diff --git a/codex-rs/Cargo.lock b/codex-rs/Cargo.lock
index 8bdb36c9d2..96316db088 100644
--- a/codex-rs/Cargo.lock
+++ b/codex-rs/Cargo.lock
@@ -3003,6 +3003,7 @@ dependencies = [
  "codex-utils-cache",
  "codex-utils-cargo-bin",
  "codex-utils-git-discovery",
+ "codex-utils-head-tail-buffer",
  "codex-utils-home-dir",
  "codex-utils-image",
  "codex-utils-output-truncation",
@@ -5184,6 +5185,13 @@ dependencies = [
  "tracing",
 ]
 
+[[package]]
+name = "codex-utils-head-tail-buffer"
+version = "0.0.0"
+dependencies = [
+ "pretty_assertions",
+]
+
 [[package]]
 name = "codex-utils-home-dir"
 version = "0.0.0"
diff --git a/codex-rs/Cargo.toml b/codex-rs/Cargo.toml
index 110914fa16..d7bd76b11c 100644
--- a/codex-rs/Cargo.toml
+++ b/codex-rs/Cargo.toml
@@ -130,6 +130,7 @@ members = [
     "utils/cache",
     "utils/critical-path",
     "utils/git-discovery",
+    "utils/head-tail-buffer",
     "utils/image",
     "utils/json-to-toml",
     "utils/home-dir",
@@ -305,6 +306,7 @@ codex-utils-critical-path = { path = "utils/critical-path" }
 codex-utils-elapsed = { path = "utils/elapsed" }
 codex-utils-fuzzy-match = { path = "utils/fuzzy-match" }
 codex-utils-git-discovery = { path = "utils/git-discovery" }
+codex-utils-head-tail-buffer = { path = "utils/head-tail-buffer" }
 codex-utils-home-dir = { path = "utils/home-dir" }
 codex-utils-image = { path = "utils/image" }
 codex-utils-json-to-toml = { path = "utils/json-to-toml" }
diff --git a/codex-rs/core/Cargo.toml b/codex-rs/core/Cargo.toml
index 245f417b18..8fc0efd30d 100644
--- a/codex-rs/core/Cargo.toml
+++ b/codex-rs/core/Cargo.toml
@@ -78,6 +78,7 @@ codex-utils-absolute-path = { workspace = true }
 codex-utils-audio = { workspace = true }
 codex-utils-cache = { workspace = true }
 codex-utils-git-discovery = { workspace = true }
+codex-utils-head-tail-buffer = { workspace = true }
 codex-utils-image = { workspace = true }
 codex-utils-home-dir = { workspace = true }
 codex-utils-output-truncation = { workspace = true }
diff --git a/codex-rs/core/src/tools/context.rs b/codex-rs/core/src/tools/context.rs
index c80a89ddd1..6a58a9c373 100644
--- a/codex-rs/core/src/tools/context.rs
+++ b/codex-rs/core/src/tools/context.rs
@@ -4,7 +4,6 @@ use crate::session::step_context::StepContext;
 use crate::session::turn_context::TurnContext;
 use crate::session::turn_context::TurnEnvironment;
 use crate::turn_diff_tracker::TurnDiffTracker;
-use crate::unified_exec::format_output_omission_marker;
 use crate::unified_exec::resolve_max_tokens;
 use codex_protocol::ResponseItemId;
 use codex_protocol::mcp::CallToolResult;
@@ -17,6 +16,7 @@ use codex_protocol::models::function_call_output_content_items_to_text;
 use codex_tools::LoadableToolSpec;
 use codex_tools::ToolName;
 use codex_utils_audio::estimate_audio_token_count;
+use codex_utils_head_tail_buffer::format_output_omission_marker;
 use codex_utils_output_truncation::TruncationPolicy;
 use codex_utils_output_truncation::approx_token_count;
 use codex_utils_output_truncation::formatted_truncate_text;
diff --git a/codex-rs/core/src/unified_exec/mod.rs b/codex-rs/core/src/unified_exec/mod.rs
index 4c47abb2de..a7a8899c5e 100644
--- a/codex-rs/core/src/unified_exec/mod.rs
+++ b/codex-rs/core/src/unified_exec/mod.rs
@@ -48,7 +48,6 @@ use codex_core_plugins::PluginMetricsSidecar;
 
 mod async_watcher;
 mod errors;
-mod head_tail_buffer;
 mod oneshot;
 mod process;
 mod process_manager;
@@ -232,10 +231,6 @@ pub(crate) fn resolve_max_tokens(max_tokens: Option<usize>) -> usize {
     max_tokens.unwrap_or(DEFAULT_MAX_OUTPUT_TOKENS)
 }
 
