## Summary
- add the guardian reviewer flow for `on-request` approvals in command,
patch, sandbox-retry, and managed-network approval paths
- keep guardian behind `features.guardian_approval` instead of exposing
a public `approval_policy = guardian` mode
- route ordinary `OnRequest` approvals to the guardian subagent when the
feature is enabled, without changing the public approval-mode surface
## Public model
- public approval modes stay unchanged
- guardian is enabled via `features.guardian_approval`
- when that feature is on, `approval_policy = on-request` keeps the same
approval boundaries but sends those approval requests to the guardian
reviewer instead of the user
- `/experimental` only persists the feature flag; it does not rewrite
`approval_policy`
- CLI and app-server no longer expose a separate `guardian` approval
mode in this PR
## Guardian reviewer
- the reviewer runs as a normal subagent and reuses the existing
subagent/thread machinery
- it is locked to a read-only sandbox and `approval_policy = never`
- it does not inherit user/project exec-policy rules
- it prefers `gpt-5.4` when the current provider exposes it, otherwise
falls back to the parent turn's active model
- it fail-closes on timeout, startup failure, malformed output, or any
other review error
- it currently auto-approves only when `risk_score < 80`
## Review context and policy
- guardian mirrors `OnRequest` approval semantics rather than
introducing a separate approval policy
- explicit `require_escalated` requests follow the same approval surface
as `OnRequest`; the difference is only who reviews them
- managed-network allowlist misses that enter the approval flow are also
reviewed by guardian
- the review prompt includes bounded recent transcript history plus
recent tool call/result evidence
- transcript entries and planned-action strings are truncated with
explicit `<guardian_truncated ... />` markers so large payloads stay
bounded
- apply-patch reviews include the full patch content (without
duplicating the structured `changes` payload)
- the guardian request layout is snapshot-tested using the same
model-visible Responses request formatter used elsewhere in core
## Guardian network behavior
- the guardian subagent inherits the parent session's managed-network
allowlist when one exists, so it can use the same approved network
surface while reviewing
- exact session-scoped network approvals are copied into the guardian
session with protocol/port scope preserved
- those copied approvals are now seeded before the guardian's first turn
is submitted, so inherited approvals are available during any immediate
review-time checks
## Out of scope / follow-ups
- the sandbox-permission validation split was pulled into a separate PR
and is not part of this diff
- a future follow-up can enable `serde_json` preserve-order in
`codex-core` and then simplify the guardian action rendering further
---------
Co-authored-by: Codex <noreply@openai.com>
## Summary
This is a purely mechanical refactor of `OtelManager` ->
`SessionTelemetry` to better convey what the struct is doing. No
behavior change.
## Why
`OtelManager` ended up sounding much broader than what this type
actually does. It doesn't manage OTEL globally; it's the session-scoped
telemetry surface for emitting log/trace events and recording metrics
with consistent session metadata (`app_version`, `model`, `slug`,
`originator`, etc.).
`SessionTelemetry` is a more accurate name, and updating the call sites
makes that boundary a lot easier to follow.
## Validation
- `just fmt`
- `cargo test -p codex-otel`
- `cargo test -p codex-core`
This is PR 3 of the app-server tracing rollout.
PRs https://github.com/openai/codex/pull/13285 and
https://github.com/openai/codex/pull/13368 gave us inbound request spans
in app-server and propagated trace context through Submission. This
change finishes the next piece in core: when a request actually starts a
turn, we now create a core-owned long-lived span that stays open for the
real lifetime of the turn.
What changed:
- `Session::spawn_task` can now optionally create a long-lived turn span
and run the spawned task inside it
- `turn/start` uses that path, so normal turn execution stays under a
single core-owned span after the async handoff
- `review/start` uses the same pattern
- added a unit test that verifies the spawned turn task inherits the
submission dispatch trace ancestry
**Why**
The app-server request span is intentionally short-lived. Once work
crosses into core, we still want one span that covers the actual
execution window until completion or interruption. This keeps that
ownership where it belongs: in the layer that owns the runtime
lifecycle.
## Why
Enterprises can already constrain approvals, sandboxing, and web search
through `requirements.toml` and MDM, but feature flags were still only
configurable as managed defaults. That meant an enterprise could suggest
feature values, but it could not actually pin them.
This change closes that gap and makes enterprise feature requirements
behave like the other constrained settings. The effective feature set
now stays consistent with enterprise requirements during config load,
when config writes are validated, and when runtime code mutates feature
flags later in the session.
