## Description
This PR adds an optional `metadata` field to `ResponseItem` for
Responses API calls. Only mechanical plumbing, no actual values
populated and sent yet. Turns out just adding a new field to
`ResponseItem` has quite a large blast radius already.
This change is backwards compatible because `metadata` is optional and
omitted when absent, so existing response items and rollout history
without it still deserialize and requests that do not set it keep the
same wire shape. For provider compatibility, we strip out `metadata`
before non-OpenAI Responses requests so Azure and AWS Bedrock never see
this field.
My followup PR here will actually make use of it to start storing and
passing along `turn_id`: https://github.com/openai/codex/pull/28360
## What changed
- Added `ResponseItemMetadata` with optional `turn_id`, plus optional
`metadata` on Responses API item variants and inter-agent communication.
- Preserved item metadata through response-item rewrites such as
truncation, missing tool-output synthesis, compaction history
rebuilding, visible-history conversion, rollout/resume, and generated
app-server schemas/types.
- Strip item metadata from non-OpenAI Responses requests while
preserving it for OpenAI-shaped requests.
- Updated the mechanical fixture/test construction churn required by the
new optional field.
## Summary
Add flat profiling fields to `codex_turn_event` so analytics can explain
where turn wall-clock time is spent without changing tool execution
behavior.
The profile reports:
- time before the first sampling request
- sampling time across all attempts and follow-ups
- overhead between sampling requests
- time blocked in the post-sampling tool drain
- time after the final sampling request
- sampling request and retry counts
## Implementation
- Extend the existing turn timing state with constant-memory phase
accounting and one RAII phase guard.
- Observe sampling and the existing post-sampling drain only at turn
orchestration boundaries.
- Keep tool runtime, tool futures, response item handling, and turn
lifecycle values unchanged.
- Add the profiling fields directly to the existing analytics turn event
without changing app-server protocol or rollout persistence.
- Use the existing turn `status` to distinguish completed, failed, and
interrupted profiles.
Exact sampling/tool overlap is intentionally omitted because measuring
tool completion accurately would require hooks in the tool execution
path.
## Validation
- Add app-server end-to-end coverage for a single-sampling turn with no
blocking tool work.
- Add app-server end-to-end coverage for `request_user_input` blocking
followed by a second sampling request.
- CI is running on the PR; tests were not executed locally per
repository guidance.
## Why
First-party backends can supply turn-scoped moderation metadata that
app-server clients need for client-side presentation. Exposing this as
an experimental typed notification lets opted-in clients consume it
without interpreting raw Responses API events.
## What changed
- forward `response.metadata.openai_chatgpt_moderation_metadata` from
Responses API SSE and WebSocket streams as turn-scoped moderation
metadata
- emit the experimental app-server v2 `turn/moderationMetadata`
notification with `{ threadId, turnId, metadata }`
- add app-server integration coverage for the typed moderation metadata
notification
## Testing
- `just test -p codex-core
build_ws_client_metadata_includes_window_lineage_and_turn_metadata`
- `just test -p codex-core` (fails locally: 46 failures and 1 timeout,
primarily missing `test_stdio_server` and shell snapshot timeouts)
- `just test -p codex-app-server-protocol`
- `just test -p codex-app-server
turn_moderation_metadata_emits_typed_notification_v2`
- `just test -p codex-app-server` (fails locally: 792 passed, 10 failed,
and 5 timed out; failures are in existing environment-sensitive tests,
primarily because nested macOS `sandbox-exec` is not permitted)
- `just write-app-server-schema --experimental --schema-root
/tmp/codex-app-server-schema-experimental`
## Why
Multi-agent v2 currently routes agent instructions through normal tool
arguments and inter-agent context. That means the parent model can emit
plaintext task text, Codex can persist it in history/rollouts, and the
recipient can receive it as ordinary assistant-message JSON.
