## Summary
Fixes#15189.
Custom model providers that set `requires_openai_auth = false` could
only use static credentials via `env_key` or
`experimental_bearer_token`. That is not enough for providers that mint
short-lived bearer tokens, because Codex had no way to run a command to
obtain a bearer token, cache it briefly in memory, and retry with a
refreshed token after a `401`.
This PR adds that provider config and wires it through the existing auth
design: request paths still go through `AuthManager.auth()` and
`UnauthorizedRecovery`, with `core` only choosing when to use a
provider-backed bearer-only `AuthManager`.
## Scope
To keep this PR reviewable, `/models` only uses provider auth for the
initial request in this change. It does **not** add a dedicated `401`
retry path for `/models`; that can be follow-up work if we still need it
after landing the main provider-token support.
## Example Usage
```toml
model_provider = "corp-openai"
[model_providers.corp-openai]
name = "Corp OpenAI"
base_url = "https://gateway.example.com/openai"
requires_openai_auth = false
[model_providers.corp-openai.auth]
command = "gcloud"
args = ["auth", "print-access-token"]
timeout_ms = 5000
refresh_interval_ms = 300000
```
The command contract is intentionally small:
- write the bearer token to `stdout`
- exit `0`
- any leading or trailing whitespace is trimmed before the token is used
## What Changed
- add `model_providers.<id>.auth` to the config model and generated
schema
- validate that command-backed provider auth is mutually exclusive with
`env_key`, `experimental_bearer_token`, and `requires_openai_auth`
- build a bearer-only `AuthManager` for `ModelClient` and
`ModelsManager` when a provider configures `auth`
- let normal Responses requests and realtime websocket connects use the
provider-backed bearer source through the same `AuthManager.auth()` path
- allow `/models` online refresh for command-auth providers and attach
the provider token to the initial `/models` request
- keep `auth.cwd` available as an advanced escape hatch and include it
in the generated config schema
## Testing
- `cargo test -p codex-core provider_auth_command`
- `cargo test -p codex-core
refresh_available_models_uses_provider_auth_token`
- `cargo test -p codex-core
test_deserialize_provider_auth_config_defaults`
## Docs
- `developers.openai.com/codex` should document the new
`[model_providers.<id>.auth]` block and the token-command contract
## Why
The Bazel-backed `argument-comment-lint` CI path had two gaps:
- Bazel wildcard target expansion skipped inline unit-test crates from
`src/` modules because the generated `*-unit-tests-bin` `rust_test`
targets are tagged `manual`.
- `argument-comment-mismatch` was still only a warning in the Bazel and
packaged-wrapper entrypoints, so a typoed `/*param_name*/` comment could
still pass CI even when the lint detected it.
That left CI blind to real linux-sandbox examples, including the missing
`/*local_port*/` comment in
`codex-rs/linux-sandbox/src/proxy_routing.rs` and typoed argument
comments in `codex-rs/linux-sandbox/src/landlock.rs`.
## What Changed
- Added `tools/argument-comment-lint/list-bazel-targets.sh` so Bazel
lint runs cover `//codex-rs/...` plus the manual `rust_test`
`*-unit-tests-bin` targets.
- Updated `just argument-comment-lint`, `rust-ci.yml`, and
`rust-ci-full.yml` to use that helper.
- Promoted both `argument-comment-mismatch` and
`uncommented-anonymous-literal-argument` to errors in every strict
entrypoint:
- `tools/argument-comment-lint/lint_aspect.bzl`
- `tools/argument-comment-lint/src/bin/argument-comment-lint.rs`
- `tools/argument-comment-lint/wrapper_common.py`
- Added wrapper/bin coverage for the stricter lint flags and documented
the behavior in `tools/argument-comment-lint/README.md`.
- Fixed the now-covered callsites in
`codex-rs/linux-sandbox/src/proxy_routing.rs`,
`codex-rs/linux-sandbox/src/landlock.rs`, and
`codex-rs/core/src/shell_snapshot_tests.rs`.
This keeps the Bazel target expansion narrow while making the Bazel and
prebuilt-linter paths enforce the same strict lint set.
## Verification
- `python3 -m unittest discover -s tools/argument-comment-lint -p
'test_*.py'`
- `cargo +nightly-2025-09-18 test --manifest-path
tools/argument-comment-lint/Cargo.toml`
- `just argument-comment-lint`
## Why
`#16193` moved the pure `tool_search` and `tool_suggest` spec builders
into `codex-tools`, but `codex-core` still owned the shared
discoverable-tool model that those builders and the `tool_suggest`
runtime both depend on. This change continues the migration by moving
that reusable model boundary out of `codex-core` as well, so the
discovery/suggestion stack uses one shared set of types and
`core/src/tools` no longer needs its own `discoverable.rs` module.