It also tightens the runtime API for managed features. `ManagedFeatures`
now follows the same constraint-oriented shape as `Constrained<T>`
instead of exposing panic-prone mutation helpers, and production code
can no longer construct it through an unconstrained `From<Features>`
path.
The PR also hardens the `compact_resume_fork` integration coverage on
Windows. After the feature-management changes,
`compact_resume_after_second_compaction_preserves_history` was
overflowing the libtest/Tokio thread stacks on Windows, so the test now
uses an explicit larger-stack harness as a pragmatic mitigation. That
may not be the ideal root-cause fix, and it merits a parallel
investigation into whether part of the async future chain should be
boxed to reduce stack pressure instead.
## What Changed
Enterprises can now pin feature values in `requirements.toml` with the
requirements-side `features` table:
```toml
[features]
personality = true
unified_exec = false
```
Only canonical feature keys are allowed in the requirements `features`
table; omitted keys remain unconstrained.
- Added a requirements-side pinned feature map to
`ConfigRequirementsToml`, threaded it through source-preserving
requirements merge and normalization in `codex-config`, and made the
TOML surface use `[features]` (while still accepting legacy
`[feature_requirements]` for compatibility).
- Exposed `featureRequirements` from `configRequirements/read`,
regenerated the JSON/TypeScript schema artifacts, and updated the
app-server README.
- Wrapped the effective feature set in `ManagedFeatures`, backed by
`ConstrainedWithSource<Features>`, and changed its API to mirror
`Constrained<T>`: `can_set(...)`, `set(...) -> ConstraintResult<()>`,
and result-returning `enable` / `disable` / `set_enabled` helpers.
- Removed the legacy-usage and bulk-map passthroughs from
`ManagedFeatures`; callers that need those behaviors now mutate a plain
`Features` value and reapply it through `set(...)`, so the constrained
wrapper remains the enforcement boundary.
- Removed the production loophole for constructing unconstrained
`ManagedFeatures`. Non-test code now creates it through the configured
feature-loading path, and `impl From<Features> for ManagedFeatures` is
restricted to `#[cfg(test)]`.
- Rejected legacy feature aliases in enterprise feature requirements,
and return a load error when a pinned combination cannot survive
dependency normalization.
- Validated config writes against enterprise feature requirements before
persisting changes, including explicit conflicting writes and
profile-specific feature states that normalize into invalid
combinations.
- Updated runtime and TUI feature-toggle paths to use the constrained
setter API and to persist or apply the effective post-constraint value
rather than the requested value.
- Updated the `core_test_support` Bazel target to include the bundled
core model-catalog fixtures in its runtime data, so helper code that
resolves `core/models.json` through runfiles works in remote Bazel test
environments.
- Renamed the core config test coverage to emphasize that effective
feature values are normalized at runtime, while conflicting persisted
config writes are rejected.
- Ran `compact_resume_after_second_compaction_preserves_history` inside
an explicit 8 MiB test thread and Tokio runtime worker stack, following
the existing larger-stack integration-test pattern, to keep the Windows
`compact_resume_fork` test slice from aborting while a parallel
investigation continues into whether some of the underlying async
futures should be boxed.
## Verification
- `cargo test -p codex-config`
- `cargo test -p codex-core feature_requirements_ -- --nocapture`
- `cargo test -p codex-core
load_requirements_toml_produces_expected_constraints -- --nocapture`
- `cargo test -p codex-core
compact_resume_after_second_compaction_preserves_history -- --nocapture`
- `cargo test -p codex-core compact_resume_fork -- --nocapture`
- Re-ran the built `codex-core` `tests/all` binary with
`RUST_MIN_STACK=262144` for
`compact_resume_after_second_compaction_preserves_history` to confirm
the explicit-stack harness fixes the deterministic low-stack repro.
- `cargo test -p codex-core`
- This still fails locally in unrelated integration areas that expect
the `codex` / `test_stdio_server` binaries or hit existing `search_tool`
wiremock mismatches.
## Docs
`developers.openai.com/codex` should document the requirements-side
`[features]` table for enterprise and MDM-managed configuration,
including that it only accepts canonical feature keys and that
conflicting config writes are rejected.
Summary
- prevent delegated review agents from re-enabling blocked tools by
explicitly disabling the Collab feature alongside web search and view
image controls
Testing
- Not run (not requested)
## Why
We currently carry multiple permission-related concepts directly on
`Config` for shell/unified-exec behavior (`approval_policy`,
`sandbox_policy`, `network`, `shell_environment_policy`,
`windows_sandbox_mode`).