This changes the v2 path so agent instructions stay encrypted between
model calls: Responses encrypts the `message` argument returned by the
model, Codex forwards only that ciphertext, and Responses decrypts it
internally for the recipient model.
## What changed
- Mark the v2 `message` parameter as encrypted for `spawn_agent`,
`send_message`, and `followup_task`.
- Treat multi-agent v2 tool `message` values as ciphertext
unconditionally.
- Store v2 inter-agent task text in
`InterAgentCommunication.encrypted_content` with empty plaintext
`content`.
- Convert encrypted inter-agent communications into the Responses
`agent_message` input item before sending the child request.
- Preserve `agent_message` items across history, rollout, compaction,
telemetry, and app-server schema paths.
- Leave multi-agent v1 unchanged.
## Message shape
The model still calls the v2 tools with a `message` argument, but that
value is now ciphertext:
```json
{
"name": "spawn_agent",
"arguments": {
"task_name": "worker",
"message": "<ciphertext>"
}
}
```
Codex stores the task as encrypted inter-agent communication:
```json
{
"author": "/root",
"recipient": "/root/worker",
"content": "",
"encrypted_content": "<ciphertext>",
"trigger_turn": true
}
```
When Codex builds the recipient request, it forwards the ciphertext
using the new Responses input item:
```json
{
"type": "agent_message",
"author": "/root",
"recipient": "/root/worker",
"content": [
{
"type": "encrypted_content",
"encrypted_content": "<ciphertext>"
}
]
}
```
Responses decrypts that item internally for the recipient model.
## Context impact
- Parent context no longer carries plaintext v2 agent task instructions
from these tool arguments.
- Codex rollout/history stores ciphertext for v2 agent instructions.
- Recipient requests receive an `agent_message` item instead of
assistant commentary JSON for encrypted task delivery.
- Plaintext completion/status notifications are still plaintext because
they are Codex-generated status messages, not encrypted model tool
arguments.
## Validation
- `just test -p codex-tools`
- `just test -p codex-protocol`
- `just test -p codex-rollout`
- `just test -p codex-rollout-trace`
- `just test -p codex-otel`
- `just write-app-server-schema`
## Why
Remote compaction v2 was still using `context_compaction` as both the
request trigger and the compacted output shape. The Responses API now
has the landed contract for this flow: Codex sends a dedicated `{
"type": "compaction_trigger" }` input item, and the backend returns the
standard `compaction` output item with encrypted content.
This aligns the v2 path with that wire contract while preserving the
existing local compacted-history post-processing behavior.
## What changed
- Add `ResponseItem::CompactionTrigger` and regenerate the app-server
protocol schema fixtures.
- Send `compaction_trigger` from `remote_compaction_v2` instead of a
payload-less `context_compaction`.
- Collect exactly one backend `compaction` output item, then reuse the
existing compacted-history rebuilding path.
- Treat the trigger item as a transient request marker rather than model
output or persisted rollout/memory content.
## Verification
- `cargo test -p codex-protocol compaction_trigger`
- `cargo test -p codex-core remote_compact_v2`
- `cargo test -p codex-core compact_remote_v2`
- `cargo test -p codex-core
responses_websocket_sends_response_processed_after_remote_compaction_v2`
- `just write-app-server-schema`
- `cargo test -p codex-app-server-protocol schema_fixtures`
## Why
While investigating `codex exec hi` startup latency, the useful
questions were not "is startup slow?" but "which durable bucket is slow
in production?"
The path we observed has a few distinct stages:
1. `thread/start` creates the session
2. startup prewarm builds the turn context, tools, and prompt
3. startup prewarm warms the websocket
4. the first real turn resolves the prewarm
5. the model produces the first token
Before this PR, production telemetry had some of the raw measurements
already:
- aggregate startup-prewarm duration / age-at-first-turn metrics
- TTFT as a metric
- websocket request telemetry
But there was no coherent production event stream for the startup
breakdown itself, and TTFT was metric-only. That made it hard to answer
the same latency questions from OpenTelemetry-backed logs without adding
one-off local instrumentation.
