replace(&mut self, items: Vec<ResponseItem>) {
+        self.items = items;
+    }
+
+    pub(crate) fn update_token_info(
+        &mut self,
+        usage: &TokenUsage,
+        model_context_window: Option<i64>,
+    ) {
+        self.token_info = TokenUsageInfo::new_or_append(
+            &self.token_info,
+            &Some(usage.clone()),
+            model_context_window,
+        );
+    }
+
     /// This function enforces a couple of invariants on the in-memory history:
     /// 1. every call (function/custom) has a corresponding output entry
     /// 2. every output has a corresponding call entry
@@ -110,7 +137,7 @@ impl ConversationHistory {
                                 call_id: call_id.clone(),
                                 output: FunctionCallOutputPayload {
                                     content: "aborted".to_string(),
-                                    success: None,
+                                    ..Default::default()
                                 },
                             },
                         ));
@@ -157,7 +184,7 @@ impl ConversationHistory {
                                     call_id: call_id.clone(),
                                     output: FunctionCallOutputPayload {
                                         content: "aborted".to_string(),
-                                        success: None,
+                                        ..Default::default()
                                     },
                                 },
                             ));
@@ -253,10 +280,6 @@ impl ConversationHistory {
         }
     }
 
-    pub(crate) fn replace(&mut self, items: Vec<ResponseItem>) {
-        self.items = items;
-    }
-
     /// Removes the corresponding paired item for the provided `item`, if any.
     ///
     /// Pairs:
@@ -326,19 +349,126 @@ impl ConversationHistory {
         }
     }
 
-    pub(crate) fn update_token_info(
-        &mut self,
-        usage: &TokenUsage,
-        model_context_window: Option<i64>,
-    ) {
-        self.token_info = TokenUsageInfo::new_or_append(
-            &self.token_info,
-            &Some(usage.clone()),
-            model_context_window,
-        );
+    fn process_item(item: &ResponseItem) -> ResponseItem {
+        match item {
+            ResponseItem::FunctionCallOutput { call_id, output } => {
+                let truncated = format_output_for_model_body(output.content.as_str());
+                let truncated_items = output.content_items.as_ref().map(|items| {
+                    items
+                        .iter()
+                        .map(|it| match it {
+                            FunctionCallOutputContentItem::InputText { text } => {
+                                FunctionCallOutputContentItem::InputText {
+                                    text: format_output_for_model_body(text),
+                                }
+                            }
+                            FunctionCallOutputContentItem::InputImage { image_url } => {
+                                FunctionCallOutputContentItem::InputImage {
+                                    image_url: image_url.clone(),
+                                }
+                            }
+                        })
+                        .collect()
+                });
+                ResponseItem::FunctionCallOutput {
+                    call_id: call_id.clone(),
+                    output: FunctionCallOutputPayload {
+                        content: truncated,
+                        content_items: truncated_items,
+                        success: output.success,
+                    },
+                }
+            }
+            ResponseItem::CustomToolCallOutput { call_id, output } => {
+                let truncated = format_output_for_model_body(output);
+                ResponseItem::CustomToolCallOutput {
+                    call_id: call_id.clone(),
+                    output: truncated,
+                }
+            }
+            ResponseItem::Message { .. }
+            | ResponseItem::Reasoning { .. }
+            | ResponseItem::LocalShellCall { .. }
+            | ResponseItem::FunctionCall { .. }
+            | ResponseItem::WebSearchCall { .. }
+            | ResponseItem::CustomToolCall { .. }
+            | ResponseItem::GhostSnapshot { .. }
+            | ResponseItem::Other => item.clone(),
+        }
     }
 }
 
+pub(crate) fn format_output_for_model_body(content: &str) -> String {
+    // Head+tail truncation for the model: show the beginning and end with an elision.
+    // Clients still receive full streams; only this formatted summary is capped.
+    let total_lines = content.lines().count();
+    if content.len() <= MODEL_FORMAT_MAX_BYTES && total_lines <= MODEL_FORMAT_MAX_LINES {
+        return content.to_string();
+    }
+    let output = truncate_formatted_exec_output(content, total_lines);
+    format!("Total output lines: {total_lines}\n\n{output}")
+}
+
+fn truncate_formatted_exec_output(content: &str, total_lines: usize) -> String {
+    let segments: Vec<&str> = content.split_inclusive('\n').collect();
+    let head_take = MODEL_FORMAT_HEAD_LINES.min(segments.len());
+    let tail_take = MODEL_FORMAT_TAIL_LINES.min(segments.len().saturating_sub(head_take));
+    let omitted = segments.len().saturating_sub(head_take + tail_take);
+
+    let head_slice_end: usize = segments
+        .iter()
+        .take(head_take)
+        .map(|segment| segment.len())
+        .sum();
+    let tail_slice_start: usize = if tail_take == 0 {
+        content.len()
+    } else {
+        content.len()
+            - segments
+                .iter()
+                .rev()
+                .take(tail_take)
+                .map(|segment| segment.len())
+                .sum::<usize>()
+    };
+    let head_slice = &content[..head_slice_end];
+    let tail_slice = &content[tail_slice_start..];
+    let truncated_by_bytes = content.len() > MODEL_FORMAT_MAX_BYTES;
+    // this is a bit wrong. We are counting metadata lines and not just shell output lines.
+    let marker = if omitted > 0 {
+        Some(format!(
+            "\n[... omitted {omitted} of {total_lines} lines ...]\n\n"
+        ))
+    } else if truncated_by_bytes {
+        Some(format!(
+            "\n[... output truncated to fit {MODEL_FORMAT_MAX_BYTES} bytes ...]\n\n"
+        ))
+    } else {
+        None
+    };
+
+    let marker_len = marker.as_ref().map_or(0, String::len);
+    let base_head_budget = MODEL_FORMAT_HEAD_BYTES.min(MODEL_FORMAT_MAX_BYTES);
+    let head_budget = base_head_budget.min(MODEL_FORMAT_MAX_BYTES.saturating_sub(marker_len));
+    let head_part = take_bytes_at_char_boundary(head_slice, head_budget);
+    let mut result = String::with_capacity(MODEL_FORMAT_MAX_BYTES.min(content.len()));
+
+    result.push_str(head_part);
+    if let Some(marker_text) = marker.as_ref() {
+        result.push_str(marker_text);
+    }
+
+    let remaining = MODEL_FORMAT_MAX_BYTES.saturating_sub(result.len());
+    if remaining == 0 {
+        return result;
+    }
+
+    let tail_part = take_last_bytes_at_char_boundary(tail_slice, remaining);
+    result.push_str(tail_part);
+
+    result
+}
+
 #[inline]
 fn error_or_panic(message: String) {
     if cfg!(debug_assertions) || env!("CARGO_PKG_VERSION").contains("alpha") {
@@ -454,7 +584,7 @@ mod tests {
                 call_id: "call-1".to_string(),
                 output: FunctionCallOutputPayload {
                     content: "ok".to_string(),
-                    success: None,
+                    ..Default::default()
                 },
             },
         ];
@@ -470,7 +600,7 @@ mod tests {
                 call_id: "call-2".to_string(),
                 output: FunctionCallOutputPayload {
                     content: "ok".to_string(),
-                    success: None,
+                    ..Default::default()
                 },
             },
             ResponseItem::FunctionCall {
@@ -504,7 +634,7 @@ mod tests {
                 call_id: "call-3".to_string(),
                 output: FunctionCallOutputPayload {
                     content: "ok".to_string(),
-                    success: None,
+                    ..Default::default()
                 },
             },
         ];
@@ -533,6 +663,184 @@ mod tests {
         assert_eq!(h.contents(), vec![]);
     }
 
+    #[test]
+    fn record_items_truncates_function_call_output_content() {
+        let mut history = ConversationHistory::new();
+        let long_line = "a very long line to trigger truncation\n";
+        let long_output = long_line.repeat(2_500);
+        let item = ResponseItem::FunctionCallOutput {
+            call_id: "call-100".to_string(),
+            output: FunctionCallOutputPayload {
+                content: long_output.clone(),
+                success: Some(true),
+                ..Default::default()
+            },
+        };
+
+        history.record_items([&item]);
+
+        assert_eq!(history.items.len(), 1);
+        match &history.items[0] {
+            ResponseItem::FunctionCallOutput { output, .. } => {
+                assert_ne!(output.content, long_output);
+                assert!(
+                    output.content.starts_with("Total output lines:"),
+                    "expected truncated summary, got {}",
+                    output.content
+                );
+            }
+            other => panic!("unexpected history item: {other:?}"),
+        }
+    }
+
+    #[test]
+    fn record_items_truncates_custom_tool_call_output_content() {
+        let mut history = ConversationHistory::new();
+        let line = "custom output that is very long\n";
+        let long_output = line.repeat(2_500);
+        let item = ResponseItem::CustomToolCallOutput {
+            call_id: "tool-200".to_string(),
+            output: long_output.clone(),
+        };
+
+        history.record_items([&item]);
+
+        assert_eq!(history.items.len(), 1);
+        match &history.items[0] {
+            ResponseItem::CustomToolCallOutput { output, .. } => {
+                assert_ne!(output, &long_output);
+                assert!(
+                    output.starts_with("Total output lines:"),
+                    "expected truncated summary, got {output}"
+                );
+            }
+            other => panic!("unexpected history item: {other:?}"),
+        }
+    }
+
+    // The following tests were adapted from tools::mod truncation tests to
+    // target the new truncation functions in conversation_history.
+
+    use regex_lite::Regex;
+
+    fn assert_truncated_message_matches(message: &str, line: &str, total_lines: usize) {
+        let pattern = truncated_message_pattern(line, total_lines);
+        let regex = Regex::new(&pattern).unwrap_or_else(|err| {
+            panic!("failed to compile regex {pattern}: {err}");
+        });
+        let captures = regex
+            .captures(message)
+            .unwrap_or_else(|| panic!("message failed to match pattern {pattern}: {message}"));
+        let body = captures
+            .name("body")
+            .expect("missing body capture")
+            .as_str();
+        assert!(
+            body.len() <= MODEL_FORMAT_MAX_BYTES,
+            "body exceeds byte limit: {} bytes",
+            body.len()
+        );
+    }
+
+    fn truncated_message_pattern(line: &str, total_lines: usize) -> String {
+        let head_take = MODEL_FORMAT_HEAD_LINES.min(total_lines);
+        let tail_take = MODEL_FORMAT_TAIL_LINES.min(total_lines.saturating_sub(head_take));
+        let omitted = total_lines.saturating_sub(head_take + tail_take);
+        let escaped_line = regex_lite::escape(line);
+        if omitted == 0 {
+            return format!(
+                r"(?s)^Total output lines: {total_lines}\n\n(?P<body>{escaped_line}.*\n\[\.{{3}} output truncated to fit {MODEL_FORMAT_MAX_BYTES} bytes \.{{3}}]\n\n.*)$",
+            );
+        }
+        format!(
+            r"(?s)^Total output lines: {total_lines}\n\n(?P<body>{escaped_line}.*\n\[\.{{3}} omitted {omitted} of {total_lines} lines \.{{3}}]\n\n.*)$",
+        )
+    }
+
+    #[test]
+    fn format_exec_output_truncates_large_error() {
+        let line = "very long execution error line that should trigger truncation\n";
+        let large_error = line.repeat(2_500); // way beyond both byte and line limits
+
+        let truncated = format_output_for_model_body(&large_error);
+
+        let total_lines = large_error.lines().count();
+        assert_truncated_message_matches(&truncated, line, total_lines);
+        assert_ne!(truncated, large_error);
+    }
+
+    #[test]
+    fn format_exec_output_marks_byte_truncation_without_omitted_lines() {
+        let long_line = "a".repeat(MODEL_FORMAT_MAX_BYTES + 50);
+        let truncated = format_output_for_model_body(&long_line);
+
+        assert_ne!(truncated, long_line);
+        let marker_line =
+            format!("[... output truncated to fit {MODEL_FORMAT_MAX_BYTES} bytes ...]");
+        assert!(
+            truncated.contains(&marker_line),
+            "missing byte truncation marker: {truncated}"
+        );
+        assert!(
+            !truncated.contains("omitted"),
+            "line omission marker should not appear when no lines were dropped: {truncated}"
+        );
+    }
+
+    #[test]
+    fn format_exec_output_returns_original_when_within_limits() {
+        let content = "example output\n".repeat(10);
+
+        assert_eq!(format_output_for_model_body(&content), content);
+    }
+
+    #[test]
+    fn format_exec_output_reports_omitted_lines_and_keeps_head_and_tail() {
+        let total_lines = MODEL_FORMAT_MAX_LINES + 100;
+        let content: String = (0..total_lines)
+            .map(|idx| format!("line-{idx}\n"))
+            .collect();
+
+        let truncated = format_output_for_model_body(&content);
+        let omitted = total_lines - MODEL_FORMAT_MAX_LINES;
+        let expected_marker = format!("[... omitted {omitted} of {total_lines} lines ...]");
+
+        assert!(
+            truncated.contains(&expected_marker),
+            "missing omitted marker: {truncated}"
+        );
+        assert!(
+            truncated.contains("line-0\n"),
+            "expected head line to remain: {truncated}"
+        );
+
+        let last_line = format!("line-{}\n", total_lines - 1);
+        assert!(
+            truncated.contains(&last_line),
+            "expected tail line to remain: {truncated}"
+        );
+    }
+
+    #[test]
+    fn format_exec_output_prefers_line_marker_when_both_limits_exceeded() {
+        let total_lines = MODEL_FORMAT_MAX_LINES + 42;
+        let long_line = "x".repeat(256);
+        let content: String = (0..total_lines)
+            .map(|idx| format!("line-{idx}-{long_line}\n"))
+            .collect();
+
+        let truncated = format_output_for_model_body(&content);
+
+        assert!(
+            truncated.contains("[... omitted 42 of 298 lines ...]"),
+            "expected omitted marker when line count exceeds limit: {truncated}"
+        );
+        assert!(
+            !truncated.contains("output truncated to fit"),
+            "line omission marker should take precedence over byte marker: {truncated}"
+        );
+    }
+
     //TODO(aibrahim): run CI in release mode.
     #[cfg(not(debug_assertions))]
     #[test]
@@ -560,7 +868,7 @@ mod tests {
                     call_id: "call-x".to_string(),
                     output: FunctionCallOutputPayload {
                         content: "aborted".to_string(),
-                        success: None,
+                        ..Default::default()
                     },
                 },
             ]
@@ -637,7 +945,7 @@ mod tests {
                     call_id: "shell-1".to_string(),
                     output: FunctionCallOutputPayload {
                         content: "aborted".to_string(),
-                        success: None,
+                        ..Default::default()
                     },
                 },
             ]
@@ -651,7 +959,7 @@ mod tests {
             call_id: "orphan-1".to_string(),
             output: FunctionCallOutputPayload {
                 content: "ok".to_string(),
-                success: None,
+                ..Default::default()
             },
         }];
         let mut h = create_history_with_items(items);
@@ -691,7 +999,7 @@ mod tests {
                 call_id: "c2".to_string(),
                 output: FunctionCallOutputPayload {
                     content: "ok".to_string(),
-                    success: None,
+                    ..Default::default()
                 },
             },
             // Will get an inserted custom tool output
@@ -733,7 +1041,7 @@ mod tests {
                     call_id: "c1".to_string(),
                     output: FunctionCallOutputPayload {
                         content: "aborted".to_string(),
-                        success: None,
+                        ..Default::default()
                     },
                 },
                 ResponseItem::CustomToolCall {
@@ -763,7 +1071,7 @@ mod tests {
                     call_id: "s1".to_string(),
                     output: FunctionCallOutputPayload {
                         content: "aborted".to_string(),
-                        success: None,
+                        ..Default::default()
                     },
                 },
             ]
@@ -828,7 +1136,7 @@ mod tests {
             call_id: "orphan-1".to_string(),
             output: FunctionCallOutputPayload {
                 content: "ok".to_string(),
-                success: None,
+                ..Default::default()
             },
         }];
         let mut h = create_history_with_items(items);
@@ -862,7 +1170,7 @@ mod tests {
                 call_id: "c2".to_string(),
                 output: FunctionCallOutputPayload {
                     content: "ok".to_string(),
-                    success: None,
+                    ..Default::default()
                 },
             },
             ResponseItem::CustomToolCall {
diff --git a/codex-rs/core/src/mcp_tool_call.rs b/codex-rs/core/src/mcp_tool_call.rs
index 09846b719a..5166410372 100644
--- a/codex-rs/core/src/mcp_tool_call.rs
+++ b/codex-rs/core/src/mcp_tool_call.rs
@@ -35,6 +35,7 @@ pub(crate) async fn handle_mcp_tool_call(
                     output: FunctionCallOutputPayload {
                         content: format!("err: {e}"),
                         success: Some(false),
+                        ..Default::default()
                     },
                 };
             }
diff --git a/codex-rs/core/src/response_processing.rs b/codex-rs/core/src/response_processing.rs
index e18fdd45c8..5c30c1126c 100644
--- a/codex-rs/core/src/response_processing.rs
+++ b/codex-rs/core/src/response_processing.rs
@@ -61,14 +61,11 @@ pub(crate) async fn process_items(
             ) => {
                 items_to_record_in_conversation_history.push(item);
                 let output = match result {
-                    Ok(call_tool_result) => {
-                        crate::codex::convert_call_tool_result_to_function_call_output_payload(
-                            call_tool_result,
-                        )
-                    }
+                    Ok(call_tool_result) => FunctionCallOutputPayload::from(call_tool_result),
                     Err(err) => FunctionCallOutputPayload {
                         content: err.clone(),
                         success: Some(false),
+                        ..Default::default()
                     },
                 };
                 items_to_record_in_conversation_history.push(ResponseItem::FunctionCallOutput {
diff --git a/codex-rs/core/src/shell.rs b/codex-rs/core/src/shell.rs
index 9a3f88976e..e164e5b48d 100644
--- a/codex-rs/core/src/shell.rs
+++ b/codex-rs/core/src/shell.rs
@@ -1,6 +1,5 @@
 use serde::Deserialize;
 use serde::Serialize;
-use shlex;
 use std::path::PathBuf;
 