## What changed
- Moved `DiscoverableTool`, `DiscoverablePluginInfo`, and
`filter_tool_suggest_discoverable_tools_for_client()` into
`codex-rs/tools/src/tool_discovery.rs` alongside the extracted
discovery/suggestion spec builders.
- Added `codex-app-server-protocol` as a `codex-tools` dependency so the
shared discoverable-tool model can own the connector-side `AppInfo`
variant directly.
- Updated `core/src/tools/handlers/tool_suggest.rs`,
`core/src/tools/spec.rs`, `core/src/tools/router.rs`,
`core/src/connectors.rs`, and `core/src/codex.rs` to consume the shared
`codex-tools` model instead of the old core-local declarations.
- Changed `core/src/plugins/discoverable.rs` to return
`DiscoverablePluginInfo` directly, moved the pure client-filter coverage
into `tool_discovery_tests.rs`, and deleted the old
`core/src/tools/discoverable.rs` module.
- Updated `codex-rs/tools/README.md` so the crate boundary documents
that `codex-tools` now owns the discoverable-tool models in addition to
the discovery/suggestion spec builders.
## Test plan
- `cargo test -p codex-tools`
- `CARGO_TARGET_DIR=/tmp/codex-core-discoverable-model cargo test -p
codex-core --lib tools::handlers::tool_suggest::`
- `CARGO_TARGET_DIR=/tmp/codex-core-discoverable-model cargo test -p
codex-core --lib tools::spec::`
- `CARGO_TARGET_DIR=/tmp/codex-core-discoverable-model cargo test -p
codex-core --lib plugins::discoverable::`
- `just bazel-lock-check`
- `just argument-comment-lint`
## References
- #16193
- #16154
- #15923
- #15928
- #15944
- #15953
- #16031
- #16047
- #16129
- #16132
- #16138
- #16141
## Why
`core/src/tools/spec.rs` still owned the pure `tool_search` and
`tool_suggest` spec builders even though that logic no longer needed
`codex-core` runtime state. This change continues the `codex-tools`
migration by moving the reusable discovery and suggestion spec
construction out of `codex-core` so `spec.rs` is left with the
core-owned policy decisions about when these tools are exposed and what
metadata is available.
## What changed
- Added `codex-rs/tools/src/tool_discovery.rs` with the shared
`tool_search` and `tool_suggest` spec builders, plus focused unit tests
in `tool_discovery_tests.rs`.
- Moved the shared `DiscoverableToolAction` and `DiscoverableToolType`
declarations into `codex-tools` so the `tool_suggest` handler and the
extracted spec builders use the same wire-model enums.
- Updated `core/src/tools/spec.rs` to translate `ToolInfo` and
`DiscoverableTool` values into neutral `codex-tools` inputs and delegate
the actual spec building there.
- Removed the old template-based description rendering helpers from
`core/src/tools/spec.rs` and deleted the now-dead helper methods in
`core/src/tools/discoverable.rs`.
- Updated `codex-rs/tools/README.md` to document that discovery and
suggestion models/spec builders now live in `codex-tools`.
## Test plan
- `cargo test -p codex-tools`
- `CARGO_TARGET_DIR=/tmp/codex-core-discovery-specs cargo test -p
codex-core --lib tools::spec::`
- `CARGO_TARGET_DIR=/tmp/codex-core-discovery-specs cargo test -p
codex-core --lib tools::handlers::tool_suggest::`
- `just argument-comment-lint`
## References
- #16154
- #15923
- #15928
- #15944
- #15953
- #16031
- #16047
- #16129
- #16132
- #16138
- #16141
## Why
The previous `codex-tools` migration steps moved the shared schema
models, local-host specs, collaboration specs, and related adapters out
of `codex-core`, but `core/src/tools/spec.rs` still contained a grab bag
of pure utility tool builders. Those specs do not need session state or
handler logic; they only describe wire shapes for tools that
`codex-core` already knows how to execute.
Moving that remaining low-coupling layer into `codex-tools` keeps the
migration moving in meaningful chunks and trims another large block of
passive tool-spec construction out of `codex-core` without touching the
runtime-coupled handlers.