## What changed
Add durable production telemetry on the existing `SessionTelemetry`
path:
- new `codex.startup_phase` OTel log/trace events plus
`codex.startup.phase.duration_ms`
- new `codex.turn_ttft` OTel log/trace events while preserving the
existing TTFT metric
The startup phase event is emitted for the coarse buckets we actually
observed while running `exec hi`:
- `thread_start_create_thread`
- `startup_prewarm_total`
- `startup_prewarm_create_turn_context`
- `startup_prewarm_build_tools`
- `startup_prewarm_build_prompt`
- `startup_prewarm_websocket_warmup`
- `startup_prewarm_resolve`
These phases are intentionally low-cardinality so they remain safe as
production telemetry tags.
## Why this shape
This keeps the instrumentation on the same production path as the rest
of the session telemetry instead of adding a local debug-only trace
mode. It also avoids changing startup behavior:
- prewarm still runs
- no control flow changes
- no extra remote calls
- no user-visible behavior changes
One boundary is intentional: very early process bootstrap that happens
before a session exists is not included here, because this PR uses
session-scoped production telemetry. The expensive buckets we were
trying to understand after `thread/start` are now covered durably.
## Verification
- `cargo test -p codex-otel`
- `cargo test -p codex-core turn_timing`
- `cargo test -p codex-core
regular_turn_emits_turn_started_without_waiting_for_startup_prewarm`
- `cargo test -p codex-core
interrupting_regular_turn_waiting_on_startup_prewarm_emits_turn_aborted`
- `cargo test -p codex-app-server thread_start`
- `just fix -p codex-otel -p codex-core -p codex-app-server`
I also ran `cargo test -p codex-core`; it built successfully and then
hit an existing unrelated stack overflow in
`tools::handlers::multi_agents::tests::tool_handlers_cascade_close_and_resume_and_keep_explicitly_closed_subtrees_closed`.
## Why
Tool families already disagree on what their existing `duration` fields
mean, so lifecycle latency should live on the shared item envelope
instead of being inferred from per-tool execution fields. Carrying that
envelope through app-server notifications gives downstream consumers one
reusable timing signal without pretending every tool has the same
execution semantics.
## What changed
- Adds `started_at_ms` to core `ItemStartedEvent` values and
`completed_at_ms` to core `ItemCompletedEvent` values.
- Populates those timestamps in the shared session lifecycle emitters,
so protocol-native items get timing without each producer tracking its
own clock state.
- Exposes `startedAtMs` on app-server `item/started` notifications and
`completedAtMs` on `item/completed` notifications.
- Maps the lifecycle timestamps through the app-server boundary while
leaving legacy-converted notifications nullable when no lifecycle
timestamp exists.
- Regenerates the app-server JSON schema and TypeScript fixtures for the
notification-envelope change and updates downstream fixtures that
construct those notifications directly.
- Extends the existing web-search and image-generation integration flows
to assert the new lifecycle timestamps on the native item events.
## Verification
- `cargo check -p codex-protocol -p codex-core -p
codex-app-server-protocol -p codex-app-server -p codex-tui -p codex-exec
-p codex-app-server-client`
- `cargo test -p codex-core --test all web_search_item_is_emitted`
- `cargo test -p codex-core --test all
image_generation_call_event_is_emitted`
- `cargo test -p codex-app-server-protocol`
---
[//]: # (BEGIN SAPLING FOOTER)
Stack created with [Sapling](https://sapling-scm.com). Best reviewed
with [ReviewStack](https://reviewstack.dev/openai/codex/pull/20514).
* #18748
* #18747
* #17090
* #17089
* __->__ #20514
## Why
This adds the `remote_compaction_v2` client path so remote compaction
can run through the normal Responses stream and install a
`context_compaction` item that trigger a compaction.
