## Why
After `hooks/list` exposes the hook inventory, clients need a way to
persist user hook preferences, make those changes effective in
already-open sessions, and distinguish user-controllable hooks from
managed requirements without adding another bespoke app-server write
API.
## What
- Extends `hooks/list` entries with effective `enabled` state.
- Persists user-level hook state under `hooks.state.<hook-id>` so the
model can grow beyond a single boolean over time.
- Uses the existing `config/batchWrite` path for hook state updates
instead of introducing a dedicated hook write RPC.
- Refreshes live session hook engines after config writes so
already-open threads observe updated enablement without a restart.
## Stack
1. openai/codex#19705
2. openai/codex#19778
3. This PR - openai/codex#19840
4. openai/codex#19882
## Reviewer Notes
The generated schema files account for much of the raw diff. The core
behavior is in:
- `hooks/src/config_rules.rs`, which resolves per-hook user state from
the config layer stack.
- `hooks/src/engine/discovery.rs`, which projects effective enablement
into `hooks/list` from source-derived managedness.
- `config/src/hook_config.rs`, which defines the new `hooks.state`
representation.
- `core/src/session/mod.rs`, which rebuilds live hook state after user
config reloads.
---------
Co-authored-by: Codex <noreply@openai.com>
## Why
This bug is exposed by Guardian/auto-review approvals. With the managed
network proxy enabled, a blocked network request can be reported back
through the network approval service as an approval denial after the
command has already started. Before this change, the shell and unified
exec runtimes registered those network approval calls, but did not have
a way to observe an async proxy denial as a cancellation/failure signal
for the running process.
The result was confusing: Guardian/auto-review could correctly deny
network access, but the command path could keep running or unregister
the approval without surfacing the denial as the command failure.
## What Changed
- `NetworkApprovalService` now attaches a cancellation token to active
and deferred network approvals.
- Proxy-denial outcomes are recorded only for active registrations,
cancel the owning token, and are consumed when the approval is
finalized.
- The shell runtime combines the normal command timeout with the
network-denial cancellation token.
- Unified exec stores the deferred network approval object, terminates
tracked processes when the proxy denial arrives, and returns the denial
as a process failure while polling or completing the process.
- Tool orchestration passes the active network approval cancellation
token into the sandbox attempt and preserves deferred approval errors
instead of silently unregistering them.
- App-server `command/exec` now handles the combined
timeout-or-cancellation expiration variant used by the runtime.
## Verification
- `cargo test -p codex-core network_approval --lib`
- `cargo clippy -p codex-app-server --all-targets -- -D warnings`
- `cargo clippy -p codex-core --all-targets -- -D warnings`
---------
Co-authored-by: Codex <noreply@openai.com>
- Fetches and caches remote /installed plugin state
- Lets skills/list load skills from remote-installed cached plugins
without requiring a local marketplace entry
- Routes plugin list/startup/install/uninstall changes through async
plugin cache invalidation and MCP refresh
## Summary
- include the live auto-review trunk rollout when `/feedback` uploads
logs
- upload that attachment as
`auto-review-rollout-<parent-thread-id>.jsonl` so it is distinguishable
from the parent rollout
- show the same auto-review attachment name in the TUI consent popup
## Scope
- this only covers the live cached auto-review trunk for the current
parent thread
- it does not add durable historical parent->auto-review lookup
- it does not add persisted rollout support for ephemeral parallel
review forks
## UI
<img width="599" height="185" alt="Screenshot 2026-04-28 at 1 17 18 PM"
src="https://github.com/user-attachments/assets/6a0e79c2-5d21-4702-8a89-f765778bc9e9"
/>
## Validation
- `cargo test -p codex-core
cached_guardian_subagent_exposes_its_rollout_path`
- `cargo test -p codex-feedback`
- `cargo test -p codex-app-server`
- `cargo test -p codex-tui feedback_upload_consent_popup_snapshot`
- `cargo test -p codex-tui
feedback_good_result_consent_popup_includes_connectivity_diagnostics_filename`
## Known unrelated local failures
- `cargo test -p codex-core` currently fails in the pre-existing proxy
env snapshot test
`tools::runtimes::tests::maybe_wrap_shell_lc_with_snapshot_keeps_user_proxy_env_when_proxy_inactive`
- `cargo test -p codex-tui` currently hits pre-existing `status::*`
snapshot drift unrelated to this change
## Follow-Up
- persist parallel auto-review fork sessions so /feedback can include
their rollout history too
- attach each persisted fork as its own clearly named file, for example
auto-review-rollout-<parent-thread-id>-fork <n>.jsonl, instead of
merging multiple Guardian sessions into one attachment
- keep the same live-session-only scope initially; durable historical
parent -> auto-review lookup can remain a separate decision if we later
need feedback from resumed sessions
Summary:
- Add codex-thread-manager-sample, a one-shot binary that starts a
ThreadManager thread, submits a prompt, and prints the final assistant
output.
- Pass ThreadStore into ThreadManager::new and expose
thread_store_from_config for existing callsites.
- Build the sample Config directly with only --model and prompt inputs.
Verification:
- just fmt
- cargo check -p codex-thread-manager-sample -p codex-app-server -p
codex-mcp-server
- git diff --check
Tests: Not run per request.
## Why
`agents.max_depth` is a legacy multi-agent v1 guard. Multi-agent v2 uses
task-path routing and its own session/thread limits, so v2 should not
reject nested `spawn_agent` calls just because the thread-spawn depth
has reached the v1 maximum.
Keeping the v1 depth guard active in v2 prevents deeper task trees even
though the v2 path still needs the depth value only for lineage and
task-path metadata.
## What Changed
- Removed the depth-limit rejection from the multi-agent v2
`spawn_agent` handler while still computing child depth for lineage/path
metadata.
- Made the depth-based disabling of legacy `SpawnCsv`/`Collab` tools
apply only when `Feature::MultiAgentV2` is disabled.
- Added `multi_agent_v2_spawn_agent_ignores_configured_max_depth` to
cover a v2 child spawning another agent when `agent_max_depth = 1`,
while the existing v1 depth-limit tests continue to enforce the legacy
behavior.