## What changed
- extended `codex-tools` to own the pure spec builders for:
- code-mode `exec` / `wait`
- `js_repl` / `js_repl_reset`
- MCP resource tools `list_mcp_resources`,
`list_mcp_resource_templates`, and `read_mcp_resource`
- utility tools `list_dir` and `test_sync_tool`
- split those builders across small module files with sibling
`*_tests.rs` coverage, keeping `src/lib.rs` exports-only
- rewired `core/src/tools/spec.rs` to call the extracted builders and
deleted the duplicated core-local implementations
- moved the direct JS REPL grammar seam test out of
`core/src/tools/spec_tests.rs` so it now lives with the extracted
implementation in `codex-tools`
- updated `codex-rs/tools/README.md` so the documented crate boundary
matches the new utility-spec surface
## Test plan
- `CARGO_TARGET_DIR=/tmp/codex-tools-utility-specs cargo test -p
codex-tools`
- `CARGO_TARGET_DIR=/tmp/codex-core-utility-specs cargo test -p
codex-core --lib tools::spec::`
- `just fix -p codex-tools -p codex-core`
- `just argument-comment-lint`
## References
- #15923
- #15928
- #15944
- #15953
- #16031
- #16047
- #16129
- #16132
- #16138
- #16141
## Why
The recent `codex-tools` migration steps have moved shared tool models
and low-coupling spec helpers out of `codex-core`, but
`core/src/tools/spec.rs` still owned a large block of pure
collaboration-tool spec construction. Those builders do not need session
state or runtime behavior; they only need a small amount of core-owned
configuration injected at the seam.
Moving that cohesive slice into `codex-tools` makes the crate boundary
more honest and removes a substantial amount of passive tool-spec logic
from `codex-core` without trying to move the runtime-coupled multi-agent
handlers at the same time.
## What changed
- added `agent_tool.rs`, `request_user_input_tool.rs`, and
`agent_job_tool.rs` to `codex-tools`, with sibling `*_tests.rs` coverage
and an exports-only `lib.rs`
- moved the pure `ToolSpec` builders for:
- collaboration tools such as `spawn_agent`, `send_input`,
`send_message`, `assign_task`, `resume_agent`, `wait_agent`,
`list_agents`, and `close_agent`
- `request_user_input`
- agent-job specs `spawn_agents_on_csv` and `report_agent_job_result`
- rewired `core/src/tools/spec.rs` to call the extracted builders while
still supplying the core-owned inputs, such as spawn-agent role
descriptions and wait timeout bounds
- updated the `core/src/tools/spec.rs` seam tests to build expected
collaboration specs through `codex-tools`
- updated `codex-rs/tools/README.md` so the crate documentation reflects
the broader collaboration-tool boundary
## Test plan
- `CARGO_TARGET_DIR=/tmp/codex-tools-collab-specs cargo test -p
codex-tools`
- `CARGO_TARGET_DIR=/tmp/codex-core-collab-specs cargo test -p
codex-core --lib tools::spec::`
- `just fix -p codex-tools -p codex-core`
- `just argument-comment-lint`
## References
- #15923
- #15928
- #15944
- #15953
- #16031
- #16047
- #16129
- #16132
- #16138
## Why
`core/src/tools/spec.rs` still bundled a set of pure local-host tool
builders with the orchestration that actually decides when those tools
are exposed and which handlers back them. That made `codex-core`
responsible for JSON/tool-shape construction that does not depend on
session state, and it kept the `codex-tools` migration from taking a
meaningfully larger bite out of `spec.rs`.
This PR moves that reusable spec-building layer into `codex-tools` while
leaving feature gating, handler registration, and runtime-coupled
descriptions in `codex-core`.