-pub(crate) fn format_output_omission_marker(omitted_bytes: usize) -> String {
-    format!("... {omitted_bytes} bytes omitted ...")
-}
-
 pub(crate) fn generate_chunk_id() -> String {
     let mut rng = rng();
     (0..6)
diff --git a/codex-rs/core/src/unified_exec/process.rs b/codex-rs/core/src/unified_exec/process.rs
index b3ddb02952..ed223ac7b4 100644
--- a/codex-rs/core/src/unified_exec/process.rs
+++ b/codex-rs/core/src/unified_exec/process.rs
@@ -24,6 +24,7 @@ use codex_protocol::protocol::TruncationPolicy;
 use codex_sandboxing::SandboxType;
 use codex_sandboxing::is_likely_sandbox_denied;
 use codex_sandboxing::record_filesystem_sandbox_violation;
+use codex_utils_head_tail_buffer::HeadTailBuffer;
 use codex_utils_output_truncation::formatted_truncate_text;
 use codex_utils_pty::ExecCommandSession;
 use codex_utils_pty::ProcessSignal as PtyProcessSignal;
@@ -32,7 +33,6 @@ use codex_utils_pty::SpawnedPty;
 use super::UNIFIED_EXEC_OUTPUT_MAX_BYTES;
 use super::UNIFIED_EXEC_OUTPUT_MAX_TOKENS;
 use super::UnifiedExecError;
-use super::head_tail_buffer::HeadTailBuffer;
 use super::process_state::ProcessState;
 use crate::shell_snapshot::ShellSnapshotFile;
 
diff --git a/codex-rs/core/src/unified_exec/process_manager.rs b/codex-rs/core/src/unified_exec/process_manager.rs
index 0ba92cb486..b07b410be6 100644
--- a/codex-rs/core/src/unified_exec/process_manager.rs
+++ b/codex-rs/core/src/unified_exec/process_manager.rs
@@ -64,7 +64,6 @@ use crate::unified_exec::async_watcher::spawn_exit_watcher;
 use crate::unified_exec::async_watcher::start_streaming_output;
 use crate::unified_exec::clamp_yield_time;
 use crate::unified_exec::generate_chunk_id;
-use crate::unified_exec::head_tail_buffer::HeadTailBuffer;
 use crate::unified_exec::process::OutputBuffers;
 use crate::unified_exec::process::OutputHandles;
 use crate::unified_exec::process::UnifiedExecProcess;
@@ -89,6 +88,7 @@ use codex_protocol::shell_environment::is_non_inheritable_env_var;
 use codex_sandboxing::SandboxCommand;
 use codex_shell_command::is_dangerous_command::DangerousCommandPlatform;
 use codex_tools::ToolName;
+use codex_utils_head_tail_buffer::HeadTailBuffer;
 use codex_utils_output_truncation::approx_tokens_from_byte_count;
 use codex_utils_path_uri::PathUri;
 
diff --git a/codex-rs/utils/head-tail-buffer/BUILD.bazel b/codex-rs/utils/head-tail-buffer/BUILD.bazel
new file mode 100644
index 0000000000..0be34d31dc
--- /dev/null
+++ b/codex-rs/utils/head-tail-buffer/BUILD.bazel
@@ -0,0 +1,6 @@
+load("//:defs.bzl", "codex_rust_crate")
+
+codex_rust_crate(
+    name = "head-tail-buffer",
+    crate_name = "codex_utils_head_tail_buffer",
+)
diff --git a/codex-rs/utils/head-tail-buffer/Cargo.toml b/codex-rs/utils/head-tail-buffer/Cargo.toml
new file mode 100644
index 0000000000..85304d4dea
--- /dev/null
+++ b/codex-rs/utils/head-tail-buffer/Cargo.toml
@@ -0,0 +1,14 @@
+[package]
+name = "codex-utils-head-tail-buffer"
+version.workspace = true
+edition.workspace = true
+license.workspace = true
+
+[lints]
+workspace = true
+
+[dev-dependencies]
+pretty_assertions = { workspace = true }
+
+[lib]
+doctest = false
diff --git a/codex-rs/core/src/unified_exec/head_tail_buffer_tests.rs b/codex-rs/utils/head-tail-buffer/src/head_tail_buffer_tests.rs
similarity index 100%
rename from codex-rs/core/src/unified_exec/head_tail_buffer_tests.rs
rename to codex-rs/utils/head-tail-buffer/src/head_tail_buffer_tests.rs
diff --git a/codex-rs/core/src/unified_exec/head_tail_buffer.rs b/codex-rs/utils/head-tail-buffer/src/lib.rs
similarity index 87%
rename from codex-rs/core/src/unified_exec/head_tail_buffer.rs
rename to codex-rs/utils/head-tail-buffer/src/lib.rs
index 0c9df7c1be..1b4ee78078 100644
--- a/codex-rs/core/src/unified_exec/head_tail_buffer.rs
+++ b/codex-rs/utils/head-tail-buffer/src/lib.rs
@@ -1,14 +1,14 @@
-use crate::unified_exec::UNIFIED_EXEC_OUTPUT_MAX_BYTES;
-use crate::unified_exec::format_output_omission_marker;
+//! A capped byte buffer that preserves a stable prefix and the most recent suffix.
+//! Retained bytes never exceed the caller's const-generic budget.
+
 use std::collections::VecDeque;
 