## Verification
- `cargo test -p codex-core
multi_agent_v2_spawn_agent_ignores_configured_max_depth -- --nocapture`
- `cargo test -p codex-core depth_limit -- --nocapture`
- `cargo test -p codex-core tools::handlers::multi_agents::tests --
--nocapture`
## Why
Unsupported features must fail closed and Codex must not expose
OpenAI-hosted fallback paths when the active provider cannot support
them. In practice, Bedrock should not surface app connectors, MCP
servers, tool search/suggestions, image generation, web search, or JS
REPL until those paths are explicitly supported for that provider.
This PR moves that decision into provider-owned capability metadata
instead of scattering Bedrock-specific checks across callers.
## What changed
- Adds `ProviderCapabilities` to `codex-model-provider`, with default
support for existing providers and a Bedrock override that disables
unsupported launch surfaces.
- Adds `ToolCapabilityBounds` to `codex-tools` so provider capability
limits can clamp otherwise-enabled tool config.
- Applies capability bounds when building session and review-thread tool
config.
- Routes MCP/app connector configuration through
`McpManager::mcp_config`, which filters configured MCP servers and app
connectors based on the active provider.
- Updates app-server MCP list/read paths to use the filtered MCP config.
- Adds coverage for default provider capabilities, Bedrock disabled
capabilities, and optional tool-surface clamping.
## Testing
built locally and verified that bedrock responses api now return without
errors calling unsupported tools.
## Summary
- Add `disable_tool_suggest` to app and plugin config, schema, and
TypeScript output
- Exclude disabled connectors and plugins from tool suggestion discovery
- Persist "never show again" tool-suggestion choices back into
`config.toml`
- Update config docs and add coverage for connector and plugin
suppression
## Testing
- Added and updated unit tests for config persistence and tool-suggest
filtering
- Not run (not requested)
## Why
Plugins can bundle lifecycle hooks, but Codex previously only discovered
hooks from user, project, and managed config layers. This adds the
plugin discovery and runtime plumbing needed for plugin-bundled hooks
while keeping execution behind the `plugin_hooks` feature flag.
## What
- Discovers plugin hook sources from each plugin's default
`hooks/hooks.json`.
- Supports `plugin.json` manifest `hooks` entries as either relative
paths or inline hook objects.
- Plumbs discovered plugin hook sources through plugin loading into the
hook runtime when `plugin_hooks` is enabled.
- Marks plugin-originated hook runs as `HookSource::Plugin`.
- Injects `PLUGIN_ROOT` and `CLAUDE_PLUGIN_ROOT` into plugin hook
command environments.
- Updates generated schemas and hook source metadata for the plugin hook
source.
## Stack
1. This PR - openai/codex#19705
2. openai/codex#19778
3. openai/codex#19840
4. openai/codex#19882
## Reviewer Notes
- Core logic is in `codex-rs/core-plugins/src/loader.rs` and
`codex-rs/hooks/src/engine/discovery.rs`
- Moved existing / adding new tests to
`codex-rs/core-plugins/src/loader_tests.rs` hence the large diff there
- Otherwise mostly plumbing and minor schema updates
### Core Changes
The `codex-rs/core` changes are limited to wiring plugin hook support
into existing core flows:
- `core/src/session/session.rs` conditionally pulls effective plugin
hook sources and plugin hook load warnings from `PluginsManager` when
`plugin_hooks` is enabled, then passes them into `HooksConfig`.
- `core/src/hook_runtime.rs` adds the `plugin` metric tag for
`HookSource::Plugin`.
- `core/config.schema.json` picks up the new `plugin_hooks` feature
flag, and `core/src/plugins/manager_tests.rs` updates fixtures for the
added plugin hook fields.
---------
Co-authored-by: Codex <noreply@openai.com>
## Why
MultiAgentV2 `wait_agent` currently clamps short waits to a fixed 10
second minimum. That default is still useful for preventing tight
polling loops, but it is too rigid for environments that need faster
mailbox wake-up checks or a larger minimum to discourage frequent
polling.
This PR makes the minimum wait timeout configurable from the existing
MultiAgentV2 feature config section, so operators can tune the behavior
without changing the legacy multi-agent tool surface.
## What Changed
- Added `features.multi_agent_v2.min_wait_timeout_ms`.
- Defaulted the new setting to the existing 10 second floor.
- Validated the configured value as `1..=3600000`, matching the existing
one hour maximum wait bound.
- Applied the configured minimum to MultiAgentV2 `wait_agent` runtime
clamping.
- Plumbed the configured minimum into the `wait_agent` tool schema,
including the effective default when the minimum is above the normal 30
second default.
- Regenerated `core/config.schema.json`.
## Verification
- `cargo test -p codex-features`
- `cargo test -p codex-tools`
- `cargo test -p codex-core --lib multi_agent_v2`
- `just fix -p codex-core`
## Why
Slow Codex turns are easier to debug when token usage is visible in the
trace itself, without joining against separate analytics. This adds
token usage to existing turn-handling spans for regular user turns only.
[Example
turn](https://openai.datadoghq.com/apm/trace/9d353efa2cb5de1f4c5b93dc33c3df04?colorBy=service&graphType=flamegraph&shouldShowLegend=true&sort=time&spanID=3555541504891512675&spanViewType=metadata&traceQuery=)
<img width="1447" height="967" alt="Screenshot 2026-04-24 at 3 03 07 PM"
src="https://github.com/user-attachments/assets/ab7bb187-e7fc-41f0-a366-6c44610b2b2c"
/>
## What Changed
Added response-level token fields on completed handle_responses spans:
gen_ai.usage.input_tokens
gen_ai.usage.cache_read.input_tokens
gen_ai.usage.output_tokens
codex.usage.reasoning_output_tokens
codex.usage.total_tokens
Added aggregate token fields on regular turn spans:
codex.turn.token_usage.*
Added an explicit regular-turn opt-in via
SessionTask::records_turn_token_usage_on_span() so this is not coupled
to span-name strings.