## What changed
- added `codex-rs/tools/src/local_tool.rs` for the pure builders for
`exec_command`, `write_stdin`, `shell`, `shell_command`, and
`request_permissions`
- added `codex-rs/tools/src/view_image.rs` for the `view_image` tool
spec and output schema so the extracted modules stay right-sized
- rewired `codex-rs/core/src/tools/spec.rs` to call those extracted
builders instead of constructing these specs inline
- kept the `request_permissions` description source in `codex-core`,
with `codex-tools` taking the description as input so the crate boundary
does not grow a dependency on handler/runtime code
- moved the direct constructor coverage for this slice from
`codex-rs/core/src/tools/spec_tests.rs` into
`codex-rs/tools/src/local_tool_tests.rs` and
`codex-rs/tools/src/view_image_tests.rs`
- updated `codex-rs/tools/README.md` to reflect that `codex-tools` now
owns this local-host spec layer
## Test plan
- `CARGO_TARGET_DIR=/tmp/codex-tools-local-host cargo test -p
codex-tools`
- `CARGO_TARGET_DIR=/tmp/codex-core-local-tools cargo test -p codex-core
--lib tools::spec::`
- `just argument-comment-lint`
## References
- #15923
- #15928
- #15944
- #15953
- #16031
- #16047
- #16129
- #16132
## Why
The longer-term `codex-tools` migration is to move pure tool-definition
and tool-spec plumbing out of `codex-core` while leaving session- and
runtime-coupled orchestration behind.
The remaining code-mode adapter layer in
`core/src/tools/code_mode_description.rs` was a good next extraction
seam because it only transformed `ToolSpec` values for code mode and
already delegated the low-level description rendering to
`codex-code-mode`.
## What Changed
- added `codex-rs/tools/src/code_mode.rs` with
`augment_tool_spec_for_code_mode()` and
`tool_spec_to_code_mode_tool_definition()`
- added focused unit coverage in `codex-rs/tools/src/code_mode_tests.rs`
- rewired `core/src/tools/spec.rs` and `core/src/tools/code_mode/mod.rs`
to use the extracted adapters from `codex-tools`
- removed the old `core/src/tools/code_mode_description.rs` shim and its
test file from `codex-core`
- added the `codex-code-mode` dependency to `codex-tools`, updated
`Cargo.lock`, and refreshed the `codex-tools` README to reflect the
expanded boundary
## Test Plan
- `cargo test -p codex-tools`
- `CARGO_TARGET_DIR=/tmp/codex-core-code-mode-adapters cargo test -p
codex-core --lib tools::spec::`
- `CARGO_TARGET_DIR=/tmp/codex-core-code-mode-adapters cargo test -p
codex-core --lib tools::code_mode::`
- `just bazel-lock-update`
- `just bazel-lock-check`
- `just argument-comment-lint`
## References
- #15923
- #15928
- #15944
- #15953
- #16031
- #16047
- #16129
## Why
The `plugin/list` force-sync path can race app-server startup's curated
plugin cache refresh.
Startup was capturing the configured curated plugin IDs from the initial
config snapshot. If `plugin/list` with `forceRemoteSync` removed curated
plugin entries from `config.toml` while that background refresh was
still in flight, the startup task could recreate cache directories for
plugins that had just been uninstalled.
That leaves the `plugin/list` response logically correct but the on-disk
cache stale, which matches the flaky Ubuntu arm failure seen in
`codex-app-server::all
suite::v2::plugin_list::plugin_list_force_remote_sync_reconciles_curated_plugin_state`
while validating [#16047](https://github.com/openai/codex/pull/16047).
## What
- change `codex-rs/core/src/plugins/manager.rs` so startup curated-repo
refresh rereads the current user `config.toml` before deciding which
curated plugin cache entries to refresh
- factor the configured-plugin parsing so the same logic can be reused
from either the config layer stack or the persisted user config value
- add a regression test that verifies curated plugin IDs are read from
the latest user config state before cache refresh runs
## Testing
- `cargo test -p codex-core
configured_curated_plugin_ids_from_codex_home_reads_latest_user_config
-- --nocapture`
- `cargo test -p codex-app-server
suite::v2::plugin_list::plugin_list_force_remote_sync_reconciles_curated_plugin_state
-- --nocapture`
- `just argument-comment-lint`
## Why
This continues the `codex-tools` migration by moving another passive
tool-spec layer out of `codex-core`.
After `ToolSpec` moved into `codex-tools`, `codex-core` still owned
`ConfiguredToolSpec` and `create_tools_json_for_responses_api()`. Both
are data-model and serialization helpers rather than runtime
orchestration, so keeping them in `core/src/tools/registry.rs` and
`core/src/tools/spec.rs` left passive tool-definition code coupled to
`codex-core` longer than necessary.