Consolidating these into one in-memory struct makes permission handling
easier to reason about and sets up the next step: supporting named
permission profiles (`[permissions.PROFILE_NAME]`) without changing
behavior now.
This change is mostly mechanical: it updates existing callsites to go
through `config.permissions`, but it does not yet refactor those
callsites to take a single `Permissions` value in places where multiple
permission fields are still threaded separately.
This PR intentionally **does not** change the on-disk `config.toml`
format yet and keeps compatibility with legacy config keys.
## What Changed
- Introduced `Permissions` in `core/src/config/mod.rs`.
- Added `Config::permissions` and moved effective runtime permission
fields under it:
- `approval_policy`
- `sandbox_policy`
- `network`
- `shell_environment_policy`
- `windows_sandbox_mode`
- Updated config loading/building so these effective values are still
derived from the same existing config inputs and constraints.
- Updated Windows sandbox helpers/resolution to read/write via
`permissions`.
- Threaded the new field through all permission consumers across core
runtime, app-server, CLI/exec, TUI, and sandbox summary code.
- Updated affected tests to reference `config.permissions.*`.
- Renamed the struct/field from
`EffectivePermissions`/`effective_permissions` to
`Permissions`/`permissions` and aligned variable naming accordingly.
## Verification
- `just fix -p codex-core -p codex-tui -p codex-cli -p codex-app-server
-p codex-exec -p codex-utils-sandbox-summary`
- `cargo build -p codex-core -p codex-tui -p codex-cli -p
codex-app-server -p codex-exec -p codex-utils-sandbox-summary`
- Defer rollout persistence for fresh threads (`InitialHistory::New`):
keep rollout events in memory and only materialize rollout file + state
DB row on first `EventMsg::UserMessage`.
- Keep precomputed rollout path available before materialization.
- Change `thread/start` to build thread response from live config
snapshot and optional precomputed path.
- Improve pre-materialization behavior in app-server/TUI: clearer
invalid-request errors for file-backed ops and a friendlier `/fork` “not
ready yet” UX.
- Update tests to match deferred semantics across
start/read/archive/unarchive/fork/resume/review flows.
- Improved resilience of user_shell test, which should be unrelated to
this change but must be affected by timing changes
For Reviewers:
* The primary change is in recorder.rs
* Most of the other changes were to fix up broken assumptions in
existing tests
Testing:
* Manually tested CLI
* Exercised app server paths by manually running IDE Extension with
rebuilt CLI binary
* Only user-visible change is that `/fork` in TUI generates visible
error if used prior to first turn
This PR makes it possible to disable live web search via an enterprise
config even if the user is running in `--yolo` mode (though cached web
search will still be available). To do this, create
`/etc/codex/requirements.toml` as follows:
```toml
# "live" is not allowed; "disabled" is allowed even though not listed explicitly.
allowed_web_search_modes = ["cached"]
```
Or set `requirements_toml_base64` MDM as explained on
https://developers.openai.com/codex/security/#locations.
### Why
- Enforce admin/MDM/`requirements.toml` constraints on web-search
behavior, independent of user config and per-turn sandbox defaults.
- Ensure per-turn config resolution and review-mode overrides never
crash when constraints are present.
### What
- Add `allowed_web_search_modes` to requirements parsing and surface it
in app-server v2 `ConfigRequirements` (`allowedWebSearchModes`), with
fixtures updated.
- Define a requirements allowlist type (`WebSearchModeRequirement`) and
normalize semantics:
- `disabled` is always implicitly allowed (even if not listed).
- An empty list is treated as `["disabled"]`.
- Make `Config.web_search_mode` a `Constrained<WebSearchMode>` and apply
requirements via `ConstrainedWithSource<WebSearchMode>`.
- Update per-turn resolution (`resolve_web_search_mode_for_turn`) to:
- Prefer `Live → Cached → Disabled` when
`SandboxPolicy::DangerFullAccess` is active (subject to requirements),
unless the user preference is explicitly `Disabled`.
- Otherwise, honor the user’s preferred mode, falling back to an allowed
mode when necessary.
- Update TUI `/debug-config` and app-server mapping to display
normalized `allowed_web_search_modes` (including implicit `disabled`).
- Fix web-search integration tests to assert cached behavior under
`SandboxPolicy::ReadOnly` (since `DangerFullAccess` legitimately prefers
`live` when allowed).