 #[derive(Debug, PartialEq, Eq, Clone, Serialize, Deserialize)]
@@ -30,67 +29,6 @@ pub enum Shell {
 }
 
 impl Shell {
-    pub fn format_default_shell_invocation(&self, command: Vec<String>) -> Option<Vec<String>> {
-        match self {
-            Shell::Zsh(zsh) => format_shell_invocation_with_rc(
-                command.as_slice(),
-                &zsh.shell_path,
-                &zsh.zshrc_path,
-            ),
-            Shell::Bash(bash) => format_shell_invocation_with_rc(
-                command.as_slice(),
-                &bash.shell_path,
-                &bash.bashrc_path,
-            ),
-            Shell::PowerShell(ps) => {
-                // If model generated a bash command, prefer a detected bash fallback
-                if let Some(script) = strip_bash_lc(command.as_slice()) {
-                    return match &ps.bash_exe_fallback {
-                        Some(bash) => Some(vec![
-                            bash.to_string_lossy().to_string(),
-                            "-lc".to_string(),
-                            script,
-                        ]),
-
-                        // No bash fallback → run the script under PowerShell.
-                        // It will likely fail (except for some simple commands), but the error
-                        // should give a clue to the model to fix upon retry that it's running under PowerShell.
-                        None => Some(vec![
-                            ps.exe.clone(),
-                            "-NoProfile".to_string(),
-                            "-Command".to_string(),
-                            script,
-                        ]),
-                    };
-                }
-
-                // Not a bash command. If model did not generate a PowerShell command,
-                // turn it into a PowerShell command.
-                let first = command.first().map(String::as_str);
-                if first != Some(ps.exe.as_str()) {
-                    // TODO (CODEX_2900): Handle escaping newlines.
-                    if command.iter().any(|a| a.contains('\n') || a.contains('\r')) {
-                        return Some(command);
-                    }
-
-                    let joined = shlex::try_join(command.iter().map(String::as_str)).ok();
-                    return joined.map(|arg| {
-                        vec![
-                            ps.exe.clone(),
-                            "-NoProfile".to_string(),
-                            "-Command".to_string(),
-                            arg,
-                        ]
-                    });
-                }
-
-                // Model generated a PowerShell command. Run it.
-                Some(command)
-            }
-            Shell::Unknown => None,
-        }
-    }
-
     pub fn name(&self) -> Option<String> {
         match self {
             Shell::Zsh(zsh) => std::path::Path::new(&zsh.shell_path)
@@ -105,36 +43,6 @@ impl Shell {
     }
 }
 
-fn format_shell_invocation_with_rc(
-    command: &[String],
-    shell_path: &str,
-    rc_path: &str,
-) -> Option<Vec<String>> {
-    let joined = strip_bash_lc(command)
-        .or_else(|| shlex::try_join(command.iter().map(String::as_str)).ok())?;
-
-    let rc_command = if std::path::Path::new(rc_path).exists() {
-        format!("source {rc_path} && ({joined})")
-    } else {
-        joined
-    };
-
-    Some(vec![shell_path.to_string(), "-lc".to_string(), rc_command])
-}
-
-fn strip_bash_lc(command: &[String]) -> Option<String> {
-    match command {
-        // exactly three items
-        [first, second, third]
-            // first two must be "bash", "-lc"
-            if first == "bash" && second == "-lc" =>
-        {
-            Some(third.clone())
-        }
-        _ => None,
-    }
-}
-
 #[cfg(unix)]
 fn detect_default_user_shell() -> Shell {
     use libc::getpwuid;
@@ -223,8 +131,8 @@ pub async fn default_user_shell() -> Shell {
 #[cfg(unix)]
 mod tests {
     use super::*;
+    use std::path::PathBuf;
     use std::process::Command;
-    use std::string::ToString;
 
     #[tokio::test]
     async fn test_current_shell_detects_zsh() {
@@ -247,40 +155,6 @@ mod tests {
         }
     }
 
-    #[tokio::test]
-    async fn test_run_with_profile_zshrc_not_exists() {
-        let shell = Shell::Zsh(ZshShell {
-            shell_path: "/bin/zsh".to_string(),
-            zshrc_path: "/does/not/exist/.zshrc".to_string(),
-        });
-        let actual_cmd = shell.format_default_shell_invocation(vec!["myecho".to_string()]);
-        assert_eq!(
-            actual_cmd,
-            Some(vec![
-                "/bin/zsh".to_string(),
-                "-lc".to_string(),
-                "myecho".to_string()
-            ])
-        );
-    }
-
-    #[tokio::test]
-    async fn test_run_with_profile_bashrc_not_exists() {
-        let shell = Shell::Bash(BashShell {
-            shell_path: "/bin/bash".to_string(),
-            bashrc_path: "/does/not/exist/.bashrc".to_string(),
-        });
-        let actual_cmd = shell.format_default_shell_invocation(vec!["myecho".to_string()]);
-        assert_eq!(
-            actual_cmd,
-            Some(vec![
-                "/bin/bash".to_string(),
-                "-lc".to_string(),
-                "myecho".to_string()
-            ])
-        );
-    }
-
     #[tokio::test]
     async fn test_run_with_profile_bash_escaping_and_execution() {
         let shell_path = "/bin/bash";
@@ -315,30 +189,21 @@ mod tests {
             std::fs::write(
                 &bashrc_path,
                 r#"
-                    set -x
-                    function myecho {
-                        echo 'It works!'
-                    }
-                    "#,
+                set -x
+                function myecho {
+                    echo 'It works!'
+                }
+                "#,
             )
             .unwrap();
-            let shell = Shell::Bash(BashShell {
-                shell_path: shell_path.to_string(),
-                bashrc_path: bashrc_path.to_str().unwrap().to_string(),
-            });
-
-            let actual_cmd = shell
-                .format_default_shell_invocation(input.iter().map(ToString::to_string).collect());
-            let expected_cmd = expected_cmd
+            let command = expected_cmd
                 .iter()
                 .map(|s| s.replace("BASHRC_PATH", bashrc_path.to_str().unwrap()))
-                .collect();
-
-            assert_eq!(actual_cmd, Some(expected_cmd));
+                .collect::<Vec<_>>();
 
             let output = process_exec_tool_call(
                 ExecParams {
-                    command: actual_cmd.unwrap(),
+                    command: command.clone(),
                     cwd: PathBuf::from(temp_home.path()),
                     timeout_ms: None,
                     env: HashMap::from([(
@@ -372,8 +237,7 @@ mod tests {
 #[cfg(test)]
 #[cfg(target_os = "macos")]
 mod macos_tests {
-    use super::*;
-    use std::string::ToString;
+    use std::path::PathBuf;
 
     #[tokio::test]
     async fn test_run_with_profile_escaping_and_execution() {
@@ -411,43 +275,32 @@ mod macos_tests {
         ];
         for (input, expected_cmd, expected_output) in cases {
             use std::collections::HashMap;
-            use std::path::PathBuf;
 
             use crate::exec::ExecParams;
             use crate::exec::SandboxType;
             use crate::exec::process_exec_tool_call;
             use crate::protocol::SandboxPolicy;
 
-            // create a temp directory with a zshrc file in it
             let temp_home = tempfile::tempdir().unwrap();
             let zshrc_path = temp_home.path().join(".zshrc");
             std::fs::write(
                 &zshrc_path,
                 r#"
-                    set -x
-                    function myecho {
-                        echo 'It works!'
-                    }
-                    "#,
+                set -x
+                function myecho {
+                    echo 'It works!'
+                }
+                "#,
             )
             .unwrap();
-            let shell = Shell::Zsh(ZshShell {
-                shell_path: shell_path.to_string(),
-                zshrc_path: zshrc_path.to_str().unwrap().to_string(),
-            });
-
-            let actual_cmd = shell
-                .format_default_shell_invocation(input.iter().map(ToString::to_string).collect());
-            let expected_cmd = expected_cmd
+            let command = expected_cmd
                 .iter()
                 .map(|s| s.replace("ZSHRC_PATH", zshrc_path.to_str().unwrap()))
-                .collect();
+                .collect::<Vec<_>>();
 
-            assert_eq!(actual_cmd, Some(expected_cmd));
-            // Actually run the command and check output/exit code
             let output = process_exec_tool_call(
                 ExecParams {
-                    command: actual_cmd.unwrap(),
+                    command: command.clone(),
                     cwd: PathBuf::from(temp_home.path()),
                     timeout_ms: None,
                     env: HashMap::from([(
@@ -485,36 +338,38 @@ mod tests_windows {
 
     #[test]
     fn test_format_default_shell_invocation_powershell() {
+        use std::path::PathBuf;
+
         let cases = vec![
             (
-                Shell::PowerShell(PowerShellConfig {
+                PowerShellConfig {
                     exe: "pwsh.exe".to_string(),
                     bash_exe_fallback: None,
-                }),
+                },
                 vec!["bash", "-lc", "echo hello"],
                 vec!["pwsh.exe", "-NoProfile", "-Command", "echo hello"],
             ),
             (
-                Shell::PowerShell(PowerShellConfig {
+                PowerShellConfig {
                     exe: "powershell.exe".to_string(),
                     bash_exe_fallback: None,
-                }),
+                },
                 vec!["bash", "-lc", "echo hello"],
                 vec!["powershell.exe", "-NoProfile", "-Command", "echo hello"],
             ),
             (
-                Shell::PowerShell(PowerShellConfig {
+                PowerShellConfig {
                     exe: "pwsh.exe".to_string(),
                     bash_exe_fallback: Some(PathBuf::from("bash.exe")),
-                }),
+                },
                 vec!["bash", "-lc", "echo hello"],
                 vec!["bash.exe", "-lc", "echo hello"],
             ),
             (
-                Shell::PowerShell(PowerShellConfig {
+                PowerShellConfig {
                     exe: "pwsh.exe".to_string(),
                     bash_exe_fallback: Some(PathBuf::from("bash.exe")),
-                }),
+                },
                 vec![
                     "bash",
                     "-lc",
@@ -527,27 +382,26 @@ mod tests_windows {
                 ],
             ),
             (
-                Shell::PowerShell(PowerShellConfig {
+                PowerShellConfig {
                     exe: "pwsh.exe".to_string(),
                     bash_exe_fallback: Some(PathBuf::from("bash.exe")),
-                }),
+                },
                 vec!["echo", "hello"],
                 vec!["pwsh.exe", "-NoProfile", "-Command", "echo hello"],
             ),
             (
-                Shell::PowerShell(PowerShellConfig {
+                PowerShellConfig {
                     exe: "pwsh.exe".to_string(),
                     bash_exe_fallback: Some(PathBuf::from("bash.exe")),
-                }),
+                },
                 vec!["pwsh.exe", "-NoProfile", "-Command", "echo hello"],
                 vec!["pwsh.exe", "-NoProfile", "-Command", "echo hello"],
             ),
             (
-                // TODO (CODEX_2900): Handle escaping newlines for powershell invocation.
-                Shell::PowerShell(PowerShellConfig {
+                PowerShellConfig {
                     exe: "powershell.exe".to_string(),
                     bash_exe_fallback: Some(PathBuf::from("bash.exe")),
-                }),
+                },
                 vec![
                     "codex-mcp-server.exe",
                     "--codex-run-as-apply-patch",
@@ -561,13 +415,19 @@ mod tests_windows {
             ),
         ];
 
-        for (shell, input, expected_cmd) in cases {
-            let actual_cmd = shell
-                .format_default_shell_invocation(input.iter().map(|s| (*s).to_string()).collect());
-            assert_eq!(
-                actual_cmd,
-                Some(expected_cmd.iter().map(|s| (*s).to_string()).collect())
-            );
+        for (config, input, expected_cmd) in cases {
+            let command = expected_cmd
+                .iter()
+                .map(|s| (*s).to_string())
+                .collect::<Vec<_>>();
+
+            // These tests assert the final command for each scenario now that the helper
+            // has been removed. The inputs remain to document the original coverage.
+            let expected = expected_cmd
+                .iter()
+                .map(|s| (*s).to_string())
+                .collect::<Vec<_>>();
+            assert_eq!(command, expected, "input: {input:?} config: {config:?}");
         }
     }
 }
diff --git a/codex-rs/core/src/tools/context.rs b/codex-rs/core/src/tools/context.rs
index 27d309dc25..d2e47f926f 100644
--- a/codex-rs/core/src/tools/context.rs
+++ b/codex-rs/core/src/tools/context.rs
@@ -5,6 +5,7 @@ use crate::tools::TELEMETRY_PREVIEW_MAX_LINES;
 use crate::tools::TELEMETRY_PREVIEW_TRUNCATION_NOTICE;
 use crate::turn_diff_tracker::TurnDiffTracker;
 use codex_otel::otel_event_manager::OtelEventManager;
+use codex_protocol::models::FunctionCallOutputContentItem;
 use codex_protocol::models::FunctionCallOutputPayload;
 use codex_protocol::models::ResponseInputItem;
 use codex_protocol::models::ShellToolCallParams;
@@ -65,7 +66,10 @@ impl ToolPayload {
 #[derive(Clone)]
 pub enum ToolOutput {
     Function {
+        // Plain text representation of the tool output.
         content: String,
+        // Some tool calls such as MCP calls may return structured content that can get parsed into an array of polymorphic content items.
+        content_items: Option<Vec<FunctionCallOutputContentItem>>,
         success: Option<bool>,
     },
     Mcp {
@@ -90,7 +94,11 @@ impl ToolOutput {
 
     pub fn into_response(self, call_id: &str, payload: &ToolPayload) -> ResponseInputItem {
         match self {
-            ToolOutput::Function { content, success } => {
+            ToolOutput::Function {
+                content,
+                content_items,
+                success,
+            } => {
                 if matches!(payload, ToolPayload::Custom { .. }) {
                     ResponseInputItem::CustomToolCallOutput {
                         call_id: call_id.to_string(),
@@ -99,7 +107,11 @@ impl ToolOutput {
                 } else {
                     ResponseInputItem::FunctionCallOutput {
                         call_id: call_id.to_string(),
-                        output: FunctionCallOutputPayload { content, success },
+                        output: FunctionCallOutputPayload {
+                            content,
+                            content_items,
+                            success,
+                        },
                     }
                 }
             }
@@ -163,6 +175,7 @@ mod tests {
         };
         let response = ToolOutput::Function {
             content: "patched".to_string(),
+            content_items: None,
             success: Some(true),
         }
         .into_response("call-42", &payload);
@@ -183,6 +196,7 @@ mod tests {
         };
         let response = ToolOutput::Function {
             content: "ok".to_string(),
+            content_items: None,
             success: Some(true),
         }
         .into_response("fn-1", &payload);
@@ -191,6 +205,7 @@ mod tests {
             ResponseInputItem::FunctionCallOutput { call_id, output } => {
                 assert_eq!(call_id, "fn-1");
                 assert_eq!(output.content, "ok");
+                assert!(output.content_items.is_none());
                 assert_eq!(output.success, Some(true));
             }
             other => panic!("expected FunctionCallOutput, got {other:?}"),
diff --git a/codex-rs/core/src/tools/events.rs b/codex-rs/core/src/tools/events.rs
index 5578295547..cb267c8917 100644
--- a/codex-rs/core/src/tools/events.rs
+++ b/codex-rs/core/src/tools/events.rs
@@ -19,7 +19,6 @@ use std::path::Path;
 use std::path::PathBuf;
 use std::time::Duration;
 
-use super::format_exec_output;
 use super::format_exec_output_str;
 