## What changed
- moved `ConfiguredToolSpec` into `codex-rs/tools/src/tool_spec.rs`
- moved `create_tools_json_for_responses_api()` into
`codex-rs/tools/src/tool_spec.rs`
- re-exported the new surface from `codex-rs/tools/src/lib.rs`, which
remains exports-only
- updated `core/src/client.rs`, `core/src/tools/registry.rs`, and
`core/src/tools/router.rs` to consume the extracted types and serializer
from `codex-tools`
- moved the tool-list serialization test into
`codex-rs/tools/src/tool_spec_tests.rs`
- added focused unit coverage for `ConfiguredToolSpec::name()`
- simplified `core/src/tools/spec_tests.rs` to use the extracted
`ConfiguredToolSpec::name()` directly and removed the now-redundant
local `tool_name()` helper
- updated `codex-rs/tools/README.md` so the crate boundary reflects the
newly extracted tool-spec wrapper and serialization helper
## Test plan
- `cargo test -p codex-tools`
- `CARGO_TARGET_DIR=/tmp/codex-core-configured-spec cargo test -p
codex-core --lib tools::spec::`
- `CARGO_TARGET_DIR=/tmp/codex-core-configured-spec cargo test -p
codex-core --lib client::`
- `just fix -p codex-tools -p codex-core`
- `just argument-comment-lint`
## References
- #15923
- #15928
- #15944
- #15953
- #16031
- #16047
## Why
This continues the `codex-tools` migration by moving another passive
tool-definition layer out of `codex-core`.
After `ResponsesApiTool` and the lower-level schema adapters moved into
`codex-tools`, `core/src/client_common.rs` was still owning `ToolSpec`
and the web-search request wire types even though they are serialized
data models rather than runtime orchestration. Keeping those types in
`codex-core` makes the crate boundary look smaller than it really is and
leaves non-runtime tool-shape code coupled to core.
## What changed
- moved `ToolSpec`, `ResponsesApiWebSearchFilters`, and
`ResponsesApiWebSearchUserLocation` into
`codex-rs/tools/src/tool_spec.rs`
- added focused unit tests in `codex-rs/tools/src/tool_spec_tests.rs`
for:
- `ToolSpec::name()`
- web-search config conversions
- `ToolSpec` serialization for `web_search` and `tool_search`
- kept `codex-rs/tools/src/lib.rs` exports-only by re-exporting the new
module from `lib.rs`
- reduced `core/src/client_common.rs` to a compatibility shim that
re-exports the extracted tool-spec types for current core call sites
- updated `core/src/tools/spec_tests.rs` to consume the extracted
web-search types directly from `codex-tools`
- updated `codex-rs/tools/README.md` so the crate contract reflects that
`codex-tools` now owns the passive tool-spec request models in addition
to the lower-level Responses API structs
## Test plan
- `cargo test -p codex-tools`
- `cargo test -p codex-core --lib tools::spec::`
- `cargo test -p codex-core --lib client_common::`
- `just fix -p codex-tools -p codex-core`
- `just argument-comment-lint`
## References
- #15923
- #15928
- #15944
- #15953
- #16031
## Summary
- remove protocol and core support for discovering and listing custom
prompts
- simplify the TUI slash-command flow and command popup to built-in
commands only
- delete obsolete custom prompt tests, helpers, and docs references
- clean up downstream event handling for the removed protocol events
## Why
`argument-comment-lint` was green in CI even though the repo still had
many uncommented literal arguments. The main gap was target coverage:
the repo wrapper did not force Cargo to inspect test-only call sites, so
examples like the `latest_session_lookup_params(true, ...)` tests in
`codex-rs/tui_app_server/src/lib.rs` never entered the blocking CI path.
This change cleans up the existing backlog, makes the default repo lint
path cover all Cargo targets, and starts rolling that stricter CI
enforcement out on the platform where it is currently validated.
## What changed
- mechanically fixed existing `argument-comment-lint` violations across
the `codex-rs` workspace, including tests, examples, and benches
- updated `tools/argument-comment-lint/run-prebuilt-linter.sh` and
`tools/argument-comment-lint/run.sh` so non-`--fix` runs default to
`--all-targets` unless the caller explicitly narrows the target set
- fixed both wrappers so forwarded cargo arguments after `--` are
preserved with a single separator
- documented the new default behavior in
`tools/argument-comment-lint/README.md`
- updated `rust-ci` so the macOS lint lane keeps the plain wrapper
invocation and therefore enforces `--all-targets`, while Linux and
Windows temporarily pass `-- --lib --bins`
That temporary CI split keeps the stricter all-targets check where it is
already cleaned up, while leaving room to finish the remaining Linux-
and Windows-specific target-gated cleanup before enabling
`--all-targets` on those runners. The Linux and Windows failures on the
intermediate revision were caused by the wrapper forwarding bug, not by
additional lint findings in those lanes.