## Testing
- `cargo test -p codex-otel`
- `cargo test -p codex-core
turn_and_completed_response_spans_record_token_usage`
- `just fmt`
- `just fix -p codex-core`
- `just fix -p codex-otel`
- Manual local Electron/app-server smoke test: regular user turn emits
the new span fields
Known status: `cargo test -p codex-core` was attempted and failed in
unrelated existing areas: config approvals, request-permissions,
git-info ordering, and subagent metadata persistence.
## Why
Several bug reports describe thread shutdown (including subagent
threads) leaving stdio MCP server processes behind. These reports all
point at the same lifecycle gap: Codex launches stdio MCP servers, but
the session-level shutdown path does not explicitly close MCP clients or
terminate the server process tree.
Fixes#12491Fixes#12976Fixes#18881Fixes#19469
## History
This is best understood as a regression/coverage gap in MCP session
lifecycle management, not as stdio MCP cleanup being absent all along.
#10710 added process-group cleanup for stdio MCP servers, but that
cleanup only runs when the `RmcpClient`/transport is dropped. The older
reports (#12491 and #12976) came after that cleanup existed, which
suggests the remaining problem was that some higher-level shutdown paths
kept the MCP manager alive or replaced it without explicitly draining
clients. The newer reports (#18881 and #19469) exposed the same family
around manager replacement and shutdown.
## What changed
- Added an explicit stdio MCP process handle in `codex-rmcp-client` so
local MCP servers terminate their process group and executor-backed MCP
servers call the executor process terminator.
- Added `RmcpClient::shutdown()` and manager-level MCP shutdown draining
so session shutdown, channel-close fallback, MCP refresh, and connector
probing stop owned MCP clients.
- Added regression coverage that starts a stdio MCP server, begins an
in-flight blocking tool call, shuts down the client, and asserts the
server process exits.
## Verification
- `cargo test -p codex-rmcp-client`
- `cargo test -p codex-mcp`
- `just fix -p codex-rmcp-client`
- `just fix -p codex-mcp`
- `just fix -p codex-core`
- Manual before/after validation with a temporary repro script:
- Pre-fix binary from `HEAD^` (`fed0a8f4fa`): reproduced the leak with
surviving MCP server and child PIDs, `survivors=[77583, 77592]`,
`leaked=true`.
- Post-fix binary from this branch (`67e318148b`): verified both MCP
processes were gone after interrupting `codex exec`, `survivors=[]`,
`leaked=false`.
## Why
Phase 2 still needs to choose the most relevant stage-1 memory outputs
by usage and recency, but exposing that ranking as the rendered
`raw_memories.md` order creates unnecessary large diff. Usage-count or
timestamp changes can reshuffle otherwise unchanged memories, making the
workspace diff noisy and giving the consolidation prompt a misleading
recency signal from file position.
This fix will reduce token consumption
## What Changed
- Keep the existing top-N Phase 2 selection ranking by `usage_count`,
`last_usage`, `source_updated_at`, and `thread_id`.
- Return the selected rows in stable ascending `thread_id` order before
syncing Phase 2 filesystem inputs.
- Update the memory README, raw memories header, and consolidation
prompt so they describe the stable order and tell the prompt to use
metadata and workspace diffs instead of file order as the recency
signal.
- Adjust the memory runtime tests to use deterministic thread IDs and
assert the stable return order separately from the ranked selection
semantics.
## Test Coverage
- Existing memory runtime tests in
`codex-rs/state/src/runtime/memories.rs` now cover the stable returned
ordering for Phase 2 inputs.
---------
Co-authored-by: Codex <noreply@openai.com>
Keep extracting memories out of core and moving the write trigger in the
app-server
This is temporary and it should move at the client level as a follow-up
This makes core fully independant from `codex-memories-write`
---------
Co-authored-by: Codex <noreply@openai.com>
## Why
MultiAgentV2 sessions need startup guidance that matches the role of the
thread that is actually being created. Root agents and subagents have
different responsibilities, and forked subagents can inherit parent
rollout history. If the parent hint is carried into the child context,
the child can see stale or conflicting developer guidance before its own
session-specific context is added.
## What changed
- Added `features.multi_agent_v2.root_agent_usage_hint_text` and
`features.multi_agent_v2.subagent_usage_hint_text` config fields,
including schema/config parsing support.
- Injected the matching root or subagent hint into the initial context
as its own developer message when `multi_agent_v2` is enabled.
- Filtered configured MultiAgentV2 usage-hint developer messages out of
forked parent history so a child thread receives fresh guidance for its
own session source/config.
- Added targeted coverage for config parsing, initial-context rendering,
feature-config deserialization, and forked-history filtering.
## Context examples
With this config:
```toml
[features.multi_agent_v2]
enabled = true
root_agent_usage_hint_text = "Root guidance."
subagent_usage_hint_text = "Subagent guidance."
```
A root thread initial context renders the root hint as a standalone
developer message:
```text
[developer]
<existing developer context, when present>
[developer]
Root guidance.
```
A subagent thread initial context renders the subagent hint instead:
```text
[developer]
<existing developer context, when present>
[developer]
Subagent guidance.
```
When a subagent forks parent history, any parent developer message whose
text exactly matches the configured MultiAgentV2 root or subagent hint
is omitted from the forked history before the child receives its fresh
subagent hint.
## Why
`ThreadConfigSnapshot` is used by app-server and thread metadata code as
a stable view of active runtime settings. Keeping both `sandbox_policy`
and `permission_profile` in the snapshot duplicates permission state and
makes it possible for the legacy projection to drift from the canonical
profile.
The legacy `sandbox` value is still needed at app-server compatibility
boundaries, so this PR derives it on demand from the snapshot profile
and cwd instead of storing it.
## What Changed
- Removes `ThreadConfigSnapshot.sandbox_policy`.
- Adds `ThreadConfigSnapshot::sandbox_policy()` as a compatibility
projection from `permission_profile` plus `cwd`.