 /// A capped buffer that preserves a stable prefix ("head") and suffix ("tail"),
 /// dropping the middle once it exceeds the configured maximum. The buffer is
 /// symmetric meaning 50% of the capacity is allocated to the head and 50% is
 /// allocated to the tail.
-#[derive(Debug, Default)]
-#[cfg_attr(test, derive(Eq, PartialEq))]
-pub(crate) struct HeadTailBuffer<const MAX_BYTES: usize = UNIFIED_EXEC_OUTPUT_MAX_BYTES> {
+#[derive(Debug, Default, Eq, PartialEq)]
+pub struct HeadTailBuffer<const MAX_BYTES: usize> {
     head: Vec<u8>,
     tail: VecDeque<u8>,
     omitted_bytes: usize,
@@ -18,22 +18,18 @@ impl<const MAX_BYTES: usize> HeadTailBuffer<MAX_BYTES> {
     const HEAD_BUDGET: usize = MAX_BYTES / 2;
     const TAIL_BUDGET: usize = MAX_BYTES.saturating_sub(Self::HEAD_BUDGET);
 
-    // Used for tests.
-    #[allow(dead_code)]
     /// Total bytes currently retained by the buffer (head + tail).
-    pub(crate) fn retained_bytes(&self) -> usize {
+    pub fn retained_bytes(&self) -> usize {
         self.head.len().saturating_add(self.tail.len())
     }
 
-    // Used for tests.
-    #[allow(dead_code)]
     /// Total bytes that were dropped from the middle due to the size cap.
-    pub(crate) fn omitted_bytes(&self) -> usize {
+    pub fn omitted_bytes(&self) -> usize {
         self.omitted_bytes
     }
 
     /// Total bytes observed by the buffer, including bytes omitted by the cap.
-    pub(crate) fn total_bytes(&self) -> usize {
+    pub fn total_bytes(&self) -> usize {
         self.retained_bytes().saturating_add(self.omitted_bytes)
     }
 
@@ -42,7 +38,7 @@ impl<const MAX_BYTES: usize> HeadTailBuffer<MAX_BYTES> {
     /// Bytes are first added to the head until the head budget is full; any
     /// remaining bytes are added to the tail, with older tail bytes being
     /// dropped to preserve the tail budget.
-    pub(crate) fn push_chunk(&mut self, chunk: &[u8]) {
+    pub fn push_chunk(&mut self, chunk: &[u8]) {
         let chunk = self.fill_head(chunk);
         self.push_tail(chunk);
     }
@@ -94,7 +90,7 @@ impl<const MAX_BYTES: usize> HeadTailBuffer<MAX_BYTES> {
     ///
     /// The output is formed by concatenating head chunks, then tail chunks.
     /// Omitted bytes are not represented in the returned value.
-    pub(crate) fn to_bytes(&self) -> Vec<u8> {
+    pub fn to_bytes(&self) -> Vec<u8> {
         let mut out = Vec::with_capacity(self.retained_bytes());
         out.extend_from_slice(&self.head);
         out.extend(self.tail.iter().copied());
@@ -103,7 +99,7 @@ impl<const MAX_BYTES: usize> HeadTailBuffer<MAX_BYTES> {
 
     /// Return the retained output with an explicit marker between the head and
     /// tail when bytes were omitted.
-    pub(crate) fn to_bytes_with_omission_marker(&self) -> Vec<u8> {
+    pub fn to_bytes_with_omission_marker(&self) -> Vec<u8> {
         if self.omitted_bytes == 0 {
             return self.to_bytes();
         }
@@ -125,7 +121,7 @@ impl<const MAX_BYTES: usize> HeadTailBuffer<MAX_BYTES> {
 
     /// Append a later buffer with the same budget. This preserves the summary
     /// of the original concatenated output, including its omission count.
-    pub(crate) fn push_buffer(&mut self, buffer: Self) {
+    pub fn push_buffer(&mut self, buffer: Self) {
         let Self {
             head,
             tail,
@@ -164,6 +160,11 @@ impl<const MAX_BYTES: usize> HeadTailBuffer<MAX_BYTES> {
     }
 }
 
+/// Render the byte omission notice used by callers of the buffer.
+pub fn format_output_omission_marker(omitted_bytes: usize) -> String {
+    format!("... {omitted_bytes} bytes omitted ...")
+}
+
 #[cfg(test)]
 #[path = "head_tail_buffer_tests.rs"]
 mod tests;