## Validation
- `bash -n tools/argument-comment-lint/run.sh`
- `bash -n tools/argument-comment-lint/run-prebuilt-linter.sh`
- shell-level wrapper forwarding check for `-- --lib --bins`
- shell-level wrapper forwarding check for `-- --tests`
- `just argument-comment-lint`
- `cargo test` in `tools/argument-comment-lint`
- `cargo test -p codex-terminal-detection`
## Follow-up
- Clean up remaining Linux-only target-gated callsites, then switch the
Linux lint lane back to the plain wrapper invocation.
- Clean up remaining Windows-only target-gated callsites, then switch
the Windows lint lane back to the plain wrapper invocation.
This is the part 1 of 2 PRs that will delete the `tui` /
`tui_app_server` split. This part simply deletes the existing `tui`
directory and marks the `tui_app_server` feature flag as removed. I left
the `tui_app_server` feature flag in place for now so its presence
doesn't result in an error. It is simply ignored.
Part 2 will rename the `tui_app_server` directory `tui`. I did this as
two parts to reduce visible code churn.
apply_patch sometimes provides additional parent dir as a writable root
when it is already writable. This is mostly a no-op on Mac/Linux but
causes actual ACL churn on Windows that is best avoided. We are also
seeing some actual failures with these ACLs in the wild, which I haven't
fully tracked down, but it's safe/best to avoid doing it altogether.
- [x] Auto / unspecified approval mode: read-only tools now skip before
guardian routing.
- [x] Approve / always-allow mode: read-only tools still skip, now via
the shared early return.
- [x] Prompt mode: read-only tools no longer skip; they continue to
approval.
## Why
The previous extraction steps moved shared tool-schema parsing into
`codex-tools`, but `codex-core` still owned the generic Responses API
tool models and the last adapter layer that turned parsed tool
definitions into `ResponsesApiTool` values.
That left `core/src/tools/spec.rs` and `core/src/client_common.rs`
holding a chunk of tool-shaping code that does not need session state,
runtime plumbing, or any other `codex-core`-specific dependency. As a
result, `codex-tools` owned the parsed tool definition, but `codex-core`
still owned the generic wire model that those definitions are converted
into.
This change moves that boundary one step further. `codex-tools` now owns
the reusable Responses/tool wire structs and the shared conversion
helpers for dynamic tools, MCP tools, and deferred MCP aliases.
`codex-core` continues to own `ToolSpec` orchestration and the remaining
web-search-specific request shapes.
## What changed
- added `tools/src/responses_api.rs` to own `ResponsesApiTool`,
`FreeformTool`, `ToolSearchOutputTool`, namespace output types, and the
shared `ToolDefinition -> ResponsesApiTool` adapter helpers
- added `tools/src/responses_api_tests.rs` for deferred-loading
behavior, adapter coverage, and namespace serialization coverage
- rewired `core/src/tools/spec.rs` to use the extracted dynamic/MCP
adapter helpers instead of defining those conversions locally
- rewired `core/src/tools/handlers/tool_search.rs` to use the extracted
deferred MCP adapter and namespace output types directly
- slimmed `core/src/client_common.rs` so it now keeps `ToolSpec` and the
web-search-specific wire types, while reusing the extracted tool models
from `codex-tools`
- moved the extracted seam tests out of `core` and updated
`codex-rs/tools/README.md` plus `tools/src/lib.rs` to reflect the
expanded `codex-tools` boundary
## Test plan
- `cargo test -p codex-tools`
- `cargo test -p codex-core --lib tools::spec::`
- `cargo test -p codex-core --lib tools::handlers::tool_search::`
- `just fix -p codex-tools -p codex-core`
- `just argument-comment-lint`
## References
- [#15923](https://github.com/openai/codex/pull/15923) `codex-tools:
extract shared tool schema parsing`
- [#15928](https://github.com/openai/codex/pull/15928) `codex-tools:
extract MCP schema adapters`
- [#15944](https://github.com/openai/codex/pull/15944) `codex-tools:
extract dynamic tool adapters`
- [#15953](https://github.com/openai/codex/pull/15953) `codex-tools:
introduce named tool definitions`
## Summary
- add `self_serve_business_usage_based` and `enterprise_cbp_usage_based`
to the public/internal plan enums and regenerate the app-server + Python