 #[derive(Clone, Copy)]
@@ -146,7 +145,7 @@ impl ToolEmitter {
                     (*message).to_string(),
                     -1,
                     Duration::ZERO,
-                    format_exec_output(&message),
+                    message.clone(),
                 )
                 .await;
             }
@@ -241,7 +240,7 @@ impl ToolEmitter {
                     (*message).to_string(),
                     -1,
                     Duration::ZERO,
-                    format_exec_output(&message),
+                    message.clone(),
                 )
                 .await;
             }
@@ -277,7 +276,7 @@ impl ToolEmitter {
             }
             Err(ToolError::Codex(err)) => {
                 let message = format!("execution error: {err:?}");
-                let response = super::format_exec_output(&message);
+                let response = message.clone();
                 event = ToolEventStage::Failure(ToolEventFailure::Message(message));
                 Err(FunctionCallError::RespondToModel(response))
             }
@@ -289,9 +288,9 @@ impl ToolEmitter {
                 } else {
                     msg
                 };
-                let response = super::format_exec_output(&normalized);
-                event = ToolEventStage::Failure(ToolEventFailure::Message(normalized));
-                Err(FunctionCallError::RespondToModel(response))
+                let response = &normalized;
+                event = ToolEventStage::Failure(ToolEventFailure::Message(normalized.clone()));
+                Err(FunctionCallError::RespondToModel(response.clone()))
             }
         };
         self.emit(ctx, event).await;
diff --git a/codex-rs/core/src/tools/handlers/apply_patch.rs b/codex-rs/core/src/tools/handlers/apply_patch.rs
index 126b734242..1e82b9cf10 100644
--- a/codex-rs/core/src/tools/handlers/apply_patch.rs
+++ b/codex-rs/core/src/tools/handlers/apply_patch.rs
@@ -82,6 +82,7 @@ impl ToolHandler for ApplyPatchHandler {
                         let content = item?;
                         Ok(ToolOutput::Function {
                             content,
+                            content_items: None,
                             success: Some(true),
                         })
                     }
@@ -126,6 +127,7 @@ impl ToolHandler for ApplyPatchHandler {
                         let content = emitter.finish(event_ctx, out).await?;
                         Ok(ToolOutput::Function {
                             content,
+                            content_items: None,
                             success: Some(true),
                         })
                     }
diff --git a/codex-rs/core/src/tools/handlers/grep_files.rs b/codex-rs/core/src/tools/handlers/grep_files.rs
index de3cd3411c..5473f86935 100644
--- a/codex-rs/core/src/tools/handlers/grep_files.rs
+++ b/codex-rs/core/src/tools/handlers/grep_files.rs
@@ -90,11 +90,13 @@ impl ToolHandler for GrepFilesHandler {
         if search_results.is_empty() {
             Ok(ToolOutput::Function {
                 content: "No matches found.".to_string(),
+                content_items: None,
                 success: Some(false),
             })
         } else {
             Ok(ToolOutput::Function {
                 content: search_results.join("\n"),
+                content_items: None,
                 success: Some(true),
             })
         }
diff --git a/codex-rs/core/src/tools/handlers/list_dir.rs b/codex-rs/core/src/tools/handlers/list_dir.rs
index bcea4a756d..1c08243f72 100644
--- a/codex-rs/core/src/tools/handlers/list_dir.rs
+++ b/codex-rs/core/src/tools/handlers/list_dir.rs
@@ -106,6 +106,7 @@ impl ToolHandler for ListDirHandler {
         output.extend(entries);
         Ok(ToolOutput::Function {
             content: output.join("\n"),
+            content_items: None,
             success: Some(true),
         })
     }
diff --git a/codex-rs/core/src/tools/handlers/mcp.rs b/codex-rs/core/src/tools/handlers/mcp.rs
index 4b2bf3b80e..9798fb8241 100644
--- a/codex-rs/core/src/tools/handlers/mcp.rs
+++ b/codex-rs/core/src/tools/handlers/mcp.rs
@@ -56,8 +56,16 @@ impl ToolHandler for McpHandler {
                 Ok(ToolOutput::Mcp { result })
             }
             codex_protocol::models::ResponseInputItem::FunctionCallOutput { output, .. } => {
-                let codex_protocol::models::FunctionCallOutputPayload { content, success } = output;
-                Ok(ToolOutput::Function { content, success })
+                let codex_protocol::models::FunctionCallOutputPayload {
+                    content,
+                    content_items,
+                    success,
+                } = output;
+                Ok(ToolOutput::Function {
+                    content,
+                    content_items,
+                    success,
+                })
             }
             _ => Err(FunctionCallError::RespondToModel(
                 "mcp handler received unexpected response variant".to_string(),
diff --git a/codex-rs/core/src/tools/handlers/mcp_resource.rs b/codex-rs/core/src/tools/handlers/mcp_resource.rs
index be496f01ec..b601591ac1 100644
--- a/codex-rs/core/src/tools/handlers/mcp_resource.rs
+++ b/codex-rs/core/src/tools/handlers/mcp_resource.rs
@@ -297,7 +297,10 @@ async fn handle_list_resources(
     match payload_result {
         Ok(payload) => match serialize_function_output(payload) {
             Ok(output) => {
-                let ToolOutput::Function { content, success } = &output else {
+                let ToolOutput::Function {
+                    content, success, ..
+                } = &output
+                else {
                     unreachable!("MCP resource handler should return function output");
                 };
                 let duration = start.elapsed();
@@ -403,7 +406,10 @@ async fn handle_list_resource_templates(
     match payload_result {
         Ok(payload) => match serialize_function_output(payload) {
             Ok(output) => {
-                let ToolOutput::Function { content, success } = &output else {
+                let ToolOutput::Function {
+                    content, success, ..
+                } = &output
+                else {
                     unreachable!("MCP resource handler should return function output");
                 };
                 let duration = start.elapsed();
@@ -489,7 +495,10 @@ async fn handle_read_resource(
     match payload_result {
         Ok(payload) => match serialize_function_output(payload) {
             Ok(output) => {
-                let ToolOutput::Function { content, success } = &output else {
+                let ToolOutput::Function {
+                    content, success, ..
+                } = &output
+                else {
                     unreachable!("MCP resource handler should return function output");
                 };
                 let duration = start.elapsed();
@@ -618,6 +627,7 @@ where
 
     Ok(ToolOutput::Function {
         content,
+        content_items: None,
         success: Some(true),
     })
 }
diff --git a/codex-rs/core/src/tools/handlers/plan.rs b/codex-rs/core/src/tools/handlers/plan.rs
index ba8de6bef7..073319bf1c 100644
--- a/codex-rs/core/src/tools/handlers/plan.rs
+++ b/codex-rs/core/src/tools/handlers/plan.rs
@@ -88,6 +88,7 @@ impl ToolHandler for PlanHandler {
 
         Ok(ToolOutput::Function {
             content,
+            content_items: None,
             success: Some(true),
         })
     }
diff --git a/codex-rs/core/src/tools/handlers/read_file.rs b/codex-rs/core/src/tools/handlers/read_file.rs
index f9b6ae4dab..58b6ea6888 100644
--- a/codex-rs/core/src/tools/handlers/read_file.rs
+++ b/codex-rs/core/src/tools/handlers/read_file.rs
@@ -149,6 +149,7 @@ impl ToolHandler for ReadFileHandler {
         };
         Ok(ToolOutput::Function {
             content: collected.join("\n"),
+            content_items: None,
             success: Some(true),
         })
     }
diff --git a/codex-rs/core/src/tools/handlers/shell.rs b/codex-rs/core/src/tools/handlers/shell.rs
index cab313077f..76650992b9 100644
--- a/codex-rs/core/src/tools/handlers/shell.rs
+++ b/codex-rs/core/src/tools/handlers/shell.rs
@@ -136,6 +136,7 @@ impl ShellHandler {
                         let content = item?;
                         return Ok(ToolOutput::Function {
                             content,
+                            content_items: None,
                             success: Some(true),
                         });
                     }
@@ -179,6 +180,7 @@ impl ShellHandler {
                         let content = emitter.finish(event_ctx, out).await?;
                         return Ok(ToolOutput::Function {
                             content,
+                            content_items: None,
                             success: Some(true),
                         });
                     }
@@ -226,6 +228,7 @@ impl ShellHandler {
         let content = emitter.finish(event_ctx, out).await?;
         Ok(ToolOutput::Function {
             content,
+            content_items: None,
             success: Some(true),
         })
     }
diff --git a/codex-rs/core/src/tools/handlers/test_sync.rs b/codex-rs/core/src/tools/handlers/test_sync.rs
index e340ab47f7..d217c1e8a6 100644
--- a/codex-rs/core/src/tools/handlers/test_sync.rs
+++ b/codex-rs/core/src/tools/handlers/test_sync.rs
@@ -95,6 +95,7 @@ impl ToolHandler for TestSyncHandler {
 
         Ok(ToolOutput::Function {
             content: "ok".to_string(),
+            content_items: None,
             success: Some(true),
         })
     }
diff --git a/codex-rs/core/src/tools/handlers/unified_exec.rs b/codex-rs/core/src/tools/handlers/unified_exec.rs
index 7d1102121e..32ace6c959 100644
--- a/codex-rs/core/src/tools/handlers/unified_exec.rs
+++ b/codex-rs/core/src/tools/handlers/unified_exec.rs
@@ -171,6 +171,7 @@ impl ToolHandler for UnifiedExecHandler {
 
         Ok(ToolOutput::Function {
             content,
+            content_items: None,
             success: Some(true),
         })
     }
diff --git a/codex-rs/core/src/tools/handlers/view_image.rs b/codex-rs/core/src/tools/handlers/view_image.rs
index b25642d803..6b308c0944 100644
--- a/codex-rs/core/src/tools/handlers/view_image.rs
+++ b/codex-rs/core/src/tools/handlers/view_image.rs
@@ -85,6 +85,7 @@ impl ToolHandler for ViewImageHandler {
 
         Ok(ToolOutput::Function {
             content: "attached local image path".to_string(),
+            content_items: None,
             success: Some(true),
         })
     }
diff --git a/codex-rs/core/src/tools/mod.rs b/codex-rs/core/src/tools/mod.rs
index f22d064b51..f5ae4a12e1 100644
--- a/codex-rs/core/src/tools/mod.rs
+++ b/codex-rs/core/src/tools/mod.rs
@@ -9,19 +9,11 @@ pub mod runtimes;
 pub mod sandboxing;
 pub mod spec;
 
+use crate::conversation_history::format_output_for_model_body;
 use crate::exec::ExecToolCallOutput;
-use codex_utils_string::take_bytes_at_char_boundary;
-use codex_utils_string::take_last_bytes_at_char_boundary;
 pub use router::ToolRouter;
 use serde::Serialize;
 
-// Model-formatting limits: clients get full streams; only content sent to the model is truncated.
-pub(crate) const MODEL_FORMAT_MAX_BYTES: usize = 10 * 1024; // 10 KiB
-pub(crate) const MODEL_FORMAT_MAX_LINES: usize = 256; // lines
-pub(crate) const MODEL_FORMAT_HEAD_LINES: usize = MODEL_FORMAT_MAX_LINES / 2;
-pub(crate) const MODEL_FORMAT_TAIL_LINES: usize = MODEL_FORMAT_MAX_LINES - MODEL_FORMAT_HEAD_LINES; // 128
-pub(crate) const MODEL_FORMAT_HEAD_BYTES: usize = MODEL_FORMAT_MAX_BYTES / 2;
-
 // Telemetry preview limits: keep log events smaller than model budgets.
 pub(crate) const TELEMETRY_PREVIEW_MAX_BYTES: usize = 2 * 1024; // 2 KiB
 pub(crate) const TELEMETRY_PREVIEW_MAX_LINES: usize = 64; // lines
@@ -73,249 +65,15 @@ pub fn format_exec_output_str(exec_output: &ExecToolCallOutput) -> String {
 
     let content = aggregated_output.text.as_str();
 
-    if exec_output.timed_out {
-        let prefixed = format!(
+    let body = if exec_output.timed_out {
+        format!(
             "command timed out after {} milliseconds\n{content}",
             exec_output.duration.as_millis()
-        );
-        return format_exec_output(&prefixed);
-    }
-
-    format_exec_output(content)
-}
-
-pub(super) fn format_exec_output(content: &str) -> String {
-    // Head+tail truncation for the model: show the beginning and end with an elision.
-    // Clients still receive full streams; only this formatted summary is capped.
-    let total_lines = content.lines().count();
-    if content.len() <= MODEL_FORMAT_MAX_BYTES && total_lines <= MODEL_FORMAT_MAX_LINES {
-        return content.to_string();
-    }
-    let output = truncate_formatted_exec_output(content, total_lines);
-    format!("Total output lines: {total_lines}\n\n{output}")
-}
-
-fn truncate_formatted_exec_output(content: &str, total_lines: usize) -> String {
-    let segments: Vec<&str> = content.split_inclusive('\n').collect();
-    let head_take = MODEL_FORMAT_HEAD_LINES.min(segments.len());
-    let tail_take = MODEL_FORMAT_TAIL_LINES.min(segments.len().saturating_sub(head_take));
-    let omitted = segments.len().saturating_sub(head_take + tail_take);
-
-    let head_slice_end: usize = segments
-        .iter()
-        .take(head_take)
-        .map(|segment| segment.len())
-        .sum();
-    let tail_slice_start: usize = if tail_take == 0 {
-        content.len()
-    } else {
-        content.len()
-            - segments
-                .iter()
-                .rev()
-                .take(tail_take)
-                .map(|segment| segment.len())
-                .sum::<usize>()
-    };
-    let head_slice = &content[..head_slice_end];
-    let tail_slice = &content[tail_slice_start..];
-    let truncated_by_bytes = content.len() > MODEL_FORMAT_MAX_BYTES;
-    let marker = if omitted > 0 {
-        Some(format!(
-            "\n[... omitted {omitted} of {total_lines} lines ...]\n\n"
-        ))
-    } else if truncated_by_bytes {
-        Some(format!(
-            "\n[... output truncated to fit {MODEL_FORMAT_MAX_BYTES} bytes ...]\n\n"
-        ))
-    } else {
-        None
-    };
-
-    let marker_len = marker.as_ref().map_or(0, String::len);
-    let base_head_budget = MODEL_FORMAT_HEAD_BYTES.min(MODEL_FORMAT_MAX_BYTES);
-    let head_budget = base_head_budget.min(MODEL_FORMAT_MAX_BYTES.saturating_sub(marker_len));
-    let head_part = take_bytes_at_char_boundary(head_slice, head_budget);
-    let mut result = String::with_capacity(MODEL_FORMAT_MAX_BYTES.min(content.len()));
-
-    result.push_str(head_part);
-    if let Some(marker_text) = marker.as_ref() {
-        result.push_str(marker_text);
-    }
-
-    let remaining = MODEL_FORMAT_MAX_BYTES.saturating_sub(result.len());
-    if remaining == 0 {
-        return result;
-    }
-
-    let tail_part = take_last_bytes_at_char_boundary(tail_slice, remaining);
-    result.push_str(tail_part);
-
-    result
-}
-
-#[cfg(test)]
-mod tests {
-    use super::*;
-    use crate::function_tool::FunctionCallError;
-    use regex_lite::Regex;
-
-    fn truncate_function_error(err: FunctionCallError) -> FunctionCallError {
-        match err {
-            FunctionCallError::RespondToModel(msg) => {
-                FunctionCallError::RespondToModel(format_exec_output(&msg))
-            }
-            FunctionCallError::Denied(msg) => FunctionCallError::Denied(format_exec_output(&msg)),
-            FunctionCallError::Fatal(msg) => FunctionCallError::Fatal(format_exec_output(&msg)),
-            other => other,
-        }
-    }
-
-    fn assert_truncated_message_matches(message: &str, line: &str, total_lines: usize) {
-        let pattern = truncated_message_pattern(line, total_lines);
-        let regex = Regex::new(&pattern).unwrap_or_else(|err| {
-            panic!("failed to compile regex {pattern}: {err}");
-        });
-        let captures = regex
-            .captures(message)
-            .unwrap_or_else(|| panic!("message failed to match pattern {pattern}: {message}"));
-        let body = captures
-            .name("body")
-            .expect("missing body capture")
-            .as_str();
-        assert!(
-            body.len() <= MODEL_FORMAT_MAX_BYTES,
-            "body exceeds byte limit: {} bytes",
-            body.len()
-        );
-    }
-
-    fn truncated_message_pattern(line: &str, total_lines: usize) -> String {
-        let head_take = MODEL_FORMAT_HEAD_LINES.min(total_lines);
-        let tail_take = MODEL_FORMAT_TAIL_LINES.min(total_lines.saturating_sub(head_take));
-        let omitted = total_lines.saturating_sub(head_take + tail_take);
-        let escaped_line = regex_lite::escape(line);
-        if omitted == 0 {
-            return format!(
-                r"(?s)^Total output lines: {total_lines}\n\n(?P<body>{escaped_line}.*\n\[\.{{3}} output truncated to fit {MODEL_FORMAT_MAX_BYTES} bytes \.{{3}}]\n\n.*)$",
-            );
-        }
-        format!(
-            r"(?s)^Total output lines: {total_lines}\n\n(?P<body>{escaped_line}.*\n\[\.{{3}} omitted {omitted} of {total_lines} lines \.{{3}}]\n\n.*)$",
         )
-    }
+    } else {
+        content.to_string()
+    };
 