- Updates app-server response/metadata code and tests to call the
projection only where legacy fields still exist.
- Keeps snapshot construction profile-only so split filesystem rules,
disabled enforcement, and external enforcement remain represented by the
canonical profile.
## Verification
- `cargo test -p codex-app-server
thread_response_permission_profile_preserves_enforcement --lib`
- `cargo test -p codex-core
dispatch_reclaims_stale_global_lock_and_starts_consolidation --lib`
---
[//]: # (BEGIN SAPLING FOOTER)
Stack created with [Sapling](https://sapling-scm.com). Best reviewed
with [ReviewStack](https://reviewstack.dev/openai/codex/pull/19775).
* #19900
* #19899
* #19776
* __->__ #19775
## Why
`SessionConfiguredEvent` is the internal event that tells clients what
permissions are active for a session. Emitting both `sandbox_policy` and
`permission_profile` leaves two possible authorities and forces every
consumer to decide which one to honor. At this point in the migration,
the profile is expressive enough to represent managed, disabled, and
external sandbox enforcement, so the internal event can be profile-only.
The wire compatibility concern is older serialized events or rollout
data that only contain `sandbox_policy`; those still need to
deserialize.
## What Changed
- Removes `sandbox_policy` from `SessionConfiguredEvent` and makes
`permission_profile` required.
- Adds custom deserialization so old payloads with only `sandbox_policy`
are upgraded to a cwd-anchored `PermissionProfile`.
- Updates core event emission and TUI session handling to sync
permissions from the profile directly.
- Updates app-server response construction to derive the legacy
`sandbox` response field from the active thread snapshot instead of from
`SessionConfiguredEvent`.
- Updates yolo-mode display logic to treat both
`PermissionProfile::Disabled` and managed unrestricted filesystem plus
enabled network as full-access, while still preserving the distinction
between no sandbox and external sandboxing.
## Verification
- `cargo test -p codex-protocol session_configured_event --lib`
- `cargo test -p codex-protocol serialize_event --lib`
- `cargo test -p codex-exec session_configured --lib`
- `cargo test -p codex-app-server
thread_response_permission_profile_preserves_enforcement --lib`
- `cargo test -p codex-core
session_configured_reports_permission_profile_for_external_sandbox
--lib`
- `cargo test -p codex-tui session_configured --lib`
- `cargo test -p codex-tui
yolo_mode_includes_managed_full_access_profiles --lib`
---
[//]: # (BEGIN SAPLING FOOTER)
Stack created with [Sapling](https://sapling-scm.com). Best reviewed
with [ReviewStack](https://reviewstack.dev/openai/codex/pull/19774).
* #19900
* #19899
* #19776
* #19775
* __->__ #19774
## Why
Fixes#19475.
`codex exec` can finish successfully and then emit an `ERROR` on stderr:
```text
failed to record rollout items: thread <id> not found
```
That happens because shutdown closes the live thread writer before
emitting `ShutdownComplete`. The terminal event was still using the
normal `send_event_raw` path, so it tried to append rollout items
through a recorder that had already been removed. The answer is correct,
but wrappers that treat stderr as failure can retry completed exec runs.
This looks like a likely recent regression from
[#18882](https://github.com/openai/codex/pull/18882), which routed live
thread writes through `ThreadStore` and added the shutdown-time live
writer close. I have not bisected this, so the PR treats #18882 as the
likely source based on the affected shutdown code path rather than a
proven first-bad commit.
## What Changed
`ShutdownComplete` now bypasses rollout persistence after thread
shutdown and is delivered directly to clients. The shutdown path still
records the protocol event in the rollout trace before delivery,
preserving trace visibility without attempting a post-shutdown
thread-store append.
The change also adds a regression test with the in-memory thread store
to assert that shutdown creates and shuts down the live thread without
appending another item after shutdown.
## 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
The experimental `PermissionProfile` API had both `:cwd` and
`:project_roots` special filesystem paths, which made the permission
root ambiguous. This PR removes the unstable `current_working_directory`
special path before the permissions API is stabilized, so callers use
`:project_roots` for symbolic project-root access.
## What changed
- Removes `FileSystemSpecialPath::CurrentWorkingDirectory` from protocol
and app-server protocol models, plus regenerated app-server
JSON/TypeScript schemas.
- Replaces internal `:cwd` permission entries with `:project_roots`
entries.
- Keeps the existing cwd-update behavior for legacy-shaped
workspace-write profiles, while removing the deleted
`CurrentWorkingDirectory` case from that compatibility path.
- Keeps `PermissionProfile::workspace_write()` as the reusable symbolic
workspace-write helper, with docs noting that `:project_roots` entries
resolve at enforcement time.
- Updates app-server docs/examples and approval UI labeling to stop
advertising `:cwd` as a permission token.
## Compatibility
Persisted rollout items may contain the old
`{"kind":"current_working_directory"}` tag from earlier experimental
`permissionProfile` snapshots. This PR keeps that tag as a
deserialize-only alias for `ProjectRoots { subpath: None }`, while
continuing to serialize only the new `project_roots` tag.
## Follow-up
This PR intentionally does not introduce an explicit project-root set on
`SessionConfiguration` or runtime sandbox resolution. Today, the
resolver still uses the active cwd as the single implicit project root.
A follow-up should model project roots separately from tool cwd so
`:project_roots` entries can resolve against the configured project
roots, and resolve to no entries when there are no project roots.
## Verification
- `cargo test -p codex-protocol permissions:: --lib`
- `cargo test -p codex-app-server-protocol`
- `cargo test -p codex-sandboxing -p codex-exec-server --lib`
- `cargo test -p codex-core session_configuration_apply_ --lib`
- `cargo test -p codex-app-server
command_exec_permission_profile_project_roots_use_command_cwd --test
all`
- `cargo test -p codex-tui
thread_read_session_state_does_not_reuse_primary_permission_profile
--lib`
- `cargo test -p codex-tui
preset_matching_accepts_workspace_write_with_extra_roots --lib`
- `cargo test -p codex-config --lib`
fixes#19486
### Problem
Right now dynamic deferred tools are filtered at normal-turn prompt
building time, rather than upstream while building the `ToolRouter`
itself. This causes issues because dynamic deferred tools are then
wrongly included in the router's `model_visible_specs`, which is what
the compaction request-building flow relies on.