SDK artifacts
- map both plans through JWT login and backend rate-limit payloads, then
bucket them with the existing Team/Business entitlement behavior in
cloud requirements, usage-limit copy, tooltips, and status display
- keep the earlier display-label remap commit on this branch so the new
Team-like and Business-like plans render consistently in the UI
## Testing
- `just write-app-server-schema`
- `uv run --project sdk/python python
sdk/python/scripts/update_sdk_artifacts.py generate-types`
- `just fix -p codex-protocol -p codex-login -p codex-core -p
codex-backend-client -p codex-cloud-requirements -p codex-tui -p
codex-tui-app-server -p codex-backend-openapi-models`
- `just fmt`
- `just argument-comment-lint`
- `cargo test -p codex-protocol
usage_based_plan_types_use_expected_wire_names`
- `cargo test -p codex-login usage_based`
- `cargo test -p codex-backend-client usage_based`
- `cargo test -p codex-cloud-requirements usage_based`
- `cargo test -p codex-core usage_limit_reached_error_formats_`
- `cargo test -p codex-tui plan_type_display_name_remaps_display_labels`
- `cargo test -p codex-tui remapped`
- `cargo test -p codex-tui-app-server
plan_type_display_name_remaps_display_labels`
- `cargo test -p codex-tui-app-server remapped`
- `cargo test -p codex-tui-app-server
preserves_usage_based_plan_type_wire_name`
## Notes
- a broader multi-crate `cargo test` run still hits unrelated existing
guardian-approval config failures in
`codex-rs/core/src/config/config_tests.rs`
1. Keep curated plugin staging directories under TempDir ownership until
activation succeeds, so failed git/HTTP sync attempts do not leak
plugins-clone-*.
2. Best-effort clean up stale plugins-clone-* directories before
creating a new staged repo, using a conservative age threshold.
3. Emit OTEL counters for curated plugin startup sync transport attempts
and final outcome across git and HTTP paths.
#15999 introduced a Windows-only `\r\n` mismatch in review-exit template
handling. This PR normalizes those template newlines and separates that
fix from [#16014](https://github.com/openai/codex/pull/16014) so it can
be reviewed independently.
## Why
This continues the `codex-tools` migration by moving one more piece of
generic tool-definition bookkeeping out of `codex-core`.
The earlier extraction steps moved shared schema parsing into
`codex-tools`, but `core/src/tools/spec.rs` still had to supply tool
names separately and perform ad hoc rewrites for deferred MCP aliases.
That meant the crate boundary was still awkward: the parsed shape coming
back from `codex-tools` was missing part of the definition that
`codex-core` ultimately needs to assemble a `ResponsesApiTool`.
This change introduces a named `ToolDefinition` in `codex-tools` so both
MCP tools and dynamic tools cross the crate boundary in the same
reusable model. `codex-core` still owns the final `ResponsesApiTool`
assembly, but less of the generic tool-definition shaping logic stays
behind in `core`.
## What changed
- replaced `ParsedToolDefinition` with a named `ToolDefinition` in
`codex-rs/tools/src/tool_definition.rs`
- added `codex-rs/tools/src/tool_definition_tests.rs` for `renamed()`
and `into_deferred()`
- updated `parse_dynamic_tool()` and `parse_mcp_tool()` to return
`ToolDefinition`
- simplified `codex-rs/core/src/tools/spec.rs` so it adapts
`ToolDefinition` into `ResponsesApiTool` instead of rewriting names and
deferred fields inline
- updated parser tests and `codex-rs/tools/README.md` to reflect the
named tool-definition model
## Test plan
- `cargo test -p codex-tools`
- `cargo test -p codex-core --lib tools::spec::`
## Why
`codex-tools` already owned the shared JSON schema parser and the MCP
tool schema adapter, but `core/src/tools/spec.rs` still parsed dynamic
tools directly.
That left the tool-schema boundary split in two different ways:
- MCP tools flowed through `codex-tools`, while dynamic tools were still
parsed in `codex-core`
- the extracted dynamic-tool path initially introduced a
dynamic-specific parsed shape even though `codex-tools` already had very
similar MCP adapter output
This change finishes that extraction boundary in one step. `codex-core`
still owns `ResponsesApiTool` assembly, but both MCP tools and dynamic
tools now enter that layer through `codex-tools` using the same parsed
tool-definition shape.