### What
add wiring for `phase` field on `ResponseItem::Message` to lay
groundwork for differentiating model preambles and final messages.
currently optional.
follows pattern in #9698.
updated schemas with `just write-app-server-schema` so we can see type
changes.
### Tests
Updated existing tests for SSE parsing and hydrating from history
web_search can now be updated per-turn, for things like changes to
sandbox policy.
`SandboxPolicy::DangerFullAccess` now sets web_search to `live`, and the
default is still `cached`.
Added integration tests.
Users of Azure endpoints are reporting that when they use `/review`,
they sometimes see an error "Invalid 'input[3].id". I suspect this is
specific to the Azure implementation of the `responses` API. The Azure
team generally copies the OpenAI code for this endpoint, but they do
have minor differences and sometimes lag in rolling out bug fixes or
updates.
The error appears to be triggered because the `/review` implementation
is using a user ID with a colon in it.
Addresses #9360
moving `web_search` rollout serverside, so need a way to explicitly
disable search + signal eligibility from the client.
- Add `x‑oai‑web‑search‑eligible` header that signifies whether the
request can have web search.
- Only attach the `web_search` tool when the resolved `WebSearchMode` is
`Live` or `Cached`.
### What
Add `WebSearchMode` enum (disabled, cached live, defaults to cached) to
config + V2 protocol. This enum takes precedence over legacy flags:
`web_search_cached`, `web_search_request`, and `tools.web_search`.
Keep `--search` as live.
### Tests
Added tests
Instead of having a hard-coded default review model, use the current
model for running `/review` unless one is specified in the config.
Also inherit current reasoning effort
Include project-level AGENTS.md and skills in /review sessions so the
review sub-agent uses the same instruction pipeline as standard runs,
keeping reviewer context aligned with normal sessions.
Add metrics capabilities to Codex. The `README.md` is up to date.
This will not be merged with the metrics before this PR of course:
https://github.com/openai/codex/pull/8350
This reverts commit 291b54a762.
This commit was intended to prevent the model from making code changes
during `/review`, which is sometimes does. Unfortunately, it has other
unintended side effects that cause `/review` to fail in a variety of
ways. See #8115 and #7815. We've therefore decided to revert this
change.
- Make Config.model optional and centralize default-selection logic in
ModelsManager, including a default_model helper (with
codex-auto-balanced when available) so sessions now carry an explicit
chosen model separate from the base config.
- Resolve `model` once in `core` and `tui` from config. Then store the
state of it on other structs.
- Move refreshing models to be before resolving the default model
This PR moves `ModelsFamily` to `openai_models`. It also propagates
`ModelsManager` to session services and use it to drive model family. We
also make `derive_default_model_family` private because it's a step
towards what we want: one place that gives model configuration.
This is a second step at having one source of truth for models
information and config: `ModelsManager`.
Next steps would be to remove `ModelsFamily` from config. That's massive
because it's being used in 41 occasions mostly pre launching `codex`.
Also, we need to make `find_family_for_model` private. It's also big
because it's being used in 21 occasions ~ all tests.
Adds AgentMessageContentDelta, ReasoningContentDelta,
ReasoningRawContentDelta item streaming events while maintaining
compatibility for old events.
---------
Co-authored-by: Owen Lin <owen@openai.com>
In this PR, I am exploring migrating task kind to an invocation of
Codex. The main reason would be getting rid off multiple
`ConversationHistory` state and streamlining our context/history
management.
This approach depends on opening a channel between the sub-codex and
codex. This channel is responsible for forwarding `interactive`
(`approvals`) and `non-interactive` events. The `task` is responsible
for handling those events.
This opens the door for implementing `codex as a tool`, replacing
`compact` and `review`, and potentially subagents.
One consideration is this code is very similar to `app-server` specially
in the approval part. If in the future we wanted an interactive
`sub-codex` we should consider using `codex-mcp`
Today `sub_id` is an ID of a single incoming Codex Op submition. We then
associate all events triggered by this operation using the same
`sub_id`.
At the same time we are also creating a TurnContext per submission and
we'd like to start associating some events (item added/item completed)
with an entire turn instead of just the operation that started it.
Using turn context when sending events give us flexibility to change
notification scheme.
Adds a new ItemStarted event and delivers UserMessage as the first item
type (more to come).
Renames `InputItem` to `UserInput` considering we're using the `Item`
suffix for actual items.
Extracting tasks in a module and start abstraction behind a Trait (more
to come on this but each task will be tackled in a dedicated PR)
The goal was to drop the ActiveTask and to have a (potentially) set of
tasks during each turn
Michael Bolin