### Fix
Move the dynamic deferred tool filtering to `ToolRouter` creation time
to solve this problem for every request that relies on `ToolRouter` for
`model_visible_specs`, which solves the issue generically.
### Tests
Added unit + integration tests to ensure dynamic deferred tools are
omitted from `model_visible_specs` and compaction request respectively.
Tested against live `/compact` endpoint; raw deferred dynamic tools
without `tool_search` returned `400` (current bug), while the filtered
payload (this fix) returns `200`.
## Why
`features.multi_agent_v2.max_concurrent_threads_per_session` is meant to
be the MultiAgentV2-specific session thread cap: it counts the root
thread and all open subagent threads. The previous implementation kept
this surface tied to `agents.max_threads`, which made it a global
subagent-only cap and allowed the legacy setting to coexist with
MultiAgentV2.
## What Changed
- Added `max_concurrent_threads_per_session` to
`[features.multi_agent_v2]` with default `4`.
- Removed the `[agents] max_concurrent_threads_per_session` alias to
`agents.max_threads`.
- When MultiAgentV2 is enabled, reject `agents.max_threads` and derive
the existing internal subagent slot limit as
`max_concurrent_threads_per_session - 1`.
- Regenerated `core/config.schema.json` and added coverage for the new
config semantics.
## Result
```
➜ codex git:(jif/clean-multi-agent-v2-config) codex -c features.multi_agent_v2.enabled=true -c features.multi_agent_v2.max_concurrent_threads_per_session=3
╭────────────────────────────────────────────────────╮
│ >_ OpenAI Codex (v0.0.0) │
│ │
│ model: gpt-5.5 xhigh fast /model to change │
│ directory: ~/code/codex │
╰────────────────────────────────────────────────────╯
Tip: Update Required - This version will no longer be supported starting May 8th. Please upgrade to the latest version (https://github.com/openai/codex/releases/latest) using your preferred package manager.
› Can you try to spawn 4 agents
• I’ll try to start four lightweight agents at once and report exactly what the runtime accepts.
• Spawned Russell [no-apps] (gpt-5.5 xhigh)
└ Spawn probe 1: reply briefly that you started, then wait for further instructions. Do not do any repo work.
• Spawned Descartes [no-apps] (gpt-5.5 xhigh)
└ Spawn probe 2: reply briefly that you started, then wait for further instructions. Do not do any repo work.
• Agent spawn failed
└ Spawn probe 3: reply briefly that you started, then wait for further instructions. Do not do any repo work.
• Agent spawn failed
└ Spawn probe 4: reply briefly that you started, then wait for further instructions. Do not do any repo work.
• The runtime accepted the first two and rejected the next two with agent thread limit reached. I’m checking whether the two accepted probes have returned cleanly, then I’ll close them if needed.
```
---------
Co-authored-by: Codex <noreply@openai.com>
## Why
Auto-review can deny an action that the user later decides they want to
retry. Today there is no TUI surface for selecting a recent denial and
sending explicit approval context back into the session, so users have
to restate intent manually and the retry can be reviewed without the
original denied action context.
This adds a narrow TUI-driven path for approving a recent denied action
while still keeping the retry inside the normal auto-review flow.
## What Changed
- Added `/auto-review-denials` to open a picker of recent denied
auto-review actions.
- Added a small in-memory TUI store for the 10 most recent denied
auto-review events.
- Selecting a denial sends the structured denied event back through the
existing core/app-server op path.
- Core now injects a developer message containing the approved action
JSON rather than the full assessment event.
- Auto-review transcript collection now preserves this specific approval
developer message so follow-up review sessions can see the user approval
context.
- Added TUI snapshot/unit coverage for the picker and approval dispatch
path.
- Added core coverage for retaining the approval developer message in
the auto-review transcript.
## Verification
- `cargo test -p codex-core
collect_guardian_transcript_entries_keeps_manual_approval_developer_message`
- `cargo test -p codex-tui auto_review_denials`
- `cargo test -p codex-tui
approving_recent_denial_emits_structured_core_op_once`
## Notes
This intentionally keeps retries going through auto-review. The approval
signal is context for the exact previously denied action, not a blanket
bypass for similar future actions.
## Why
The remaining migration work still needs `SandboxPolicy` at a few
compatibility boundaries, but those projections should come from one
canonical path. Keeping ad hoc legacy projections scattered through
app-server, CLI, and config code makes it easy for behavior to drift as
`PermissionProfile` gains fidelity that the legacy enum cannot
represent.
## What Changed
- Adds `Permissions::legacy_sandbox_policy(cwd)` and
`Config::legacy_sandbox_policy()` as the compatibility projection from
the canonical `PermissionProfile`.
- Adds `Permissions::can_set_legacy_sandbox_policy()` so legacy inputs
are checked after they are converted into profile semantics.
- Updates app-server command handling, Windows sandbox setup, session
configuration, and sandbox summaries to use the centralized projection
helper.
- Leaves `SandboxPolicy` in place only for boundary inputs/outputs that
still speak the legacy abstraction.
## Verification
- `cargo check -p codex-config -p codex-core -p codex-sandboxing -p
codex-app-server -p codex-cli -p codex-tui`
- `cargo test -p codex-tui
permissions_selection_history_snapshot_full_access_to_default --
--nocapture`
- `cargo test -p codex-tui
permissions_selection_sends_approvals_reviewer_in_override_turn_context
-- --nocapture`
- `bazel test //codex-rs/tui:tui-unit-tests-bin
--test_arg=permissions_selection_history_snapshot_full_access_to_default
--test_output=errors`
- `bazel test //codex-rs/tui:tui-unit-tests-bin
--test_arg=permissions_selection_sends_approvals_reviewer_in_override_turn_context
--test_output=errors`
---
[//]: # (BEGIN SAPLING FOOTER)
Stack created with [Sapling](https://sapling-scm.com). Best reviewed
with [ReviewStack](https://reviewstack.dev/openai/codex/pull/19734).