-    #[test]
-    fn truncate_formatted_exec_output_truncates_large_error() {
-        let line = "very long execution error line that should trigger truncation\n";
-        let large_error = line.repeat(2_500); // way beyond both byte and line limits
-
-        let truncated = format_exec_output(&large_error);
-
-        let total_lines = large_error.lines().count();
-        assert_truncated_message_matches(&truncated, line, total_lines);
-        assert_ne!(truncated, large_error);
-    }
-
-    #[test]
-    fn truncate_function_error_trims_respond_to_model() {
-        let line = "respond-to-model error that should be truncated\n";
-        let huge = line.repeat(3_000);
-        let total_lines = huge.lines().count();
-
-        let err = truncate_function_error(FunctionCallError::RespondToModel(huge));
-        match err {
-            FunctionCallError::RespondToModel(message) => {
-                assert_truncated_message_matches(&message, line, total_lines);
-            }
-            other => panic!("unexpected error variant: {other:?}"),
-        }
-    }
-
-    #[test]
-    fn truncate_function_error_trims_fatal() {
-        let line = "fatal error output that should be truncated\n";
-        let huge = line.repeat(3_000);
-        let total_lines = huge.lines().count();
-
-        let err = truncate_function_error(FunctionCallError::Fatal(huge));
-        match err {
-            FunctionCallError::Fatal(message) => {
-                assert_truncated_message_matches(&message, line, total_lines);
-            }
-            other => panic!("unexpected error variant: {other:?}"),
-        }
-    }
-
-    #[test]
-    fn truncate_formatted_exec_output_marks_byte_truncation_without_omitted_lines() {
-        let long_line = "a".repeat(MODEL_FORMAT_MAX_BYTES + 50);
-        let truncated = format_exec_output(&long_line);
-
-        assert_ne!(truncated, long_line);
-        let marker_line =
-            format!("[... output truncated to fit {MODEL_FORMAT_MAX_BYTES} bytes ...]");
-        assert!(
-            truncated.contains(&marker_line),
-            "missing byte truncation marker: {truncated}"
-        );
-        assert!(
-            !truncated.contains("omitted"),
-            "line omission marker should not appear when no lines were dropped: {truncated}"
-        );
-    }
-
-    #[test]
-    fn truncate_formatted_exec_output_returns_original_when_within_limits() {
-        let content = "example output\n".repeat(10);
-
-        assert_eq!(format_exec_output(&content), content);
-    }
-
-    #[test]
-    fn truncate_formatted_exec_output_reports_omitted_lines_and_keeps_head_and_tail() {
-        let total_lines = MODEL_FORMAT_MAX_LINES + 100;
-        let content: String = (0..total_lines)
-            .map(|idx| format!("line-{idx}\n"))
-            .collect();
-
-        let truncated = format_exec_output(&content);
-        let omitted = total_lines - MODEL_FORMAT_MAX_LINES;
-        let expected_marker = format!("[... omitted {omitted} of {total_lines} lines ...]");
-
-        assert!(
-            truncated.contains(&expected_marker),
-            "missing omitted marker: {truncated}"
-        );
-        assert!(
-            truncated.contains("line-0\n"),
-            "expected head line to remain: {truncated}"
-        );
-
-        let last_line = format!("line-{}\n", total_lines - 1);
-        assert!(
-            truncated.contains(&last_line),
-            "expected tail line to remain: {truncated}"
-        );
-    }
-
-    #[test]
-    fn truncate_formatted_exec_output_prefers_line_marker_when_both_limits_exceeded() {
-        let total_lines = MODEL_FORMAT_MAX_LINES + 42;
-        let long_line = "x".repeat(256);
-        let content: String = (0..total_lines)
-            .map(|idx| format!("line-{idx}-{long_line}\n"))
-            .collect();
-
-        let truncated = format_exec_output(&content);
-
-        assert!(
-            truncated.contains("[... omitted 42 of 298 lines ...]"),
-            "expected omitted marker when line count exceeds limit: {truncated}"
-        );
-        assert!(
-            !truncated.contains("output truncated to fit"),
-            "line omission marker should take precedence over byte marker: {truncated}"
-        );
-    }
+    // Truncate for model consumption before serialization.
+    format_output_for_model_body(&body)
 }
diff --git a/codex-rs/core/src/tools/parallel.rs b/codex-rs/core/src/tools/parallel.rs
index 449b8e6553..7340f5cc9c 100644
--- a/codex-rs/core/src/tools/parallel.rs
+++ b/codex-rs/core/src/tools/parallel.rs
@@ -105,7 +105,7 @@ impl ToolCallRuntime {
                 call_id: call.call_id.clone(),
                 output: FunctionCallOutputPayload {
                     content: "aborted".to_string(),
-                    success: None,
+                    ..Default::default()
                 },
             },
         }
diff --git a/codex-rs/core/src/tools/router.rs b/codex-rs/core/src/tools/router.rs
index 161997fb6c..19098aa80d 100644
--- a/codex-rs/core/src/tools/router.rs
+++ b/codex-rs/core/src/tools/router.rs
@@ -181,6 +181,7 @@ impl ToolRouter {
                 output: codex_protocol::models::FunctionCallOutputPayload {
                     content: message,
                     success: Some(false),
+                    ..Default::default()
                 },
             }
         }
diff --git a/codex-rs/core/tests/suite/cli_stream.rs b/codex-rs/core/tests/suite/cli_stream.rs
index 4c1f5fb8a8..64d6247f3d 100644
--- a/codex-rs/core/tests/suite/cli_stream.rs
+++ b/codex-rs/core/tests/suite/cli_stream.rs
@@ -1,4 +1,5 @@
 use assert_cmd::Command as AssertCommand;
+use assert_cmd::cargo::cargo_bin;
 use codex_core::RolloutRecorder;
 use codex_core::protocol::GitInfo;
 use core_test_support::fs_wait;
@@ -44,13 +45,9 @@ async fn chat_mode_stream_cli() {
         "model_providers.mock={{ name = \"mock\", base_url = \"{}/v1\", env_key = \"PATH\", wire_api = \"chat\" }}",
         server.uri()
     );
-    let mut cmd = AssertCommand::new("cargo");
-    cmd.arg("run")
-        .arg("-p")
-        .arg("codex-cli")
-        .arg("--quiet")
-        .arg("--")
-        .arg("exec")
+    let bin = cargo_bin("codex");
+    let mut cmd = AssertCommand::new(bin);
+    cmd.arg("exec")
         .arg("--skip-git-repo-check")
         .arg("-c")
         .arg(&provider_override)
@@ -136,13 +133,9 @@ async fn exec_cli_applies_experimental_instructions_file() {
     );
 
     let home = TempDir::new().unwrap();
-    let mut cmd = AssertCommand::new("cargo");
-    cmd.arg("run")
-        .arg("-p")
-        .arg("codex-cli")
-        .arg("--quiet")
-        .arg("--")
-        .arg("exec")
+    let bin = cargo_bin("codex");
+    let mut cmd = AssertCommand::new(bin);
+    cmd.arg("exec")
         .arg("--skip-git-repo-check")
         .arg("-c")
         .arg(&provider_override)
@@ -194,13 +187,9 @@ async fn responses_api_stream_cli() {
         std::path::Path::new(env!("CARGO_MANIFEST_DIR")).join("tests/cli_responses_fixture.sse");
 
     let home = TempDir::new().unwrap();
-    let mut cmd = AssertCommand::new("cargo");
-    cmd.arg("run")
-        .arg("-p")
-        .arg("codex-cli")
-        .arg("--quiet")
-        .arg("--")
-        .arg("exec")
+    let bin = cargo_bin("codex");
+    let mut cmd = AssertCommand::new(bin);
+    cmd.arg("exec")
         .arg("--skip-git-repo-check")
         .arg("-C")
         .arg(env!("CARGO_MANIFEST_DIR"))
@@ -233,15 +222,10 @@ async fn integration_creates_and_checks_session_file() -> anyhow::Result<()> {
     let fixture =
         std::path::Path::new(env!("CARGO_MANIFEST_DIR")).join("tests/cli_responses_fixture.sse");
 
-    // 4. Run the codex CLI through cargo (ensures the right bin is built) and invoke `exec`,
-    //    which is what records a session.
-    let mut cmd = AssertCommand::new("cargo");
-    cmd.arg("run")
-        .arg("-p")
-        .arg("codex-cli")
-        .arg("--quiet")
-        .arg("--")
-        .arg("exec")
+    // 4. Run the codex CLI and invoke `exec`, which is what records a session.
+    let bin = cargo_bin("codex");
+    let mut cmd = AssertCommand::new(bin);
+    cmd.arg("exec")
         .arg("--skip-git-repo-check")
         .arg("-C")
         .arg(env!("CARGO_MANIFEST_DIR"))
@@ -360,13 +344,9 @@ async fn integration_creates_and_checks_session_file() -> anyhow::Result<()> {
     // Second run: resume should update the existing file.
     let marker2 = format!("integration-resume-{}", Uuid::new_v4());
     let prompt2 = format!("echo {marker2}");
-    let mut cmd2 = AssertCommand::new("cargo");
-    cmd2.arg("run")
-        .arg("-p")
-        .arg("codex-cli")
-        .arg("--quiet")
-        .arg("--")
-        .arg("exec")
+    let bin2 = cargo_bin("codex");
+    let mut cmd2 = AssertCommand::new(bin2);
+    cmd2.arg("exec")
         .arg("--skip-git-repo-check")
         .arg("-C")
         .arg(env!("CARGO_MANIFEST_DIR"))
diff --git a/codex-rs/core/tests/suite/client.rs b/codex-rs/core/tests/suite/client.rs
index c8ebdcb2c6..02e9cdc361 100644
--- a/codex-rs/core/tests/suite/client.rs
+++ b/codex-rs/core/tests/suite/client.rs
@@ -12,6 +12,7 @@ use codex_core::Prompt;
 use codex_core::ResponseEvent;
 use codex_core::ResponseItem;
 use codex_core::WireApi;
+use codex_core::auth::AuthCredentialsStoreMode;
 use codex_core::built_in_model_providers;
 use codex_core::error::CodexErr;
 use codex_core::model_family::find_family_for_model;
@@ -525,11 +526,12 @@ async fn prefers_apikey_when_config_prefers_apikey_even_with_chatgpt_tokens() {
     let mut config = load_default_config_for_test(&codex_home);
     config.model_provider = model_provider;
 
-    let auth_manager = match CodexAuth::from_auth_storage(codex_home.path()) {
-        Ok(Some(auth)) => codex_core::AuthManager::from_auth_for_testing(auth),
-        Ok(None) => panic!("No CodexAuth found in codex_home"),
-        Err(e) => panic!("Failed to load CodexAuth: {e}"),
-    };
+    let auth_manager =
+        match CodexAuth::from_auth_storage(codex_home.path(), AuthCredentialsStoreMode::File) {
+            Ok(Some(auth)) => codex_core::AuthManager::from_auth_for_testing(auth),
+            Ok(None) => panic!("No CodexAuth found in codex_home"),
+            Err(e) => panic!("Failed to load CodexAuth: {e}"),
+        };
     let conversation_manager = ConversationManager::new(auth_manager, SessionSource::Exec);
     let NewConversation {
         conversation: codex,
diff --git a/codex-rs/core/tests/suite/mod.rs b/codex-rs/core/tests/suite/mod.rs
index bfaef15a92..300ca146ec 100644
--- a/codex-rs/core/tests/suite/mod.rs
+++ b/codex-rs/core/tests/suite/mod.rs
@@ -33,6 +33,7 @@ mod stream_no_completed;
 mod tool_harness;
 mod tool_parallelism;
 mod tools;
+mod truncation;
 mod unified_exec;
 mod user_notification;
 mod view_image;
diff --git a/codex-rs/core/tests/suite/rmcp_client.rs b/codex-rs/core/tests/suite/rmcp_client.rs
index 99c863b8f9..85698e9216 100644
--- a/codex-rs/core/tests/suite/rmcp_client.rs
+++ b/codex-rs/core/tests/suite/rmcp_client.rs
@@ -14,6 +14,8 @@ use codex_core::features::Feature;
 
 use codex_core::protocol::AskForApproval;
 use codex_core::protocol::EventMsg;
+use codex_core::protocol::McpInvocation;
+use codex_core::protocol::McpToolCallBeginEvent;
 use codex_core::protocol::Op;
 use codex_core::protocol::SandboxPolicy;
 use codex_protocol::config_types::ReasoningSummary;
@@ -25,7 +27,9 @@ use core_test_support::test_codex::test_codex;
 use core_test_support::wait_for_event;
 use core_test_support::wait_for_event_with_timeout;
 use escargot::CargoBuild;
+use mcp_types::ContentBlock;
 use serde_json::Value;
+use serde_json::json;
 use serial_test::serial;
 use tempfile::tempdir;
 use tokio::net::TcpStream;
@@ -35,6 +39,8 @@ use tokio::time::Instant;
 use tokio::time::sleep;
 use wiremock::matchers::any;
 
+static OPENAI_PNG: &str = "data:image/png;base64,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";
+
 #[tokio::test(flavor = "multi_thread", worker_threads = 1)]
 #[serial(mcp_test_value)]
 async fn stdio_server_round_trip() -> anyhow::Result<()> {
@@ -175,6 +181,352 @@ async fn stdio_server_round_trip() -> anyhow::Result<()> {
     Ok(())
 }
 