The goal is to migrate some of the compaction logic on the client side
We keeps the v2 transport behind a feature flag while letting follow-up
requests reuse the compacted context instead of falling back to the
legacy compaction item shape.
## What changed
- add `ResponseItem::ContextCompaction` and refresh the generated
app-server / schema / TypeScript fixtures that expose response items on
the wire
- add `core/src/compact_remote_v2.rs` to send compaction through the
standard streamed Responses client, require exactly one
`context_compaction` output item, and install that item into compacted
history
- route manual compact and auto-compaction through the v2 path when
`remote_compaction_v2` is enabled, while keeping the existing remote
compaction path as the fallback
- preserve the new item type across history retention, follow-up request
construction, telemetry, rollout persistence, and rollout-trace
normalization
- add targeted coverage for the feature flag, `context_compaction`
serialization, rollout-trace normalization, and remote-compaction
follow-up behavior
## Verification
- added protocol tests for `context_compaction`
serialization/deserialization in `protocol/src/models.rs`
- added rollout-trace coverage for `context_compaction` normalization in
`rollout-trace/src/reducer/conversation_tests.rs`
- added remote compaction integration coverage for v2 follow-up reuse
and mixed compaction output streams in
`core/tests/suite/compact_remote.rs`
---------
Co-authored-by: Codex <noreply@openai.com>
## Why
- Without change: MCP tool calls receive
`_meta["x-codex-turn-metadata"]` with `session_id` and `turn_id`.
- Issue: MCP servers may want the turn start timestamp to measure
internal latency relative to turn start.
## What Changed
- With change: turn metadata now includes `turn_started_at_unix_ms`,
which is propagated to MCP tool calls in
`_meta["x-codex-turn-metadata"]`.
## Verification
- `codex-rs/core/src/mcp_tool_call_tests.rs`
- `codex-rs/core/src/turn_metadata_tests.rs`
- `codex-rs/core/src/turn_timing_tests.rs`
- `codex-rs/core/tests/responses_headers.rs`
- `codex-rs/core/tests/suite/search_tool.rs`
## Summary
- Remove `ghost_snapshot` / `GhostCommit` from the Responses API surface
and generated SDK/schema artifacts.
- Keep legacy config loading compatible, but make undo a no-op that
reports the feature is unavailable.
- Clean up core history, compaction, telemetry, rollout, and tests to
stop carrying ghost snapshot items.
## Testing
- Unit tests passed for `codex-protocol`, `codex-core` targeted undo and
compaction flows, `codex-rollout`, and `codex-app-server-protocol`.
- Regenerated config and app-server schemas plus Python SDK artifacts
and verified they match the checked-in outputs.
## Why
Guardian analytics includes time-to-first-token, but the Guardian
reviewer runs as a normal Codex session and `TurnCompleteEvent` did not
expose TTFT. The timing needs to flow through the standard
turn-completion protocol so Guardian review analytics can consume the
same value as the rest of the session machinery.
## What changed
Adds optional `time_to_first_token_ms` to `TurnCompleteEvent` and
populates it from `TurnTiming`. The value is carried through app-server
thread history, rollout reconstruction, TUI/app-server adapters, and
Guardian review session handling.
Guardian review analytics now captures TTFT from the reviewer
turn-complete event when available. Existing tests and fixtures are
updated to set the new optional field to `None` where TTFT is not
relevant.
## Verification
- `cargo clippy -p codex-tui --tests -- -D warnings`
- `cargo clippy -p codex-core --lib --tests -- -D warnings`
---
[//]: # (BEGIN SAPLING FOOTER)
Stack created with [Sapling](https://sapling-scm.com). Best reviewed
with [ReviewStack](https://reviewstack.dev/openai/codex/pull/17696).
* __->__ #17696
* #17695
* #17693
* #18278
* #18953
Adds new events for streaming apply_patch changes from responses api.
This is to enable clients to show progress during file writes.
Caveat: This does not work with apply_patch in function call mode, since
that required adding streaming json parsing.