* #19737
* #19736
* #19735
* __->__ #19734
# Why
Requirements support host-specific
`remote_sandbox_config.hostname_patterns`, but config loading previously
resolved and passed the system hostname through every config-loading
path even when no requirements layer used `remote_sandbox_config`. On
machines where hostname lookup is slow, startup and app-server config
reads paid for a feature that was not active.
We only need the hostname when a requirements layer actually declares
`remote_sandbox_config`, so this moves hostname resolution to the single
requirements merge point and keeps all other config callers unaware of
hostname matching.
# What
- Removed the eager `host_name` plumbing from
`load_config_layers_state`, `load_requirements_toml`, `ConfigBuilder`,
app-server `ConfigManager`, network proxy loading, and related call
sites.
- Resolve the hostname inside
`merge_requirements_with_remote_sandbox_config` only when the incoming
requirements contain `remote_sandbox_config`.
## Why
Several execution paths still converted profile-backed permissions into
`SandboxPolicy` and then rebuilt runtime permissions from that legacy
shape. Those round trips are unnecessary after the preceding PRs and can
lose split filesystem semantics. Core approval and escalation should
carry the resolved profile directly.
## What Changed
- Removes `sandbox_policy` from `ResolvedPermissionProfile`; the
resolved permission object now carries the canonical `PermissionProfile`
directly.
- Updates exec-policy fallback, shell/unified-exec interception,
escalation reruns, and related tests to pass profiles instead of legacy
policies.
- Removes legacy additional-permission merge helpers that built an
effective `SandboxPolicy` before rebuilding runtime permissions.
- Keeps legacy projections only at compatibility boundaries that still
require `SandboxPolicy`, not in core permission computation.
## Verification
- `cargo test -p codex-core direct_write_roots`
- `cargo test -p codex-core runtime_roots_to_legacy_projection`
- `cargo test -p codex-app-server
requested_permissions_trust_project_uses_permission_profile_intent`
---
[//]: # (BEGIN SAPLING FOOTER)
Stack created with [Sapling](https://sapling-scm.com). Best reviewed
with [ReviewStack](https://reviewstack.dev/openai/codex/pull/19394).
* #19737
* #19736
* #19735
* #19734
* #19395
* __->__ #19394
## Why
Runtime decisions should not infer permissions from the lossy legacy
sandbox projection once `PermissionProfile` is available. In particular,
`Disabled` and `External` need to remain distinct, and managed profiles
with split filesystem or deny-read rules should not be collapsed before
approval, network, safety, or analytics code makes decisions.
## What Changed
- Changes managed network proxy setup and network approval logic to use
`PermissionProfile` when deciding whether a managed sandbox is active.
- Migrates patch safety, Guardian/user-shell approval paths, Landlock
helper setup, analytics sandbox classification, and selected
turn/session code to profile-backed permissions.
- Validates command-level profile overrides against the constrained
`PermissionProfile` rather than a strict `SandboxPolicy` round trip.
- Preserves configured deny-read restrictions when command profiles are
narrowed.
- Adds coverage for profile-backed trust, network proxy/approval
behavior, patch safety, analytics classification, and command-profile
narrowing.
## Verification
- `cargo test -p codex-core direct_write_roots`
- `cargo test -p codex-core runtime_roots_to_legacy_projection`
- `cargo test -p codex-app-server
requested_permissions_trust_project_uses_permission_profile_intent`
---
[//]: # (BEGIN SAPLING FOOTER)
Stack created with [Sapling](https://sapling-scm.com). Best reviewed
with [ReviewStack](https://reviewstack.dev/openai/codex/pull/19393).
* #19395
* #19394
* __->__ #19393
## Why
Config loading had become split across crates: `codex-config` owned the
config types and merge logic, while `codex-core` still owned the loader
that assembled the layer stack. This change consolidates that
responsibility in `codex-config`, so the crate that defines config
behavior also owns how configs are discovered and loaded.
To make that move possible without reintroducing the old dependency
cycle, the shell-environment policy types and helpers that
`codex-exec-server` needs now live in `codex-protocol` instead of
flowing through `codex-config`.
This also makes the migrated loader tests more deterministic on machines
that already have managed or system Codex config installed by letting
tests override the system config and requirements paths instead of
reading the host's `/etc/codex`.
## What Changed
- moved the config loader implementation from `codex-core` into
`codex-config::loader` and deleted the old `core::config_loader` module
instead of leaving a compatibility shim
- moved shell-environment policy types and helpers into
`codex-protocol`, then updated `codex-exec-server` and other downstream
crates to import them from their new home
- updated downstream callers to use loader/config APIs from
`codex-config`
- added test-only loader overrides for system config and requirements
paths so loader-focused tests do not depend on host-managed config state
- cleaned up now-unused dependency entries and platform-specific cfgs
that were surfaced by post-push CI
## Testing
- `cargo test -p codex-config`
- `cargo test -p codex-core config_loader_tests::`
- `cargo test -p codex-protocol -p codex-exec-server -p
codex-cloud-requirements -p codex-rmcp-client --lib`
- `cargo test --lib -p codex-app-server-client -p codex-exec`
- `cargo test --no-run --lib -p codex-app-server`
- `cargo test -p codex-linux-sandbox --lib`
- `cargo shear`
- `just bazel-lock-check`
## Notes
- I did not chase unrelated full-suite failures outside the migrated
loader surface.
- `cargo test -p codex-core --lib` still hits unrelated proxy-sensitive
failures on this machine, and Windows CI still shows unrelated
long-running/timeouting test noise outside the loader migration itself.
## Why
After #19391, `PermissionProfile` and the split filesystem/network
policies could still be stored in parallel. That creates drift risk: a
profile can preserve deny globs, external enforcement, or split
filesystem entries while a cached projection silently loses those
details. This PR makes the profile the runtime source and derives
compatibility views from it.