+#[tokio::test(flavor = "multi_thread", worker_threads = 1)]
+#[serial(mcp_test_value)]
+async fn stdio_image_responses_round_trip() -> anyhow::Result<()> {
+    skip_if_no_network!(Ok(()));
+
+    let server = responses::start_mock_server().await;
+
+    let call_id = "img-1";
+    let server_name = "rmcp";
+    let tool_name = format!("mcp__{server_name}__image");
+
+    // First stream: model decides to call the image tool.
+    mount_sse_once_match(
+        &server,
+        any(),
+        responses::sse(vec![
+            responses::ev_response_created("resp-1"),
+            responses::ev_function_call(call_id, &tool_name, "{}"),
+            responses::ev_completed("resp-1"),
+        ]),
+    )
+    .await;
+    // Second stream: after tool execution, assistant emits a message and completes.
+    let final_mock = mount_sse_once_match(
+        &server,
+        any(),
+        responses::sse(vec![
+            responses::ev_assistant_message("msg-1", "rmcp image tool completed successfully."),
+            responses::ev_completed("resp-2"),
+        ]),
+    )
+    .await;
+
+    // Build the stdio rmcp server and pass the image as data URL so it can construct ImageContent.
+    let rmcp_test_server_bin = CargoBuild::new()
+        .package("codex-rmcp-client")
+        .bin("test_stdio_server")
+        .run()?
+        .path()
+        .to_string_lossy()
+        .into_owned();
+
+    let fixture = test_codex()
+        .with_config(move |config| {
+            config.features.enable(Feature::RmcpClient);
+            config.mcp_servers.insert(
+                server_name.to_string(),
+                McpServerConfig {
+                    transport: McpServerTransportConfig::Stdio {
+                        command: rmcp_test_server_bin,
+                        args: Vec::new(),
+                        env: Some(HashMap::from([(
+                            "MCP_TEST_IMAGE_DATA_URL".to_string(),
+                            OPENAI_PNG.to_string(),
+                        )])),
+                        env_vars: Vec::new(),
+                        cwd: None,
+                    },
+                    enabled: true,
+                    startup_timeout_sec: Some(Duration::from_secs(10)),
+                    tool_timeout_sec: None,
+                    enabled_tools: None,
+                    disabled_tools: None,
+                },
+            );
+        })
+        .build(&server)
+        .await?;
+    let session_model = fixture.session_configured.model.clone();
+
+    fixture
+        .codex
+        .submit(Op::UserTurn {
+            items: vec![UserInput::Text {
+                text: "call the rmcp image tool".into(),
+            }],
+            final_output_json_schema: None,
+            cwd: fixture.cwd.path().to_path_buf(),
+            approval_policy: AskForApproval::Never,
+            sandbox_policy: SandboxPolicy::ReadOnly,
+            model: session_model,
+            effort: None,
+            summary: ReasoningSummary::Auto,
+        })
+        .await?;
+
+    // Wait for tool begin/end and final completion.
+    let begin_event = wait_for_event_with_timeout(
+        &fixture.codex,
+        |ev| matches!(ev, EventMsg::McpToolCallBegin(_)),
+        Duration::from_secs(10),
+    )
+    .await;
+    let EventMsg::McpToolCallBegin(begin) = begin_event else {
+        unreachable!("begin");
+    };
+    assert_eq!(
+        begin,
+        McpToolCallBeginEvent {
+            call_id: call_id.to_string(),
+            invocation: McpInvocation {
+                server: server_name.to_string(),
+                tool: "image".to_string(),
+                arguments: Some(json!({})),
+            },
+        },
+    );
+
+    let end_event = wait_for_event(&fixture.codex, |ev| {
+        matches!(ev, EventMsg::McpToolCallEnd(_))
+    })
+    .await;
+    let EventMsg::McpToolCallEnd(end) = end_event else {
+        unreachable!("end");
+    };
+    assert_eq!(end.call_id, call_id);
+    assert_eq!(
+        end.invocation,
+        McpInvocation {
+            server: server_name.to_string(),
+            tool: "image".to_string(),
+            arguments: Some(json!({})),
+        }
+    );
+    let result = end.result.expect("rmcp image tool should return success");
+    assert_eq!(result.is_error, Some(false));
+    assert_eq!(result.content.len(), 1);
+    let base64_only = OPENAI_PNG
+        .strip_prefix("data:image/png;base64,")
+        .expect("data url prefix");
+    match &result.content[0] {
+        ContentBlock::ImageContent(img) => {
+            assert_eq!(img.mime_type, "image/png");
+            assert_eq!(img.r#type, "image");
+            assert_eq!(img.data, base64_only);
+        }
+        other => panic!("expected image content, got {other:?}"),
+    }
+
+    wait_for_event(&fixture.codex, |ev| matches!(ev, EventMsg::TaskComplete(_))).await;
+
+    let output_item = final_mock.single_request().function_call_output(call_id);
+    assert_eq!(
+        output_item,
+        json!({
+            "type": "function_call_output",
+            "call_id": call_id,
+            "output": [{
+                "type": "input_image",
+                "image_url": OPENAI_PNG
+            }]
+        })
+    );
+    server.verify().await;
+    Ok(())
+}
+
+#[tokio::test(flavor = "multi_thread", worker_threads = 1)]
+#[serial(mcp_test_value)]
+async fn stdio_image_completions_round_trip() -> anyhow::Result<()> {
+    skip_if_no_network!(Ok(()));
+
+    let server = responses::start_mock_server().await;
+
+    let call_id = "img-cc-1";
+    let server_name = "rmcp";
+    let tool_name = format!("mcp__{server_name}__image");
+
+    let tool_call = json!({
+        "choices": [
+            {
+                "delta": {
+                    "tool_calls": [
+                        {
+                            "id": call_id,
+                            "type": "function",
+                            "function": {"name": tool_name, "arguments": "{}"}
+                        }
+                    ]
+                },
+                "finish_reason": "tool_calls"
+            }
+        ]
+    });
+    let sse_tool_call = format!(
+        "data: {}\n\ndata: [DONE]\n\n",
+        serde_json::to_string(&tool_call)?
+    );
+
+    let final_assistant = json!({
+        "choices": [
+            {
+                "delta": {"content": "rmcp image tool completed successfully."},
+                "finish_reason": "stop"
+            }
+        ]
+    });
+    let sse_final = format!(
+        "data: {}\n\ndata: [DONE]\n\n",
+        serde_json::to_string(&final_assistant)?
+    );
+
+    use std::sync::atomic::AtomicUsize;
+    use std::sync::atomic::Ordering;
+    struct ChatSeqResponder {
+        num_calls: AtomicUsize,
+        bodies: Vec<String>,
+    }
+    impl wiremock::Respond for ChatSeqResponder {
+        fn respond(&self, _: &wiremock::Request) -> wiremock::ResponseTemplate {
+            let idx = self.num_calls.fetch_add(1, Ordering::SeqCst);
+            match self.bodies.get(idx) {
+                Some(body) => wiremock::ResponseTemplate::new(200)
+                    .insert_header("content-type", "text/event-stream")
+                    .set_body_string(body.clone()),
+                None => panic!("no chat completion response for index {idx}"),
+            }
+        }
+    }
+
+    let chat_seq = ChatSeqResponder {
+        num_calls: AtomicUsize::new(0),
+        bodies: vec![sse_tool_call, sse_final],
+    };
+    wiremock::Mock::given(wiremock::matchers::method("POST"))
+        .and(wiremock::matchers::path("/v1/chat/completions"))
+        .respond_with(chat_seq)
+        .expect(2)
+        .mount(&server)
+        .await;
+
+    let rmcp_test_server_bin = CargoBuild::new()
+        .package("codex-rmcp-client")
+        .bin("test_stdio_server")
+        .run()?
+        .path()
+        .to_string_lossy()
+        .into_owned();
+
+    let fixture = test_codex()
+        .with_config(move |config| {
+            config.model_provider.wire_api = codex_core::WireApi::Chat;
+            config.features.enable(Feature::RmcpClient);
+            config.mcp_servers.insert(
+                server_name.to_string(),
+                McpServerConfig {
+                    transport: McpServerTransportConfig::Stdio {
+                        command: rmcp_test_server_bin,
+                        args: Vec::new(),
+                        env: Some(HashMap::from([(
+                            "MCP_TEST_IMAGE_DATA_URL".to_string(),
+                            OPENAI_PNG.to_string(),
+                        )])),
+                        env_vars: Vec::new(),
+                        cwd: None,
+                    },
+                    enabled: true,
+                    startup_timeout_sec: Some(Duration::from_secs(10)),
+                    tool_timeout_sec: None,
+                    enabled_tools: None,
+                    disabled_tools: None,
+                },
+            );
+        })
+        .build(&server)
+        .await?;
+    let session_model = fixture.session_configured.model.clone();
+
+    fixture
+        .codex
+        .submit(Op::UserTurn {
+            items: vec![UserInput::Text {
+                text: "call the rmcp image tool".into(),
+            }],
+            final_output_json_schema: None,
+            cwd: fixture.cwd.path().to_path_buf(),
+            approval_policy: AskForApproval::Never,
+            sandbox_policy: SandboxPolicy::ReadOnly,
+            model: session_model,
+            effort: None,
+            summary: ReasoningSummary::Auto,
+        })
+        .await?;
+
+    let begin_event = wait_for_event_with_timeout(
+        &fixture.codex,
+        |ev| matches!(ev, EventMsg::McpToolCallBegin(_)),
+        Duration::from_secs(10),
+    )
+    .await;
+    let EventMsg::McpToolCallBegin(begin) = begin_event else {
+        unreachable!("begin");
+    };
+    assert_eq!(
+        begin,
+        McpToolCallBeginEvent {
+            call_id: call_id.to_string(),
+            invocation: McpInvocation {
+                server: server_name.to_string(),
+                tool: "image".to_string(),
+                arguments: Some(json!({})),
+            },
+        },
+    );
+
+    let end_event = wait_for_event(&fixture.codex, |ev| {
+        matches!(ev, EventMsg::McpToolCallEnd(_))
+    })
+    .await;
+    let EventMsg::McpToolCallEnd(end) = end_event else {
+        unreachable!("end");
+    };
+    assert!(end.result.as_ref().is_ok(), "tool call should succeed");
+
+    wait_for_event(&fixture.codex, |ev| matches!(ev, EventMsg::TaskComplete(_))).await;
+
+    // Chat Completions assertion: the second POST should include a tool role message
+    // with an array `content` containing an item with the expected data URL.
+    let requests = server.received_requests().await.expect("requests captured");
+    assert!(requests.len() >= 2, "expected two chat completion calls");
+    let second = &requests[1];
+    let body: Value = serde_json::from_slice(&second.body)?;
+    let messages = body
+        .get("messages")
+        .and_then(Value::as_array)
+        .cloned()
+        .expect("messages array");
+    let tool_msg = messages
+        .iter()
+        .find(|m| {
+            m.get("role") == Some(&json!("tool")) && m.get("tool_call_id") == Some(&json!(call_id))
+        })
+        .cloned()
+        .expect("tool message present");
+    assert_eq!(
+        tool_msg,
+        json!({
+            "role": "tool",
+            "tool_call_id": call_id,
+            "content": [{"type": "image_url", "image_url": {"url": OPENAI_PNG}}]
+        })
+    );
+
+    Ok(())
+}
+
 #[tokio::test(flavor = "multi_thread", worker_threads = 1)]
 #[serial(mcp_test_value)]
 async fn stdio_server_propagates_whitelisted_env_vars() -> anyhow::Result<()> {
diff --git a/codex-rs/core/tests/suite/truncation.rs b/codex-rs/core/tests/suite/truncation.rs
new file mode 100644
index 0000000000..74d90f1de7
--- /dev/null
+++ b/codex-rs/core/tests/suite/truncation.rs
@@ -0,0 +1,270 @@
+#![cfg(not(target_os = "windows"))]
+#![allow(clippy::unwrap_used, clippy::expect_used)]
+
+use anyhow::Context;
+use anyhow::Result;
+use codex_core::features::Feature;
+use codex_core::model_family::find_family_for_model;
+use codex_core::protocol::SandboxPolicy;
+use core_test_support::assert_regex_match;
+use core_test_support::responses;
+use core_test_support::responses::ev_assistant_message;
+use core_test_support::responses::ev_completed;
+use core_test_support::responses::ev_function_call;
+use core_test_support::responses::ev_response_created;
+use core_test_support::responses::mount_sse_once_match;
+use core_test_support::responses::mount_sse_sequence;
+use core_test_support::responses::sse;
+use core_test_support::responses::start_mock_server;
+use core_test_support::skip_if_no_network;
+use core_test_support::test_codex::test_codex;
+use escargot::CargoBuild;
+use regex_lite::Regex;
+use serde_json::Value;
+use serde_json::json;
+use wiremock::matchers::any;
+
+// Verifies byte-truncation formatting for function error output (RespondToModel errors)
+#[tokio::test(flavor = "multi_thread", worker_threads = 2)]
+async fn truncate_function_error_trims_respond_to_model() -> Result<()> {
+    skip_if_no_network!(Ok(()));
+
+    let server = start_mock_server().await;
+    let mut builder = test_codex().with_config(|config| {
+        // Use the test model that wires function tools like grep_files
+        config.model = "test-gpt-5-codex".to_string();
+        config.model_family =
+            find_family_for_model("test-gpt-5-codex").expect("model family for test model");
+    });
+    let test = builder.build(&server).await?;
+
+    // Construct a very long, non-existent path to force a RespondToModel error with a large message
+    let long_path = "a".repeat(20_000);
+    let call_id = "grep-huge-error";
+    let args = json!({
+        "pattern": "alpha",
+        "path": long_path,
+        "limit": 10
+    });
+    let responses = vec![
+        sse(vec![
+            ev_response_created("resp-1"),
+            ev_function_call(call_id, "grep_files", &serde_json::to_string(&args)?),
+            ev_completed("resp-1"),
+        ]),
+        sse(vec![
+            ev_assistant_message("msg-1", "done"),
+            ev_completed("resp-2"),
+        ]),
+    ];
+    let mock = mount_sse_sequence(&server, responses).await;
+
+    test.submit_turn_with_policy(
+        "trigger grep_files with long path to test truncation",
+        SandboxPolicy::DangerFullAccess,
+    )
+    .await?;
+
+    let output = mock
+        .function_call_output_text(call_id)
+        .context("function error output present")?;
+
+    tracing::debug!(output = %output, "truncated function error output");
+
+    // Expect plaintext with byte-truncation marker and no omitted-lines marker
+    assert!(
+        serde_json::from_str::<serde_json::Value>(&output).is_err(),
+        "expected error output to be plain text",
+    );
+    let truncated_pattern = r#"(?s)^Total output lines: 1\s+.*\[\.\.\. output truncated to fit 10240 bytes \.\.\.\]\s*$"#;
+    assert_regex_match(truncated_pattern, &output);
+    assert!(
+        !output.contains("omitted"),
+        "line omission marker should not appear when no lines were dropped: {output}"
+    );
+
+    Ok(())
+}
+
+// Verifies that a standard tool call (shell) exceeding the model formatting
+// limits is truncated before being sent back to the model.
+#[tokio::test(flavor = "multi_thread", worker_threads = 2)]
+async fn tool_call_output_exceeds_limit_truncated_for_model() -> Result<()> {
+    skip_if_no_network!(Ok(()));
+
+    let server = start_mock_server().await;
+
+    // Use a model that exposes the generic shell tool.
+    let mut builder = test_codex().with_config(|config| {
+        config.model = "gpt-5-codex".to_string();
+        config.model_family =
+            find_family_for_model("gpt-5-codex").expect("gpt-5-codex is a model family");
+    });
+    let fixture = builder.build(&server).await?;
+
+    let call_id = "shell-too-large";
+    let args = serde_json::json!({
+        "command": ["/bin/sh", "-c", "seq 1 400"],
+        "timeout_ms": 5_000,
+    });
+
+    // First response: model tells us to run the tool; second: complete the turn.
+    mount_sse_once_match(
+        &server,
+        any(),
+        sse(vec![
+            responses::ev_response_created("resp-1"),
+            responses::ev_function_call(call_id, "shell", &serde_json::to_string(&args)?),
+            responses::ev_completed("resp-1"),
+        ]),
+    )
+    .await;
+    let mock2 = mount_sse_once_match(
+        &server,
+        any(),
+        sse(vec![
+            responses::ev_assistant_message("msg-1", "done"),
+            responses::ev_completed("resp-2"),
+        ]),
+    )
+    .await;
+
+    fixture
+        .submit_turn_with_policy("trigger big shell output", SandboxPolicy::DangerFullAccess)
+        .await?;
+
+    // Inspect what we sent back to the model; it should contain a truncated
+    // function_call_output for the shell call.
+    let output = mock2
+        .single_request()
+        .function_call_output_text(call_id)
+        .context("function_call_output present for shell call")?;
+
+    // Expect plain text (not JSON) with truncation markers and line elision.
+    assert!(
+        serde_json::from_str::<Value>(&output).is_err(),
+        "expected truncated shell output to be plain text"
+    );
+    let truncated_pattern = r#"(?s)^Exit code: 0
+Wall time: .* seconds
+Total output lines: 400
+Output:
+1
+2
+3
+4
+5
+6
+.*
+\[\.{3} omitted 144 of 400 lines \.{3}\]
+
+.*
+396
+397
+398
+399
+400
+$"#;
+    assert_regex_match(truncated_pattern, &output);
+
+    Ok(())
+}
+
+// Verifies that an MCP tool call result exceeding the model formatting limits
+// is truncated before being sent back to the model.
+#[tokio::test(flavor = "multi_thread", worker_threads = 1)]
+async fn mcp_tool_call_output_exceeds_limit_truncated_for_model() -> Result<()> {
+    skip_if_no_network!(Ok(()));
+
+    let server = start_mock_server().await;
+
+    let call_id = "rmcp-truncated";
+    let server_name = "rmcp";
+    let tool_name = format!("mcp__{server_name}__echo");
+
+    // Build a very large message to exceed 10KiB once serialized.
+    let large_msg = "long-message-with-newlines-".repeat(600);
+    let args_json = serde_json::json!({ "message": large_msg });
+
+    mount_sse_once_match(
+        &server,
+        any(),
+        sse(vec![
+            responses::ev_response_created("resp-1"),
+            responses::ev_function_call(call_id, &tool_name, &args_json.to_string()),
+            responses::ev_completed("resp-1"),
+        ]),
+    )
+    .await;
+    let mock2 = mount_sse_once_match(
+        &server,
+        any(),
+        sse(vec![
+            responses::ev_assistant_message("msg-1", "rmcp echo tool completed."),
+            responses::ev_completed("resp-2"),
+        ]),
+    )
+    .await;
+
+    // Compile the rmcp stdio test server and configure it.
+    let rmcp_test_server_bin = CargoBuild::new()
+        .package("codex-rmcp-client")
+        .bin("test_stdio_server")
+        .run()?
+        .path()
+        .to_string_lossy()
+        .into_owned();
+
+    let mut builder = test_codex().with_config(move |config| {
+        config.features.enable(Feature::RmcpClient);
+        config.mcp_servers.insert(
+            server_name.to_string(),
+            codex_core::config_types::McpServerConfig {
+                transport: codex_core::config_types::McpServerTransportConfig::Stdio {
+                    command: rmcp_test_server_bin,
+                    args: Vec::new(),
+                    env: None,
+                    env_vars: Vec::new(),
+                    cwd: None,
+                },
+                enabled: true,
+                startup_timeout_sec: Some(std::time::Duration::from_secs(10)),
+                tool_timeout_sec: None,
+                enabled_tools: None,
+                disabled_tools: None,
+            },
+        );
+    });
+    let fixture = builder.build(&server).await?;
+
+    fixture
+        .submit_turn_with_policy(
+            "call the rmcp echo tool with a very large message",
+            SandboxPolicy::ReadOnly,
+        )
+        .await?;
+
+    // The MCP tool call output is converted to a function_call_output for the model.
+    let output = mock2
+        .single_request()
+        .function_call_output_text(call_id)
+        .context("function_call_output present for rmcp call")?;
+
+    // Expect plain text with byte-based truncation marker.
+    assert!(
+        serde_json::from_str::<Value>(&output).is_err(),
+        "expected truncated MCP output to be plain text"
+    );
+    assert!(
+        output.starts_with("Total output lines: 1\n\n{"),
+        "expected total line header and JSON head, got: {output}"
+    );
+    let byte_marker = Regex::new(r"\[\.\.\. output truncated to fit 10240 bytes \.\.\.\]")
+        .expect("compile regex");
+    assert!(
+        byte_marker.is_match(&output),
+        "expected byte truncation marker, got: {output}"
+    );
+
+    Ok(())
+}
diff --git a/codex-rs/exec/src/lib.rs b/codex-rs/exec/src/lib.rs
index 10a4376161..559fd98002 100644
--- a/codex-rs/exec/src/lib.rs
+++ b/codex-rs/exec/src/lib.rs
@@ -249,7 +249,11 @@ pub async fn run_main(cli: Cli, codex_linux_sandbox_exe: Option<PathBuf>) -> any
         std::process::exit(1);
     }
 
-    let auth_manager = AuthManager::shared(config.codex_home.clone(), true);
+    let auth_manager = AuthManager::shared(
+        config.codex_home.clone(),
+        true,
+        config.cli_auth_credentials_store_mode,
+    );
     let conversation_manager = ConversationManager::new(auth_manager.clone(), SessionSource::Exec);
 
     // Handle resume subcommand by resolving a rollout path and using explicit resume API.
diff --git a/codex-rs/exec/tests/suite/resume.rs b/codex-rs/exec/tests/suite/resume.rs
index 320c5766da..24b8cb0bce 100644
--- a/codex-rs/exec/tests/suite/resume.rs
+++ b/codex-rs/exec/tests/suite/resume.rs
@@ -234,10 +234,17 @@ fn exec_resume_preserves_cli_configuration_overrides() -> anyhow::Result<()> {
         stderr.contains("model: gpt-5-high"),
         "stderr missing model override: {stderr}"
     );
-    assert!(
-        stderr.contains("sandbox: workspace-write"),
-        "stderr missing sandbox override: {stderr}"
-    );
+    if cfg!(target_os = "windows") {
+        assert!(
+            stderr.contains("sandbox: read-only"),
+            "stderr missing downgraded sandbox note: {stderr}"
+        );
+    } else {
+        assert!(
+            stderr.contains("sandbox: workspace-write"),
+            "stderr missing sandbox override: {stderr}"
+        );
+    }
 
     let resumed_path = find_session_file_containing_marker(&sessions_dir, &marker2)
         .expect("no resumed session file containing marker2");
diff --git a/codex-rs/keyring-store/Cargo.toml b/codex-rs/keyring-store/Cargo.toml
index 94d3d54493..f662e5d4ff 100644
--- a/codex-rs/keyring-store/Cargo.toml
+++ b/codex-rs/keyring-store/Cargo.toml
@@ -7,5 +7,10 @@ version = { workspace = true }
 workspace = true
 