## Summary
- reduce public module visibility across Rust crates, preferring private
or crate-private modules with explicit crate-root public exports
- update external call sites and tests to use the intended public crate
APIs instead of reaching through module trees
- add the module visibility guideline to AGENTS.md
## Validation
- `cargo check --workspace --all-targets --message-format=short` passed
before the final fix/format pass
- `just fix` completed successfully
- `just fmt` completed successfully
- `git diff --check` passed
## Why
PR #13783 moved the `codex.rs` unit tests into `codex_tests.rs`. This
applies the same extraction pattern across the rest of `codex-rs/core`
so the production modules stay focused on runtime code instead of large
inline test blocks.
Keeping the tests in sibling files also makes follow-up edits easier to
review because product changes no longer have to share a file with
hundreds or thousands of lines of test scaffolding.
## What changed
- replaced each inline `mod tests { ... }` in `codex-rs/core/src/**`
with a path-based module declaration
- moved each extracted unit test module into a sibling `*_tests.rs`
file, using `mod_tests.rs` for `mod.rs` modules
- preserved the existing `cfg(...)` guards and module-local structure so
the refactor remains structural rather than behavioral
## Testing
- `cargo test -p codex-core --lib` (`1653 passed; 0 failed; 5 ignored`)
- `just fix -p codex-core`
- `cargo fmt --check`
- `cargo shear`
## Why
to support a new bring your own search tool in Responses
API(https://developers.openai.com/api/docs/guides/tools-tool-search#client-executed-tool-search)
we migrating our bm25 search tool to use official way to execute search
on client and communicate additional tools to the model.
## What
- replace the legacy `search_tool_bm25` flow with client-executed
`tool_search`
- add protocol, SSE, history, and normalization support for
`tool_search_call` and `tool_search_output`
- return namespaced Codex Apps search results and wire namespaced
follow-up tool calls back into MCP dispatch
This cleans up a bunch of metric plumbing that had started to drift.
The main change is making `codex-otel` the canonical home for shared
metric definitions and metric tag helpers. I moved the `turn/thread`
metric names that were still duplicated into the OTEL metric registry,
added a shared `metrics::tags` module for common tag keys and session
tag construction, and updated `SessionTelemetry` to build its metadata
tags through that shared path.
On the codex-core side, TTFT/TTFM now use the shared metric-name
constants instead of local string definitions. I also switched the
obvious remaining turn/thread metric callsites over to the shared
constants, and added a small helper so TTFT/TTFM can attach an optional
sanitized client.name tag from TurnContext.
This should make follow-on telemetry work less ad hoc:
- one canonical place for metric names
- one canonical place for common metric tag keys/builders
- less duplication between `codex-core` and `codex-otel`
## 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`
### Summary
This adds turn-level latency metrics for the first model output and the
first completed agent message.
- `codex.turn.ttft.duration_ms` starts at turn start and records on the
first output signal we see from the model. That includes normal
assistant text, reasoning deltas, and non-text outputs like tool-call
items.
- `codex.turn.ttfm.duration_ms` also starts at turn start, but it
records when the first agent message finishes streaming rather than when
its first delta arrives.
### Implementation notes
The timing is tracked in codex-core, not app-server, so the definition
stays consistent across CLI, TUI, and app-server clients.
I reused the existing turn lifecycle boundary that already drives
`codex.turn.e2e_duration_ms`, stored the turn start timestamp in turn
state, and record each metric once per turn.
I also wired the new metric names into the OTEL runtime metrics summary
so they show up in the same in-memory/debug snapshot path as the
existing timing metrics.
Owen Lin ·
Ahmed Ibrahim ·
carlc-oai ·
jif ·
Matthew Zeng ·
rhan-oai ·
mchen-oai ·
pakrym-oai ·
Eric Traut ·
Akshay Nathan ·
Michael Bolin ·
Anton Panasenko ·