## What Changed
- Removes stored filesystem/network sandbox projections from
`Permissions` and `SessionConfiguration`; their accessors now derive
from the canonical `PermissionProfile`.
- Derives legacy `SandboxPolicy` snapshots from profiles only where an
older API still needs that field.
- Updates MCP connection and elicitation state to track
`PermissionProfile` instead of `SandboxPolicy` for auto-approval
decisions.
- Adds semantic filesystem-policy comparison so cwd changes can preserve
richer profiles while still recognizing equivalent legacy projections
independent of entry ordering.
- Updates config/session tests to assert profile-derived projections
instead of parallel stored fields.
## Verification
- `cargo test -p codex-core direct_write_roots`
- `cargo test -p codex-core runtime_roots_to_legacy_projection`
- `cargo test -p codex-app-server
requested_permissions_trust_project_uses_permission_profile_intent`
---
[//]: # (BEGIN SAPLING FOOTER)
Stack created with [Sapling](https://sapling-scm.com). Best reviewed
with [ReviewStack](https://reviewstack.dev/openai/codex/pull/19392).
* #19395
* #19394
* #19393
* __->__ #19392
## Why
This supersedes #19391. During stack repair, GitHub marked #19391 as
merged into a temporary stack branch rather than into `main`, so the
runtime-config change needed a fresh PR.
`PermissionProfile` is now the canonical permissions shape after #19231
because it can distinguish `Managed`, `Disabled`, and `External`
enforcement while also carrying filesystem rules that legacy
`SandboxPolicy` cannot represent cleanly. Core config and session state
still needed to accept profile-backed permissions without forcing every
profile through the strict legacy bridge, which rejected valid runtime
profiles such as direct write roots.
The unrelated CI/test hardening that previously rode along with this PR
has been split into #19683 so this PR stays focused on the permissions
model migration.
## What Changed
- Adds `Permissions.permission_profile` and
`SessionConfiguration.permission_profile` as constrained runtime state,
while keeping `sandbox_policy` as a legacy compatibility projection.
- Introduces profile setters that keep `PermissionProfile`, split
filesystem/network policies, and legacy `SandboxPolicy` projections
synchronized.
- Uses a compatibility projection for requirement checks and legacy
consumers instead of rejecting profiles that cannot round-trip through
`SandboxPolicy` exactly.
- Updates config loading, config overrides, session updates, turn
context plumbing, prompt permission text, sandbox tags, and exec request
construction to carry profile-backed runtime permissions.
- Preserves configured deny-read entries and `glob_scan_max_depth` when
command/session profiles are narrowed.
- Adds `PermissionProfile::read_only()` and
`PermissionProfile::workspace_write()` presets that match legacy
defaults.
## Verification
- `cargo test -p codex-core direct_write_roots`
- `cargo test -p codex-core runtime_roots_to_legacy_projection`
- `cargo test -p codex-app-server
requested_permissions_trust_project_uses_permission_profile_intent`
---
[//]: # (BEGIN SAPLING FOOTER)
Stack created with [Sapling](https://sapling-scm.com). Best reviewed
with [ReviewStack](https://reviewstack.dev/openai/codex/pull/19606).
* #19395
* #19394
* #19393
* #19392
* __->__ #19606
Some providers of Responses API forward a model-defined `end_turn`
boolean indicating explicitly the model's indication of whether it would
like to end the turn or to be inferenced again. In this PR, we update
the sampling loop to use this field correctly if it's set. If the field
is not set by the provider, we fall back to the existing sampling logic.
Adds the core runtime behavior for active goals on top of the model
tools from PR 3.
## Why
A long-running goal should be a core runtime concern, not something
every client has to implement. Core owns the turn lifecycle, tool
completion boundaries, interruptions, resume behavior, and token usage,
so it is the right place to account progress, enforce budgets, and
decide when to continue work.
## What changed
- Centralized goal lifecycle side effects behind
`Session::goal_runtime_apply(GoalRuntimeEvent::...)`.
- Starts goal continuation turns only when the session is idle; pending
user input and mailbox work take priority.
- Accounts token and wall-clock usage at turn, tool, mutation,
interrupt, and resume boundaries; `get_thread_goal` remains read-only.
- Preserves sub-second wall-clock remainder across accounting boundaries
so long-running goals do not drift downward over time.
- Treats token budget exhaustion as a soft stop by marking the goal
`budget_limited` and injecting wrap-up steering instead of aborting the
active turn.
- Suppresses budget steering when `update_goal` marks a goal complete.
- Pauses active goals on interrupt and auto-reactivates paused goals
when a thread resumes outside plan mode.
- Suppresses repeated automatic continuation when a continuation turn
makes no tool calls.
- Added continuation and budget-limit prompt templates.
## Verification
- Added focused core coverage for continuation scheduling, accounting
boundaries, budget-limit steering, completion accounting, interrupt
pause behavior, resume auto-activation, and wall-clock remainder
accounting.
Adds the model-facing goal tools on top of the app-server API from PR 2.
## Why
Once goals are persisted and exposed to clients, the model needs a
small, constrained tool surface for goal workflows. The tool contract
should let the model inspect goals, create them only when explicitly
requested, and mark them complete without giving it broad control over
user/runtime-owned state.
## What changed
- Added `get_goal`, `create_goal`, and `update_goal` tool specs behind
the `goals` feature flag.
- Added core goal tool handlers that validate objectives and token
budgets before mutating persisted state.
- Constrained `create_goal` to create only when no goal exists, with
optional `token_budget` only when a budget is explicitly provided.
- Tightened the `create_goal` instructions so the model does not infer
goals from ordinary task requests.
- Constrained `update_goal` to expose only goal completion; pause,
resume, clear, and budget-limited transitions remain user- or
runtime-controlled.
- Registered the goal tools in the tool registry and kept them out of
review contexts where they should not appear.
## Verification
- Added tool-registry coverage for feature gating and tool availability.
- Added core session tests for create/get/update behavior, duplicate
goal rejection, budget validation, and completion-only updates.