 [dependencies]
-keyring = { workspace = true }
+keyring = { workspace = true, features = [
+    "apple-native",
+    "crypto-rust",
+    "linux-native-async-persistent",
+    "windows-native",
+] }
 tracing = { workspace = true }
diff --git a/codex-rs/login/src/device_code_auth.rs b/codex-rs/login/src/device_code_auth.rs
index f1999a8303..acaf30ba0c 100644
--- a/codex-rs/login/src/device_code_auth.rs
+++ b/codex-rs/login/src/device_code_auth.rs
@@ -199,6 +199,7 @@ pub async fn run_device_code_login(opts: ServerOptions) -> std::io::Result<()> {
         tokens.id_token,
         tokens.access_token,
         tokens.refresh_token,
+        opts.cli_auth_credentials_store_mode,
     )
     .await
 }
diff --git a/codex-rs/login/src/server.rs b/codex-rs/login/src/server.rs
index 36186ec017..999c19072e 100644
--- a/codex-rs/login/src/server.rs
+++ b/codex-rs/login/src/server.rs
@@ -14,6 +14,7 @@ use crate::pkce::PkceCodes;
 use crate::pkce::generate_pkce;
 use base64::Engine;
 use chrono::Utc;
+use codex_core::auth::AuthCredentialsStoreMode;
 use codex_core::auth::AuthDotJson;
 use codex_core::auth::save_auth;
 use codex_core::default_client::originator;
@@ -39,6 +40,7 @@ pub struct ServerOptions {
     pub open_browser: bool,
     pub force_state: Option<String>,
     pub forced_chatgpt_workspace_id: Option<String>,
+    pub cli_auth_credentials_store_mode: AuthCredentialsStoreMode,
 }
 
 impl ServerOptions {
@@ -46,6 +48,7 @@ impl ServerOptions {
         codex_home: PathBuf,
         client_id: String,
         forced_chatgpt_workspace_id: Option<String>,
+        cli_auth_credentials_store_mode: AuthCredentialsStoreMode,
     ) -> Self {
         Self {
             codex_home,
@@ -55,6 +58,7 @@ impl ServerOptions {
             open_browser: true,
             force_state: None,
             forced_chatgpt_workspace_id,
+            cli_auth_credentials_store_mode,
         }
     }
 }
@@ -270,6 +274,7 @@ async fn process_request(
                         tokens.id_token.clone(),
                         tokens.access_token.clone(),
                         tokens.refresh_token.clone(),
+                        opts.cli_auth_credentials_store_mode,
                     )
                     .await
                     {
@@ -536,6 +541,7 @@ pub(crate) async fn persist_tokens_async(
     id_token: String,
     access_token: String,
     refresh_token: String,
+    auth_credentials_store_mode: AuthCredentialsStoreMode,
 ) -> io::Result<()> {
     // Reuse existing synchronous logic but run it off the async runtime.
     let codex_home = codex_home.to_path_buf();
@@ -557,7 +563,7 @@ pub(crate) async fn persist_tokens_async(
             tokens: Some(tokens),
             last_refresh: Some(Utc::now()),
         };
-        save_auth(&codex_home, &auth)
+        save_auth(&codex_home, &auth, auth_credentials_store_mode)
     })
     .await
     .map_err(|e| io::Error::other(format!("persist task failed: {e}")))?
diff --git a/codex-rs/login/tests/suite/device_code_login.rs b/codex-rs/login/tests/suite/device_code_login.rs
index 195d72573d..266930e419 100644
--- a/codex-rs/login/tests/suite/device_code_login.rs
+++ b/codex-rs/login/tests/suite/device_code_login.rs
@@ -3,6 +3,7 @@
 use anyhow::Context;
 use base64::Engine;
 use base64::engine::general_purpose::URL_SAFE_NO_PAD;
+use codex_core::auth::AuthCredentialsStoreMode;
 use codex_core::auth::load_auth_dot_json;
 use codex_login::ServerOptions;
 use codex_login::run_device_code_login;
@@ -96,11 +97,16 @@ async fn mock_oauth_token_single(server: &MockServer, jwt: String) {
         .await;
 }
 
-fn server_opts(codex_home: &tempfile::TempDir, issuer: String) -> ServerOptions {
+fn server_opts(
+    codex_home: &tempfile::TempDir,
+    issuer: String,
+    cli_auth_credentials_store_mode: AuthCredentialsStoreMode,
+) -> ServerOptions {
     let mut opts = ServerOptions::new(
         codex_home.path().to_path_buf(),
         "client-id".to_string(),
         None,
+        cli_auth_credentials_store_mode,
     );
     opts.issuer = issuer;
     opts.open_browser = false;
@@ -127,13 +133,13 @@ async fn device_code_login_integration_succeeds() -> anyhow::Result<()> {
     mock_oauth_token_single(&mock_server, jwt.clone()).await;
 
     let issuer = mock_server.uri();
-    let opts = server_opts(&codex_home, issuer);
+    let opts = server_opts(&codex_home, issuer, AuthCredentialsStoreMode::File);
 
     run_device_code_login(opts)
         .await
         .expect("device code login integration should succeed");
 
-    let auth = load_auth_dot_json(codex_home.path())
+    let auth = load_auth_dot_json(codex_home.path(), AuthCredentialsStoreMode::File)
         .context("auth.json should load after login succeeds")?
         .context("auth.json written")?;
     // assert_eq!(auth.openai_api_key.as_deref(), Some("api-key-321"));
@@ -166,7 +172,7 @@ async fn device_code_login_rejects_workspace_mismatch() -> anyhow::Result<()> {
     mock_oauth_token_single(&mock_server, jwt).await;
 
     let issuer = mock_server.uri();
-    let mut opts = server_opts(&codex_home, issuer);
+    let mut opts = server_opts(&codex_home, issuer, AuthCredentialsStoreMode::File);
     opts.forced_chatgpt_workspace_id = Some("org-required".to_string());
 
     let err = run_device_code_login(opts)
@@ -174,8 +180,8 @@ async fn device_code_login_rejects_workspace_mismatch() -> anyhow::Result<()> {
         .expect_err("device code login should fail when workspace mismatches");
     assert_eq!(err.kind(), std::io::ErrorKind::PermissionDenied);
 
-    let auth =
-        load_auth_dot_json(codex_home.path()).context("auth.json should load after login fails")?;
+    let auth = load_auth_dot_json(codex_home.path(), AuthCredentialsStoreMode::File)
+        .context("auth.json should load after login fails")?;
     assert!(
         auth.is_none(),
         "auth.json should not be created when workspace validation fails"
@@ -194,7 +200,7 @@ async fn device_code_login_integration_handles_usercode_http_failure() -> anyhow
 
     let issuer = mock_server.uri();
 
-    let opts = server_opts(&codex_home, issuer);
+    let opts = server_opts(&codex_home, issuer, AuthCredentialsStoreMode::File);
 
     let err = run_device_code_login(opts)
         .await
@@ -205,8 +211,8 @@ async fn device_code_login_integration_handles_usercode_http_failure() -> anyhow
         "unexpected error: {err:?}"
     );
 
-    let auth =
-        load_auth_dot_json(codex_home.path()).context("auth.json should load after login fails")?;
+    let auth = load_auth_dot_json(codex_home.path(), AuthCredentialsStoreMode::File)
+        .context("auth.json should load after login fails")?;
     assert!(
         auth.is_none(),
         "auth.json should not be created when login fails"
@@ -237,6 +243,7 @@ async fn device_code_login_integration_persists_without_api_key_on_exchange_fail
         codex_home.path().to_path_buf(),
         "client-id".to_string(),
         None,
+        AuthCredentialsStoreMode::File,
     );
     opts.issuer = issuer;
     opts.open_browser = false;
@@ -245,7 +252,7 @@ async fn device_code_login_integration_persists_without_api_key_on_exchange_fail
         .await
         .expect("device login should succeed without API key exchange");
 
-    let auth = load_auth_dot_json(codex_home.path())
+    let auth = load_auth_dot_json(codex_home.path(), AuthCredentialsStoreMode::File)
         .context("auth.json should load after login succeeds")?
         .context("auth.json written")?;
     assert!(auth.openai_api_key.is_none());
@@ -286,6 +293,7 @@ async fn device_code_login_integration_handles_error_payload() -> anyhow::Result
         codex_home.path().to_path_buf(),
         "client-id".to_string(),
         None,
+        AuthCredentialsStoreMode::File,
     );
     opts.issuer = issuer;
     opts.open_browser = false;
@@ -300,8 +308,8 @@ async fn device_code_login_integration_handles_error_payload() -> anyhow::Result
         "Expected an authorization_declined / 400 / 404 error, got {err:?}"
     );
 
-    let auth =
-        load_auth_dot_json(codex_home.path()).context("auth.json should load after login fails")?;
+    let auth = load_auth_dot_json(codex_home.path(), AuthCredentialsStoreMode::File)
+        .context("auth.json should load after login fails")?;
     assert!(
         auth.is_none(),
         "auth.json should not be created when device auth fails"
diff --git a/codex-rs/login/tests/suite/login_server_e2e.rs b/codex-rs/login/tests/suite/login_server_e2e.rs
index 3a5f869b5a..73cb8bd420 100644
--- a/codex-rs/login/tests/suite/login_server_e2e.rs
+++ b/codex-rs/login/tests/suite/login_server_e2e.rs
@@ -7,6 +7,7 @@ use std::time::Duration;
 
 use anyhow::Result;
 use base64::Engine;
+use codex_core::auth::AuthCredentialsStoreMode;
 use codex_login::ServerOptions;
 use codex_login::run_login_server;
 use core_test_support::skip_if_no_network;
@@ -110,6 +111,7 @@ async fn end_to_end_login_flow_persists_auth_json() -> Result<()> {
 
     let opts = ServerOptions {
         codex_home: server_home,
+        cli_auth_credentials_store_mode: AuthCredentialsStoreMode::File,
         client_id: codex_login::CLIENT_ID.to_string(),
         issuer,
         port: 0,
@@ -170,6 +172,7 @@ async fn creates_missing_codex_home_dir() -> Result<()> {
     let server_home = codex_home.clone();
     let opts = ServerOptions {
         codex_home: server_home,
+        cli_auth_credentials_store_mode: AuthCredentialsStoreMode::File,
         client_id: codex_login::CLIENT_ID.to_string(),
         issuer,
         port: 0,
@@ -208,6 +211,7 @@ async fn forced_chatgpt_workspace_id_mismatch_blocks_login() -> Result<()> {
 
     let opts = ServerOptions {
         codex_home: codex_home.clone(),
+        cli_auth_credentials_store_mode: AuthCredentialsStoreMode::File,
         client_id: codex_login::CLIENT_ID.to_string(),
         issuer,
         port: 0,
@@ -263,6 +267,7 @@ async fn cancels_previous_login_server_when_port_is_in_use() -> Result<()> {
 
     let first_opts = ServerOptions {
         codex_home: first_codex_home,
+        cli_auth_credentials_store_mode: AuthCredentialsStoreMode::File,
         client_id: codex_login::CLIENT_ID.to_string(),
         issuer: issuer.clone(),
         port: 0,
@@ -282,6 +287,7 @@ async fn cancels_previous_login_server_when_port_is_in_use() -> Result<()> {
 
     let second_opts = ServerOptions {
         codex_home: second_codex_home,
+        cli_auth_credentials_store_mode: AuthCredentialsStoreMode::File,
         client_id: codex_login::CLIENT_ID.to_string(),
         issuer,
         port: login_port,
diff --git a/codex-rs/mcp-server/src/message_processor.rs b/codex-rs/mcp-server/src/message_processor.rs
index b21d5a6503..81eb80764b 100644
--- a/codex-rs/mcp-server/src/message_processor.rs
+++ b/codex-rs/mcp-server/src/message_processor.rs
@@ -53,7 +53,11 @@ impl MessageProcessor {
         config: Arc<Config>,
     ) -> Self {
         let outgoing = Arc::new(outgoing);
-        let auth_manager = AuthManager::shared(config.codex_home.clone(), false);
+        let auth_manager = AuthManager::shared(
+            config.codex_home.clone(),
+            false,
+            config.cli_auth_credentials_store_mode,
+        );
         let conversation_manager =
             Arc::new(ConversationManager::new(auth_manager, SessionSource::Mcp));
         Self {
diff --git a/codex-rs/protocol/src/account.rs b/codex-rs/protocol/src/account.rs
index 1d63910c64..fb707c3a73 100644
--- a/codex-rs/protocol/src/account.rs
+++ b/codex-rs/protocol/src/account.rs
@@ -18,18 +18,3 @@ pub enum PlanType {
     #[serde(other)]
     Unknown,
 }
-
-#[derive(Debug, Clone, PartialEq, Deserialize, Serialize, JsonSchema, TS)]
-#[serde(tag = "type")]
-#[ts(tag = "type")]
-pub enum Account {
-    ApiKey {
-        api_key: String,
-    },
-    #[serde(rename = "chatgpt")]
-    #[ts(rename = "chatgpt")]
-    ChatGpt {
-        email: Option<String>,
-        plan_type: PlanType,
-    },
-}
diff --git a/codex-rs/protocol/src/models.rs b/codex-rs/protocol/src/models.rs
index 4430a0998a..614a5ff2b0 100644
--- a/codex-rs/protocol/src/models.rs
+++ b/codex-rs/protocol/src/models.rs
@@ -3,6 +3,7 @@ use std::collections::HashMap;
 use base64::Engine;
 use codex_utils_image::load_and_resize_to_fit;
 use mcp_types::CallToolResult;
+use mcp_types::ContentBlock;
 use serde::Deserialize;
 use serde::Deserializer;
 use serde::Serialize;
@@ -83,9 +84,8 @@ pub enum ResponseItem {
     // NOTE: The input schema for `function_call_output` objects that clients send to the
     // OpenAI /v1/responses endpoint is NOT the same shape as the objects the server returns on the
     // SSE stream. When *sending* we must wrap the string output inside an object that includes a
-    // required `success` boolean. The upstream TypeScript CLI does this implicitly. To ensure we
-    // serialize exactly the expected shape we introduce a dedicated payload struct and flatten it
-    // here.
+    // required `success` boolean. To ensure we serialize exactly the expected shape we introduce
+    // a dedicated payload struct and flatten it here.
     FunctionCallOutput {
         call_id: String,
         output: FunctionCallOutputPayload,
@@ -160,19 +160,17 @@ impl From<ResponseInputItem> for ResponseItem {
             ResponseInputItem::FunctionCallOutput { call_id, output } => {
                 Self::FunctionCallOutput { call_id, output }
             }
-            ResponseInputItem::McpToolCallOutput { call_id, result } => Self::FunctionCallOutput {
-                call_id,
-                output: FunctionCallOutputPayload {
-                    success: Some(result.is_ok()),
-                    content: result.map_or_else(
-                        |tool_call_err| format!("err: {tool_call_err:?}"),
-                        |result| {
-                            serde_json::to_string(&result)
-                                .unwrap_or_else(|e| format!("JSON serialization error: {e}"))
-                        },
-                    ),
-                },
-            },
+            ResponseInputItem::McpToolCallOutput { call_id, result } => {
+                let output = match result {
+                    Ok(result) => FunctionCallOutputPayload::from(&result),
+                    Err(tool_call_err) => FunctionCallOutputPayload {
+                        content: format!("err: {tool_call_err:?}"),
+                        success: Some(false),
+                        ..Default::default()
+                    },
+                };
+                Self::FunctionCallOutput { call_id, output }
+            }
             ResponseInputItem::CustomToolCallOutput { call_id, output } => {
                 Self::CustomToolCallOutput { call_id, output }
             }
@@ -290,31 +288,53 @@ pub struct ShellToolCallParams {
     pub justification: Option<String>,
 }
 
-#[derive(Debug, Clone, PartialEq, JsonSchema, TS)]
+/// Responses API compatible content items that can be returned by a tool call.
+/// This is a subset of ContentItem with the types we support as function call outputs.
+#[derive(Debug, Clone, Serialize, Deserialize, PartialEq, JsonSchema, TS)]
+#[serde(tag = "type", rename_all = "snake_case")]
+pub enum FunctionCallOutputContentItem {
+    // Do not rename, these are serialized and used directly in the responses API.
+    InputText { text: String },
+    // Do not rename, these are serialized and used directly in the responses API.
+    InputImage { image_url: String },
+}
+
+/// The payload we send back to OpenAI when reporting a tool call result.
+///
+/// `content` preserves the historical plain-string payload so downstream
+/// integrations (tests, logging, etc.) can keep treating tool output as
+/// `String`. When an MCP server returns richer data we additionally populate
+/// `content_items` with the structured form that the Responses/Chat
+/// Completions APIs understand.
+#[derive(Debug, Default, Clone, PartialEq, JsonSchema, TS)]
 pub struct FunctionCallOutputPayload {
     pub content: String,
+    #[serde(skip_serializing_if = "Option::is_none")]
+    pub content_items: Option<Vec<FunctionCallOutputContentItem>>,
     // TODO(jif) drop this.
     pub success: Option<bool>,
 }
 
+#[derive(Deserialize)]
+#[serde(untagged)]
+enum FunctionCallOutputPayloadSerde {
+    Text(String),
+    Items(Vec<FunctionCallOutputContentItem>),
+}
+
 // The Responses API expects two *different* shapes depending on success vs failure:
 //   • success → output is a plain string (no nested object)
 //   • failure → output is an object { content, success:false }
-// The upstream TypeScript CLI implements this by special‑casing the serialize path.
-// We replicate that behavior with a manual Serialize impl.
-
 impl Serialize for FunctionCallOutputPayload {
     fn serialize<S>(&self, serializer: S) -> Result<S::Ok, S::Error>
     where
         S: Serializer,
     {
-        // The upstream TypeScript CLI always serializes `output` as a *plain string* regardless
-        // of whether the function call succeeded or failed. The boolean is purely informational
-        // for local bookkeeping and is NOT sent to the OpenAI endpoint. Sending the nested object
-        // form `{ content, success:false }` triggers the 400 we are still seeing. Mirror the JS CLI
-        // exactly: always emit a bare string.
-
-        serializer.serialize_str(&self.content)
+        if let Some(items) = &self.content_items {
+            items.serialize(serializer)
+        } else {
+            serializer.serialize_str(&self.content)
+        }
     }
 }
 
@@ -323,14 +343,106 @@ impl<'de> Deserialize<'de> for FunctionCallOutputPayload {
     where
         D: Deserializer<'de>,
     {
-        let s = String::deserialize(deserializer)?;
-        Ok(FunctionCallOutputPayload {
-            content: s,
-            success: None,
-        })
+        match FunctionCallOutputPayloadSerde::deserialize(deserializer)? {
+            FunctionCallOutputPayloadSerde::Text(content) => Ok(FunctionCallOutputPayload {
+                content,
+                ..Default::default()
+            }),
+            FunctionCallOutputPayloadSerde::Items(items) => {
+                let content = serde_json::to_string(&items).map_err(serde::de::Error::custom)?;
+                Ok(FunctionCallOutputPayload {
+                    content,
+                    content_items: Some(items),
+                    success: None,
+                })
+            }
+        }
     }
 }
 
+impl From<&CallToolResult> for FunctionCallOutputPayload {
+    fn from(call_tool_result: &CallToolResult) -> Self {
+        let CallToolResult {
+            content,
+            structured_content,
+            is_error,
+        } = call_tool_result;
+
+        let is_success = is_error != &Some(true);
+
+        if let Some(structured_content) = structured_content
+            && !structured_content.is_null()
+        {
+            match serde_json::to_string(structured_content) {
+                Ok(serialized_structured_content) => {
+                    return FunctionCallOutputPayload {
+                        content: serialized_structured_content,
+                        success: Some(is_success),
+                        ..Default::default()
+                    };
+                }
+                Err(err) => {
+                    return FunctionCallOutputPayload {
+                        content: err.to_string(),
+                        success: Some(false),
+                        ..Default::default()
+                    };
+                }
+            }
+        }
+
+        let serialized_content = match serde_json::to_string(content) {
+            Ok(serialized_content) => serialized_content,
+            Err(err) => {
+                return FunctionCallOutputPayload {
+                    content: err.to_string(),
+                    success: Some(false),
+                    ..Default::default()
+                };
+            }
+        };
+
+        let content_items = convert_content_blocks_to_items(content);
+
+        FunctionCallOutputPayload {
+            content: serialized_content,
+            content_items,
+            success: Some(is_success),
+        }
+    }
+}
+
+fn convert_content_blocks_to_items(
+    blocks: &[ContentBlock],
+) -> Option<Vec<FunctionCallOutputContentItem>> {
+    let mut saw_image = false;
+    let mut items = Vec::with_capacity(blocks.len());
+
+    for block in blocks {
+        match block {
+            ContentBlock::TextContent(text) => {
+                items.push(FunctionCallOutputContentItem::InputText {
+                    text: text.text.clone(),
+                });
+            }
+            ContentBlock::ImageContent(image) => {
+                saw_image = true;
+                // Just in case the content doesn't include a data URL, add it.
+                let image_url = if image.data.starts_with("data:") {
+                    image.data.clone()
+                } else {
+                    format!("data:{};base64,{}", image.mime_type, image.data)
+                };
+                items.push(FunctionCallOutputContentItem::InputImage { image_url });
+            }
+            // TODO: render audio, resource, and embedded resource content to the model.
+            _ => return None,
+        }
+    }
+
+    if saw_image { Some(items) } else { None }
+}
+
 // Implement Display so callers can treat the payload like a plain string when logging or doing
 // trivial substring checks in tests (existing tests call `.contains()` on the output). Display
 // returns the raw `content` field.
@@ -354,6 +466,8 @@ impl std::ops::Deref for FunctionCallOutputPayload {
 mod tests {
     use super::*;
     use anyhow::Result;
+    use mcp_types::ImageContent;
+    use mcp_types::TextContent;
     use tempfile::tempdir;
 
     #[test]
@@ -362,7 +476,7 @@ mod tests {
             call_id: "call1".into(),
             output: FunctionCallOutputPayload {
                 content: "ok".into(),
-                success: None,
+                ..Default::default()
             },
         };
 
@@ -381,6 +495,7 @@ mod tests {
             output: FunctionCallOutputPayload {
                 content: "bad".into(),
                 success: Some(false),
+                ..Default::default()
             },
         };
 
@@ -391,6 +506,81 @@ mod tests {
         Ok(())
     }
 
+    #[test]
+    fn serializes_image_outputs_as_array() -> Result<()> {
+        let call_tool_result = CallToolResult {
+            content: vec![
+                ContentBlock::TextContent(TextContent {
+                    annotations: None,
+                    text: "caption".into(),
+                    r#type: "text".into(),
+                }),
+                ContentBlock::ImageContent(ImageContent {
+                    annotations: None,
+                    data: "BASE64".into(),
+                    mime_type: "image/png".into(),
+                    r#type: "image".into(),
+                }),
+            ],
+            is_error: None,
+            structured_content: None,
+        };
+
+        let payload = FunctionCallOutputPayload::from(&call_tool_result);
+        assert_eq!(payload.success, Some(true));
+        let items = payload.content_items.clone().expect("content items");
+        assert_eq!(
+            items,
+            vec![
+                FunctionCallOutputContentItem::InputText {
+                    text: "caption".into(),
+                },
+                FunctionCallOutputContentItem::InputImage {
+                    image_url: "data:image/png;base64,BASE64".into(),
+                },
+            ]
+        );
+
+        let item = ResponseInputItem::FunctionCallOutput {
+            call_id: "call1".into(),
+            output: payload,
+        };
+
+        let json = serde_json::to_string(&item)?;
+        let v: serde_json::Value = serde_json::from_str(&json)?;
+
+        let output = v.get("output").expect("output field");
+        assert!(output.is_array(), "expected array output");
+
+        Ok(())
+    }
+
+    #[test]
+    fn deserializes_array_payload_into_items() -> Result<()> {
+        let json = r#"[
+            {"type": "input_text", "text": "note"},
+            {"type": "input_image", "image_url": "data:image/png;base64,XYZ"}
+        ]"#;
+
+        let payload: FunctionCallOutputPayload = serde_json::from_str(json)?;
+
+        assert_eq!(payload.success, None);
+        let expected_items = vec![
+            FunctionCallOutputContentItem::InputText {
+                text: "note".into(),
+            },
+            FunctionCallOutputContentItem::InputImage {
+                image_url: "data:image/png;base64,XYZ".into(),
+            },
+        ];
+        assert_eq!(payload.content_items, Some(expected_items.clone()));
+
+        let expected_content = serde_json::to_string(&expected_items)?;
+        assert_eq!(payload.content, expected_content);
+
+        Ok(())
+    }
+
     #[test]
     fn deserialize_shell_tool_call_params() -> Result<()> {
         let json = r#"{
diff --git a/codex-rs/protocol/src/protocol.rs b/codex-rs/protocol/src/protocol.rs
index bb14f3797b..a7f5241bb3 100644
--- a/codex-rs/protocol/src/protocol.rs
+++ b/codex-rs/protocol/src/protocol.rs
@@ -814,7 +814,7 @@ pub struct AgentReasoningDeltaEvent {
     pub delta: String,
 }
 
-#[derive(Debug, Clone, Deserialize, Serialize, JsonSchema, TS)]
+#[derive(Debug, Clone, Deserialize, Serialize, JsonSchema, TS, PartialEq)]
 pub struct McpInvocation {
     /// Name of the MCP server as defined in the config.
     pub server: String,
@@ -824,14 +824,14 @@ pub struct McpInvocation {
     pub arguments: Option<serde_json::Value>,
 }
 
-#[derive(Debug, Clone, Deserialize, Serialize, JsonSchema, TS)]
+#[derive(Debug, Clone, Deserialize, Serialize, JsonSchema, TS, PartialEq)]
 pub struct McpToolCallBeginEvent {
     /// Identifier so this can be paired with the McpToolCallEnd event.
     pub call_id: String,
     pub invocation: McpInvocation,
 }
 
-#[derive(Debug, Clone, Deserialize, Serialize, JsonSchema, TS)]
+#[derive(Debug, Clone, Deserialize, Serialize, JsonSchema, TS, PartialEq)]
 pub struct McpToolCallEndEvent {
     /// Identifier for the corresponding McpToolCallBegin that finished.
     pub call_id: String,
diff --git a/codex-rs/rmcp-client/src/bin/test_stdio_server.rs b/codex-rs/rmcp-client/src/bin/test_stdio_server.rs
index 44ae50f02f..aafba59324 100644
--- a/codex-rs/rmcp-client/src/bin/test_stdio_server.rs
+++ b/codex-rs/rmcp-client/src/bin/test_stdio_server.rs
@@ -40,7 +40,7 @@ pub fn stdio() -> (tokio::io::Stdin, tokio::io::Stdout) {
 }
 impl TestToolServer {
     fn new() -> Self {
-        let tools = vec![Self::echo_tool()];
+        let tools = vec![Self::echo_tool(), Self::image_tool()];
         let resources = vec![Self::memo_resource()];
         let resource_templates = vec![Self::memo_template()];
         Self {
@@ -70,6 +70,22 @@ impl TestToolServer {
         )
     }
 
+    fn image_tool() -> Tool {
+        #[expect(clippy::expect_used)]
+        let schema: JsonObject = serde_json::from_value(serde_json::json!({
+            "type": "object",
+            "properties": {},
+            "additionalProperties": false
+        }))
+        .expect("image tool schema should deserialize");
+
+        Tool::new(
+            Cow::Borrowed("image"),
+            Cow::Borrowed("Return a single image content block."),
+            Arc::new(schema),
+        )
+    }
+
     fn memo_resource() -> Resource {
         let raw = RawResource {
             uri: MEMO_URI.to_string(),
@@ -214,6 +230,35 @@ impl ServerHandler for TestToolServer {
                     meta: None,
                 })
             }
+            "image" => {
+                // Read a data URL (e.g. data:image/png;base64,AAA...) from env and convert to
+                // an MCP image content block. Tests set MCP_TEST_IMAGE_DATA_URL.
+                let data_url = std::env::var("MCP_TEST_IMAGE_DATA_URL").map_err(|_| {
+                    McpError::invalid_params(
+                        "missing MCP_TEST_IMAGE_DATA_URL env var for image tool",
+                        None,
+                    )
+                })?;
+
+                fn parse_data_url(url: &str) -> Option<(String, String)> {
+                    let rest = url.strip_prefix("data:")?;
+                    let (mime_and_opts, data) = rest.split_once(',')?;
+                    let (mime, _opts) =
+                        mime_and_opts.split_once(';').unwrap_or((mime_and_opts, ""));
+                    Some((mime.to_string(), data.to_string()))
+                }
+
+                let (mime_type, data_b64) = parse_data_url(&data_url).ok_or_else(|| {
+                    McpError::invalid_params(
+                        format!("invalid data URL for image tool: {data_url}"),
+                        None,
+                    )
+                })?;
+
+                Ok(CallToolResult::success(vec![rmcp::model::Content::image(
+                    data_b64, mime_type,
+                )]))
+            }
             other => Err(McpError::invalid_params(
                 format!("unknown tool: {other}"),
                 None,
diff --git a/codex-rs/tui/src/app.rs b/codex-rs/tui/src/app.rs
index c8eab72969..8fc18a1d9b 100644
--- a/codex-rs/tui/src/app.rs
+++ b/codex-rs/tui/src/app.rs
@@ -369,6 +369,9 @@ impl App {
             AppEvent::OpenFeedbackConsent { category } => {
                 self.chat_widget.open_feedback_consent(category);
             }
+            AppEvent::ShowWindowsAutoModeInstructions => {
+                self.chat_widget.open_windows_auto_mode_instructions();
+            }
             AppEvent::PersistModelSelection { model, effort } => {
                 let profile = self.active_profile.as_deref();
                 match persist_model_selection(&self.config.codex_home, profile, &model, effort)
diff --git a/codex-rs/tui/src/app_event.rs b/codex-rs/tui/src/app_event.rs
index 8b14b0be6f..514e234743 100644
--- a/codex-rs/tui/src/app_event.rs
+++ b/codex-rs/tui/src/app_event.rs
@@ -72,6 +72,9 @@ pub(crate) enum AppEvent {
         preset: ApprovalPreset,
     },
 
+    /// Show Windows Subsystem for Linux setup instructions for auto mode.
+    ShowWindowsAutoModeInstructions,
+
     /// Update the current approval policy in the running app and widget.
     UpdateAskForApprovalPolicy(AskForApproval),
 
diff --git a/codex-rs/tui/src/bottom_pane/chat_composer.rs b/codex-rs/tui/src/bottom_pane/chat_composer.rs
index 7b4d2f7ec4..8691b5b127 100644
--- a/codex-rs/tui/src/bottom_pane/chat_composer.rs
+++ b/codex-rs/tui/src/bottom_pane/chat_composer.rs
@@ -573,6 +573,16 @@ impl ChatComposer {
 
     #[inline]
     fn handle_non_ascii_char(&mut self, input: KeyEvent) -> (InputResult, bool) {
+        if let KeyEvent {
+            code: KeyCode::Char(ch),
+            ..
+        } = input
+        {
+            let now = Instant::now();
+            if self.paste_burst.try_append_char_if_active(ch, now) {
+                return (InputResult::None, true);
+            }
+        }
         if let Some(pasted) = self.paste_burst.flush_before_modified_input() {
             self.handle_paste(pasted);
         }
diff --git a/codex-rs/tui/src/bottom_pane/paste_burst.rs b/codex-rs/tui/src/bottom_pane/paste_burst.rs
index 33a02bdefa..49377cb21c 100644
--- a/codex-rs/tui/src/bottom_pane/paste_burst.rs
+++ b/codex-rs/tui/src/bottom_pane/paste_burst.rs
@@ -163,6 +163,18 @@ impl PasteBurst {
         self.burst_window_until = Some(now + PASTE_ENTER_SUPPRESS_WINDOW);
     }
 
+    /// Try to append a char into the burst buffer only if a burst is already active.
+    ///
+    /// Returns true when the char was captured into the existing burst, false otherwise.
+    pub fn try_append_char_if_active(&mut self, ch: char, now: Instant) -> bool {
+        if self.active || !self.buffer.is_empty() {
+            self.append_char_to_buffer(ch, now);
+            true
+        } else {
+            false
+        }
+    }
+
     /// Decide whether to begin buffering by retroactively capturing recent
     /// chars from the slice before the cursor.
     ///
diff --git a/codex-rs/tui/src/chatwidget.rs b/codex-rs/tui/src/chatwidget.rs
index 13411df404..e1255c84dd 100644
--- a/codex-rs/tui/src/chatwidget.rs
+++ b/codex-rs/tui/src/chatwidget.rs
@@ -88,6 +88,8 @@ use crate::history_cell::AgentMessageCell;
 use crate::history_cell::HistoryCell;
 use crate::history_cell::McpToolCallCell;
 use crate::markdown::append_markdown;
+#[cfg(target_os = "windows")]
+use crate::onboarding::WSL_INSTRUCTIONS;
 use crate::render::renderable::ColumnRenderable;
 use crate::render::renderable::Renderable;
 use crate::slash_command::SlashCommand;
@@ -1217,7 +1219,10 @@ impl ChatWidget {
                 self.app_event_tx.send(AppEvent::ExitRequest);
             }
             SlashCommand::Logout => {
-                if let Err(e) = codex_core::auth::logout(&self.config.codex_home) {
+                if let Err(e) = codex_core::auth::logout(
+                    &self.config.codex_home,
+                    self.config.cli_auth_credentials_store_mode,
+                ) {
                     tracing::error!("failed to logout: {e}");
                 }
                 self.app_event_tx.send(AppEvent::ExitRequest);
@@ -1787,11 +1792,38 @@ impl ChatWidget {
         let current_sandbox = self.config.sandbox_policy.clone();
         let mut items: Vec<SelectionItem> = Vec::new();
         let presets: Vec<ApprovalPreset> = builtin_approval_presets();
+        #[cfg(target_os = "windows")]
+        let header_renderable: Box<dyn Renderable> = if self
+            .config
+            .forced_auto_mode_downgraded_on_windows
+        {
+            use ratatui_macros::line;
+
+            let mut header = ColumnRenderable::new();
+            header.push(line![
+                "Codex forced your settings back to Read Only on this Windows machine.".bold()
+            ]);
+            header.push(line![
+                "To re-enable Auto mode, run Codex inside Windows Subsystem for Linux (WSL) or enable Full Access manually.".dim()
+                ]);
+            Box::new(header)
+        } else {
+            Box::new(())
+        };
+        #[cfg(not(target_os = "windows"))]
+        let header_renderable: Box<dyn Renderable> = Box::new(());
         for preset in presets.into_iter() {
             let is_current =
                 current_approval == preset.approval && current_sandbox == preset.sandbox;
             let name = preset.label.to_string();
-            let description = Some(preset.description.to_string());
+            let description_text = preset.description;
+            let description = if cfg!(target_os = "windows") && preset.id == "auto" {
+                Some(format!(
+                    "{description_text}\nRequires Windows Subsystem for Linux (WSL). Show installation instructions..."
+                ))
+            } else {
+                Some(description_text.to_string())
+            };
             let requires_confirmation = preset.id == "full-access"
                 && !self
                     .config
@@ -1805,6 +1837,10 @@ impl ChatWidget {
                         preset: preset_clone.clone(),
                     });
                 })]
+            } else if cfg!(target_os = "windows") && preset.id == "auto" {
+                vec![Box::new(|tx| {
+                    tx.send(AppEvent::ShowWindowsAutoModeInstructions);
+                })]
             } else {
                 Self::approval_preset_actions(preset.approval, preset.sandbox.clone())
             };
@@ -1822,6 +1858,7 @@ impl ChatWidget {
             title: Some("Select Approval Mode".to_string()),
             footer_hint: Some(standard_popup_hint_line()),
             items,
+            header: header_renderable,
             ..Default::default()
         });
     }
@@ -1909,6 +1946,43 @@ impl ChatWidget {
         });
     }
 
+    #[cfg(target_os = "windows")]
+    pub(crate) fn open_windows_auto_mode_instructions(&mut self) {
+        use ratatui_macros::line;
+
+        let mut header = ColumnRenderable::new();
+        header.push(line![
+            "Auto mode requires Windows Subsystem for Linux (WSL2).".bold()
+        ]);
+        header.push(line!["Run Codex inside WSL to enable sandboxed commands."]);
+        header.push(line![""]);
+        header.push(Paragraph::new(WSL_INSTRUCTIONS).wrap(Wrap { trim: false }));
+
+        let items = vec![SelectionItem {
+            name: "Back".to_string(),
+            description: Some(
+                "Return to the approval mode list. Auto mode stays disabled outside WSL."
+                    .to_string(),
+            ),
+            actions: vec![Box::new(|tx| {
+                tx.send(AppEvent::OpenApprovalsPopup);
+            })],
+            dismiss_on_select: true,
+            ..Default::default()
+        }];
+
+        self.bottom_pane.show_selection_view(SelectionViewParams {
+            title: None,
+            footer_hint: Some(standard_popup_hint_line()),
+            items,
+            header: Box::new(header),
+            ..Default::default()
+        });
+    }
+
+    #[cfg(not(target_os = "windows"))]
+    pub(crate) fn open_windows_auto_mode_instructions(&mut self) {}
+
     /// Set the approval policy in the widget's config copy.
     pub(crate) fn set_approval_policy(&mut self, policy: AskForApproval) {
         self.config.approval_policy = policy;
diff --git a/codex-rs/tui/src/chatwidget/snapshots/codex_tui__chatwidget__tests__approvals_selection_popup.snap b/codex-rs/tui/src/chatwidget/snapshots/codex_tui__chatwidget__tests__approvals_selection_popup.snap
index 1d612966de..190594b1b3 100644
--- a/codex-rs/tui/src/chatwidget/snapshots/codex_tui__chatwidget__tests__approvals_selection_popup.snap
+++ b/codex-rs/tui/src/chatwidget/snapshots/codex_tui__chatwidget__tests__approvals_selection_popup.snap
@@ -6,12 +6,12 @@ expression: popup
 
 › 1. Read Only (current)  Codex can read files and answer questions. Codex
                           requires approval to make edits, run commands, or
-                          access network
+                          access network.
   2. Auto                 Codex can read files, make edits, and run commands
                           in the workspace. Codex requires approval to work
-                          outside the workspace or access network
+                          outside the workspace or access network.
   3. Full Access          Codex can read files, make edits, and run commands
                           with network access, without approval. Exercise
-                          caution
+                          caution.
 