Adds the app-server v2 goal API on top of the persisted goal state from
PR 1.
## Why
Clients need a stable app-server surface for reading and controlling
materialized thread goals before the model tools and TUI can use them.
Goal changes also need to be observable by app-server clients, including
clients that resume an existing thread.
## What changed
- Added v2 `thread/goal/get`, `thread/goal/set`, and `thread/goal/clear`
RPCs for materialized threads.
- Added `thread/goal/updated` and `thread/goal/cleared` notifications so
clients can keep local goal state in sync.
- Added resume/snapshot wiring so reconnecting clients see the current
goal state for a thread.
- Added app-server handlers that reconcile persisted rollout state
before direct goal mutations.
- Updated the app-server README plus generated JSON and TypeScript
schema fixtures for the new API surface.
## Verification
- Added app-server v2 coverage for goal get/set/clear behavior,
notification emission, resume snapshots, and non-local thread-store
interactions.
## Why
`ReadOnlyAccess` was a transitional legacy shape on `SandboxPolicy`:
`FullAccess` meant the historical read-only/workspace-write modes could
read the full filesystem, while `Restricted` tried to carry partial
readable roots. The partial-read model now belongs in
`FileSystemSandboxPolicy` and `PermissionProfile`, so keeping it on
`SandboxPolicy` makes every legacy projection reintroduce lossy
read-root bookkeeping and creates unnecessary noise in the rest of the
permissions migration.
This PR makes the legacy policy model narrower and explicit:
`SandboxPolicy::ReadOnly` and `SandboxPolicy::WorkspaceWrite` represent
the old full-read sandbox modes only. Split readable roots, deny-read
globs, and platform-default/minimal read behavior stay in the runtime
permissions model.
## What changed
- Removes `ReadOnlyAccess` from
`codex_protocol::protocol::SandboxPolicy`, including the generated
`access` and `readOnlyAccess` API fields.
- Updates legacy policy/profile conversions so restricted filesystem
reads are represented only by `FileSystemSandboxPolicy` /
`PermissionProfile` entries.
- Keeps app-server v2 compatible with legacy `fullAccess` read-access
payloads by accepting and ignoring that no-op shape, while rejecting
legacy `restricted` read-access payloads instead of silently widening
them to full-read legacy policies.
- Carries Windows sandbox platform-default read behavior with an
explicit override flag instead of depending on
`ReadOnlyAccess::Restricted`.
- Refreshes generated app-server schema/types and updates tests/docs for
the simplified legacy policy shape.
## Verification
- `cargo check -p codex-app-server-protocol --tests`
- `cargo check -p codex-windows-sandbox --tests`
- `cargo test -p codex-app-server-protocol sandbox_policy_`
---
[//]: # (BEGIN SAPLING FOOTER)
Stack created with [Sapling](https://sapling-scm.com). Best reviewed
with [ReviewStack](https://reviewstack.dev/openai/codex/pull/19449).
* #19395
* #19394
* #19393
* #19392
* #19391
* __->__ #19449
- Route cold thread/resume and thread/fork source loading through
ThreadStore reads instead of direct rollout path operations
- Keep lookups that explicitly specify a rollout-path using the local
thread store methods but return an invalid-request error for remote
ThreadStore configurations
- Add some additional unit tests for code path coverage
## Why
The profile conversion path still required a `cwd` even when it was only
translating a legacy `SandboxPolicy` into a `PermissionProfile`. That
made profile producers invent an ambient `cwd`, which is exactly the
anchoring we are trying to remove from permission-profile data. A legacy
workspace-write policy can be represented symbolically instead: `:cwd =
write` plus read-only `:project_roots` metadata subpaths.
This PR creates that cwd-free base so the rest of the stack can stop
threading cwd through profile construction. Callers that actually need a
concrete runtime filesystem policy for a specific cwd still have an
explicitly named cwd-bound conversion.
## What Changed
- `PermissionProfile::from_legacy_sandbox_policy` now takes only
`&SandboxPolicy`.
- `FileSystemSandboxPolicy::from_legacy_sandbox_policy` is now the
symbolic, cwd-free projection for profiles.
- The old concrete projection is retained as
`FileSystemSandboxPolicy::from_legacy_sandbox_policy_for_cwd` for
runtime/boundary code that must materialize legacy cwd behavior.
- Workspace-write profiles preserve `CurrentWorkingDirectory` and
`ProjectRoots` special entries instead of materializing cwd into
absolute paths.
## Verification
- `cargo check -p codex-protocol -p codex-core -p
codex-app-server-protocol -p codex-app-server -p codex-exec -p
codex-exec-server -p codex-tui -p codex-sandboxing -p
codex-linux-sandbox -p codex-analytics --tests`
- `just fix -p codex-protocol -p codex-core -p codex-app-server-protocol
-p codex-app-server -p codex-exec -p codex-exec-server -p codex-tui -p
codex-sandboxing -p codex-linux-sandbox -p codex-analytics`
---
[//]: # (BEGIN SAPLING FOOTER)
Stack created with [Sapling](https://sapling-scm.com). Best reviewed
with [ReviewStack](https://reviewstack.dev/openai/codex/pull/19414).
* #19395
* #19394
* #19393
* #19392
* #19391
* __->__ #19414
## Summary
- Thread `agent_max_threads` into `ToolsConfig` and
`SpawnAgentToolOptions`.
- Render the configured `max_concurrent_threads_per_session` value in
the MultiAgentV2 `spawn_agent` description.
- Cover the description text in `codex-tools` unit tests and
`codex-core` tool spec tests.
## Validation
- `just fmt`
- `cargo test -p codex-tools`
- `cargo test -p codex-core spawn_agent_description`
- `git diff --check`
## Notes
- `cargo test -p codex-core` was also attempted, but unrelated
environment-sensitive tests failed with the active local environment.
Examples: approvals reviewer defaults observed `AutoReview` instead of
`User`, request-permissions event tests did not emit events, and
proxy-env tests saw `http://127.0.0.1:50604` from the active proxy
environment.
Co-authored-by: Codex <noreply@openai.com>