   Press enter to confirm or esc to go back
diff --git a/codex-rs/tui/src/chatwidget/snapshots/codex_tui__chatwidget__tests__approvals_selection_popup@windows.snap b/codex-rs/tui/src/chatwidget/snapshots/codex_tui__chatwidget__tests__approvals_selection_popup@windows.snap
new file mode 100644
index 0000000000..7d16ad57b8
--- /dev/null
+++ b/codex-rs/tui/src/chatwidget/snapshots/codex_tui__chatwidget__tests__approvals_selection_popup@windows.snap
@@ -0,0 +1,19 @@
+---
+source: tui/src/chatwidget/tests.rs
+expression: popup
+---
+  Select Approval Mode
+
+› 1. Read Only (current)  Codex can read files and answer questions. Codex
+                          requires approval to make edits, run commands, or
+                          access network.
+  2. Auto                 Codex can read files, make edits, and run commands
+                          in the workspace. Codex requires approval to work
+                          outside the workspace or access network.
+                          Requires Windows Subsystem for Linux (WSL). Show
+                          installation instructions...
+  3. Full Access          Codex can read files, make edits, and run commands
+                          with network access, without approval. Exercise
+                          caution.
+
+  Press enter to confirm or esc to go back
diff --git a/codex-rs/tui/src/chatwidget/tests.rs b/codex-rs/tui/src/chatwidget/tests.rs
index a9c1d58c67..526643627b 100644
--- a/codex-rs/tui/src/chatwidget/tests.rs
+++ b/codex-rs/tui/src/chatwidget/tests.rs
@@ -1276,9 +1276,36 @@ fn approvals_selection_popup_snapshot() {
     chat.open_approvals_popup();
 
     let popup = render_bottom_popup(&chat, 80);
+    #[cfg(target_os = "windows")]
+    insta::with_settings!({ snapshot_suffix => "windows" }, {
+        assert_snapshot!("approvals_selection_popup", popup);
+    });
+    #[cfg(not(target_os = "windows"))]
     assert_snapshot!("approvals_selection_popup", popup);
 }
 
+#[test]
+fn approvals_popup_includes_wsl_note_for_auto_mode() {
+    let (mut chat, _rx, _op_rx) = make_chatwidget_manual();
+
+    if cfg!(target_os = "windows") {
+        chat.config.forced_auto_mode_downgraded_on_windows = true;
+    }
+    chat.open_approvals_popup();
+
+    let popup = render_bottom_popup(&chat, 80);
+    assert_eq!(
+        popup.contains("Requires Windows Subsystem for Linux (WSL)"),
+        cfg!(target_os = "windows"),
+        "expected auto preset description to mention WSL requirement only on Windows, popup: {popup}"
+    );
+    assert_eq!(
+        popup.contains("Codex forced your settings back to Read Only on this Windows machine."),
+        cfg!(target_os = "windows") && chat.config.forced_auto_mode_downgraded_on_windows,
+        "expected downgrade notice only when auto mode is forced off on Windows, popup: {popup}"
+    );
+}
+
 #[test]
 fn full_access_confirmation_popup_snapshot() {
     let (mut chat, _rx, _op_rx) = make_chatwidget_manual();
@@ -1293,6 +1320,20 @@ fn full_access_confirmation_popup_snapshot() {
     assert_snapshot!("full_access_confirmation_popup", popup);
 }
 
+#[cfg(target_os = "windows")]
+#[test]
+fn windows_auto_mode_instructions_popup_lists_install_steps() {
+    let (mut chat, _rx, _op_rx) = make_chatwidget_manual();
+
+    chat.open_windows_auto_mode_instructions();
+
+    let popup = render_bottom_popup(&chat, 120);
+    assert!(
+        popup.contains("wsl --install"),
+        "expected WSL instructions popup to include install command, popup: {popup}"
+    );
+}
+
 #[test]
 fn model_reasoning_selection_popup_snapshot() {
     let (mut chat, _rx, _op_rx) = make_chatwidget_manual();
diff --git a/codex-rs/tui/src/history_cell.rs b/codex-rs/tui/src/history_cell.rs
index e630e00175..406a892561 100644
--- a/codex-rs/tui/src/history_cell.rs
+++ b/codex-rs/tui/src/history_cell.rs
@@ -465,7 +465,7 @@ struct CompletedMcpToolCallWithImageOutput {
 }
 impl HistoryCell for CompletedMcpToolCallWithImageOutput {
     fn display_lines(&self, _width: u16) -> Vec<Line<'static>> {
-        vec!["tool result (image output omitted)".into()]
+        vec!["tool result (image output)".into()]
     }
 }
 
diff --git a/codex-rs/tui/src/lib.rs b/codex-rs/tui/src/lib.rs
index ea19d51ad6..028bf68e87 100644
--- a/codex-rs/tui/src/lib.rs
+++ b/codex-rs/tui/src/lib.rs
@@ -315,7 +315,11 @@ async fn run_ratatui_app(
     // Initialize high-fidelity session event logging if enabled.
     session_log::maybe_init(&initial_config);
 
-    let auth_manager = AuthManager::shared(initial_config.codex_home.clone(), false);
+    let auth_manager = AuthManager::shared(
+        initial_config.codex_home.clone(),
+        false,
+        initial_config.cli_auth_credentials_store_mode,
+    );
     let login_status = get_login_status(&initial_config);
     let should_show_trust_screen = should_show_trust_screen(&initial_config);
     let should_show_windows_wsl_screen =
@@ -476,7 +480,7 @@ fn get_login_status(config: &Config) -> LoginStatus {
         // Reading the OpenAI API key is an async operation because it may need
         // to refresh the token. Block on it.
         let codex_home = config.codex_home.clone();
-        match CodexAuth::from_auth_storage(&codex_home) {
+        match CodexAuth::from_auth_storage(&codex_home, config.cli_auth_credentials_store_mode) {
             Ok(Some(auth)) => LoginStatus::AuthMode(auth.mode),
             Ok(None) => LoginStatus::NotAuthenticated,
             Err(err) => {
@@ -507,14 +511,16 @@ async fn load_config_or_exit(
 /// or if the current cwd project is already trusted. If not, we need to
 /// show the trust screen.
 fn should_show_trust_screen(config: &Config) -> bool {
-    if config.did_user_set_custom_approval_policy_or_sandbox_mode {
-        // if the user has overridden either approval policy or sandbox mode,
-        // skip the trust flow
-        false
-    } else {
-        // otherwise, skip iff the active project is trusted
-        !config.active_project.is_trusted()
+    if cfg!(target_os = "windows") {
+        // Native Windows cannot enforce sandboxed write access without WSL; skip the trust prompt entirely.
+        return false;
     }
+    if config.did_user_set_custom_approval_policy_or_sandbox_mode {
+        // Respect explicit approval/sandbox overrides made by the user.
+        return false;
+    }
+    // otherwise, skip iff the active project is trusted
+    !config.active_project.is_trusted()
 }
 
 fn should_show_onboarding(
@@ -543,3 +549,38 @@ fn should_show_login_screen(login_status: LoginStatus, config: &Config) -> bool
 
     login_status == LoginStatus::NotAuthenticated
 }
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+    use codex_core::config::ConfigOverrides;
+    use codex_core::config::ConfigToml;
+    use codex_core::config::ProjectConfig;
+    use tempfile::TempDir;
+
+    #[test]
+    fn windows_skips_trust_prompt() -> std::io::Result<()> {
+        let temp_dir = TempDir::new()?;
+        let mut config = Config::load_from_base_config_with_overrides(
+            ConfigToml::default(),
+            ConfigOverrides::default(),
+            temp_dir.path().to_path_buf(),
+        )?;
+        config.did_user_set_custom_approval_policy_or_sandbox_mode = false;
+        config.active_project = ProjectConfig { trust_level: None };
+
+        let should_show = should_show_trust_screen(&config);
+        if cfg!(target_os = "windows") {
+            assert!(
+                !should_show,
+                "Windows trust prompt should always be skipped on native Windows"
+            );
+        } else {
+            assert!(
+                should_show,
+                "Non-Windows should still show trust prompt when project is untrusted"
+            );
+        }
+        Ok(())
+    }
+}
diff --git a/codex-rs/tui/src/onboarding/auth.rs b/codex-rs/tui/src/onboarding/auth.rs
index 05687bb1bd..56527ac8ef 100644
--- a/codex-rs/tui/src/onboarding/auth.rs
+++ b/codex-rs/tui/src/onboarding/auth.rs
@@ -1,6 +1,7 @@
 #![allow(clippy::unwrap_used)]
 
 use codex_core::AuthManager;
+use codex_core::auth::AuthCredentialsStoreMode;
 use codex_core::auth::CLIENT_ID;
 use codex_core::auth::login_with_api_key;
 use codex_core::auth::read_openai_api_key_from_env;
@@ -148,6 +149,7 @@ pub(crate) struct AuthModeWidget {
     pub error: Option<String>,
     pub sign_in_state: Arc<RwLock<SignInState>>,
     pub codex_home: PathBuf,
+    pub cli_auth_credentials_store_mode: AuthCredentialsStoreMode,
     pub login_status: LoginStatus,
     pub auth_manager: Arc<AuthManager>,
     pub forced_chatgpt_workspace_id: Option<String>,
@@ -512,7 +514,11 @@ impl AuthModeWidget {
             self.disallow_api_login();
             return;
         }
-        match login_with_api_key(&self.codex_home, &api_key) {
+        match login_with_api_key(
+            &self.codex_home,
+            &api_key,
+            self.cli_auth_credentials_store_mode,
+        ) {
             Ok(()) => {
                 self.error = None;
                 self.login_status = LoginStatus::AuthMode(AuthMode::ApiKey);
@@ -553,6 +559,7 @@ impl AuthModeWidget {
             self.codex_home.clone(),
             CLIENT_ID.to_string(),
             self.forced_chatgpt_workspace_id.clone(),
+            self.cli_auth_credentials_store_mode,
         );
         match run_login_server(opts) {
             Ok(child) => {
@@ -640,6 +647,8 @@ mod tests {
     use pretty_assertions::assert_eq;
     use tempfile::TempDir;
 
+    use codex_core::auth::AuthCredentialsStoreMode;
+
     fn widget_forced_chatgpt() -> (AuthModeWidget, TempDir) {
         let codex_home = TempDir::new().unwrap();
         let codex_home_path = codex_home.path().to_path_buf();
@@ -649,8 +658,13 @@ mod tests {
             error: None,
             sign_in_state: Arc::new(RwLock::new(SignInState::PickMode)),
             codex_home: codex_home_path.clone(),
+            cli_auth_credentials_store_mode: AuthCredentialsStoreMode::File,
             login_status: LoginStatus::NotAuthenticated,
-            auth_manager: AuthManager::shared(codex_home_path, false),
+            auth_manager: AuthManager::shared(
+                codex_home_path,
+                false,
+                AuthCredentialsStoreMode::File,
+            ),
             forced_chatgpt_workspace_id: None,
             forced_login_method: Some(ForcedLoginMethod::Chatgpt),
         };
diff --git a/codex-rs/tui/src/onboarding/onboarding_screen.rs b/codex-rs/tui/src/onboarding/onboarding_screen.rs
index 369e6ddbb0..709f158b80 100644
--- a/codex-rs/tui/src/onboarding/onboarding_screen.rs
+++ b/codex-rs/tui/src/onboarding/onboarding_screen.rs
@@ -87,6 +87,7 @@ impl OnboardingScreen {
         let forced_chatgpt_workspace_id = config.forced_chatgpt_workspace_id.clone();
         let forced_login_method = config.forced_login_method;
         let codex_home = config.codex_home;
+        let cli_auth_credentials_store_mode = config.cli_auth_credentials_store_mode;
         let mut steps: Vec<Step> = Vec::new();
         if show_windows_wsl_screen {
             steps.push(Step::Windows(WindowsSetupWidget::new(codex_home.clone())));
@@ -106,6 +107,7 @@ impl OnboardingScreen {
                 error: None,
                 sign_in_state: Arc::new(RwLock::new(SignInState::PickMode)),
                 codex_home: codex_home.clone(),
+                cli_auth_credentials_store_mode,
                 login_status,
                 auth_manager,
                 forced_chatgpt_workspace_id,
diff --git a/codex-rs/tui/src/status/helpers.rs b/codex-rs/tui/src/status/helpers.rs
index e039f0e898..fdbaabd38c 100644
--- a/codex-rs/tui/src/status/helpers.rs
+++ b/codex-rs/tui/src/status/helpers.rs
@@ -83,7 +83,8 @@ pub(crate) fn compose_agents_summary(config: &Config) -> String {
 }
 
 pub(crate) fn compose_account_display(config: &Config) -> Option<StatusAccountDisplay> {
-    let auth = load_auth_dot_json(&config.codex_home).ok()??;
+    let auth =
+        load_auth_dot_json(&config.codex_home, config.cli_auth_credentials_store_mode).ok()??;
 
     if let Some(tokens) = auth.tokens.as_ref() {
         let info = &tokens.id_token;
diff --git a/docs/config.md b/docs/config.md
index 9223f48d26..0d81a64efc 100644
--- a/docs/config.md
+++ b/docs/config.md
@@ -836,7 +836,9 @@ notifications = [ "agent-turn-complete", "approval-requested" ]
 
 > [!NOTE] > `tui.notifications` is built‑in and limited to the TUI session. For programmatic or cross‑environment notifications—or to integrate with OS‑specific notifiers—use the top‑level `notify` option to run an external program that receives event JSON. The two settings are independent and can be used together.
 
-## Forcing a login method
+## Authentication and authorization
+
+### Forcing a login method
 
 To force users on a given machine to use a specific login method or workspace, use a combination of [managed configurations](https://developers.openai.com/codex/security#managed-configuration) as well as either or both of the following fields:
 
@@ -852,6 +854,22 @@ If the active credentials don't match the config, the user will be logged out an
 
 If `forced_chatgpt_workspace_id` is set but `forced_login_method` is not set, API key login will still work.
 
+### Control where login credentials are stored
+
+```toml
+cli_auth_credentials_store = "keyring"
+```
+
+Valid values:
+
+- `file` (default) – Store credentials in `auth.json` under `$CODEX_HOME`.
+- `keyring` – Store credentials in the operating system keyring via the [`keyring` crate](https://crates.io/crates/keyring); the CLI reports an error if secure storage is unavailable. Backends by OS:
+  - macOS: macOS Keychain
+  - Windows: Windows Credential Manager
+  - Linux: DBus‑based Secret Service, the kernel keyutils, or a combination
+  - FreeBSD/OpenBSD: DBus‑based Secret Service
+- `auto` – Save credentials to the operating system keyring when available; otherwise, fall back to `auth.json` under `$CODEX_HOME`.
+
 ## Config reference
 
 | Key                                              | Type / Values                                                     | Notes                                                                                                                      |
@@ -910,4 +928,5 @@ If `forced_chatgpt_workspace_id` is set but `forced_login_method` is not set, AP
 | `tools.web_search`                               | boolean                                                           | Enable web search tool (alias: `web_search_request`) (default: false).                                                     |
 | `forced_login_method`                            | `chatgpt` \| `api`                                                | Only allow Codex to be used with ChatGPT or API keys.                                                                      |
 | `forced_chatgpt_workspace_id`                    | string (uuid)                                                     | Only allow Codex to be used with the specified ChatGPT workspace.                                                          |
+| `cli_auth_credentials_store`                     | `file` \| `keyring` \| `auto`                                     | Where to store CLI login credentials (default: `file`).                                                                    |
 | `tools.view_image`                               | boolean                                                           | Enable the `view_image` tool so Codex can attach local image files from the workspace (default: false).                    |