## What changed
- added `tools/src/dynamic_tool.rs` and sibling
`tools/src/dynamic_tool_tests.rs`
- introduced `parse_dynamic_tool()` in `codex-tools` and switched
`core/src/tools/spec.rs` to use it for dynamic tools
- added `tools/src/parsed_tool_definition.rs` so both MCP and dynamic
adapters return the same `ParsedToolDefinition`
- updated `core/src/tools/spec.rs` to build `ResponsesApiTool` through a
shared local adapter helper instead of separate MCP and dynamic assembly
paths
- expanded `core/src/tools/spec_tests.rs` so the dynamic-tool adapter
test asserts the full converted `ResponsesApiTool`, including
`defer_loading`
- updated `codex-rs/tools/README.md` to reflect the shared parsed
tool-definition boundary
## Test plan
- `cargo test -p codex-tools`
- `cargo test -p codex-core --lib tools::spec::`
---
[//]: # (BEGIN SAPLING FOOTER)
Stack created with [Sapling](https://sapling-scm.com). Best reviewed
with [ReviewStack](https://reviewstack.dev/openai/codex/pull/15944).
* #15953
* __->__ #15944
## Summary
- split the joined `PATH` before running system `bwrap` lookup
- keep the existing workspace-local `bwrap` skip behavior intact
- add regression tests that exercise real multi-entry search paths
## Why
The PATH-based lookup added in #15791 still wrapped the raw `PATH`
environment value as a single `PathBuf` before passing it through
`join_paths()`. On Unix, a normal multi-entry `PATH` contains `:`, so
that wrapper path is invalid as one path element and the lookup returns
`None`.
That made Codex behave as if no system `bwrap` was installed even when
`bwrap` was available on `PATH`, which is what users in #15340 were
still hitting on `0.117.0-alpha.25`.
## Impact
System `bwrap` discovery now works with normal multi-entry `PATH` values
instead of silently falling back to the vendored binary.
Fixes#15340.
## Validation
- `just fmt`
- `cargo test -p codex-sandboxing`
- `cargo test -p codex-linux-sandbox`
- `just fix -p codex-sandboxing`
- `just argument-comment-lint`
## Summary
This PR replaces the legacy network allow/deny list model with explicit
rule maps for domains and unix sockets across managed requirements,
permissions profiles, the network proxy config, and the app server
protocol.
Concretely, it:
- introduces typed domain (`allow` / `deny`) and unix socket permission
(`allow` / `none`) entries instead of separate `allowed_domains`,
`denied_domains`, and `allow_unix_sockets` lists
- updates config loading, managed requirements merging, and exec-policy
overlays to read and upsert rule entries consistently
- exposes the new shape through protocol/schema outputs, debug surfaces,
and app-server config APIs
- rejects the legacy list-based keys and updates docs/tests to reflect
the new config format
## Why
The previous representation split related network policy across multiple
parallel lists, which made merging and overriding rules harder to reason
about. Moving to explicit keyed permission maps gives us a single source
of truth per host/socket entry, makes allow/deny precedence clearer, and
gives protocol consumers access to the full rule state instead of
derived projections only.
## Backward Compatibility
### Backward compatible
- Managed requirements still accept the legacy
`experimental_network.allowed_domains`,
`experimental_network.denied_domains`, and
`experimental_network.allow_unix_sockets` fields. They are normalized
into the new canonical `domains` and `unix_sockets` maps internally.
- App-server v2 still deserializes legacy `allowedDomains`,
`deniedDomains`, and `allowUnixSockets` payloads, so older clients can
continue reading managed network requirements.
- App-server v2 responses still populate `allowedDomains`,
`deniedDomains`, and `allowUnixSockets` as legacy compatibility views
derived from the canonical maps.
- `managed_allowed_domains_only` keeps the same behavior after
normalization. Legacy managed allowlists still participate in the same
enforcement path as canonical `domains` entries.
### Not backward compatible
- Permissions profiles under `[permissions.<profile>.network]` no longer
accept the legacy list-based keys. Those configs must use the canonical
`[domains]` and `[unix_sockets]` tables instead of `allowed_domains`,
`denied_domains`, or `allow_unix_sockets`.
- Managed `experimental_network` config cannot mix canonical and legacy
forms in the same block. For example, `domains` cannot be combined with
`allowed_domains` or `denied_domains`, and `unix_sockets` cannot be
combined with `allow_unix_sockets`.
- The canonical format can express explicit `"none"` entries for unix
sockets, but those entries do not round-trip through the legacy
compatibility fields because the legacy fields only represent allow/deny
lists.
## Testing
`/target/debug/codex sandbox macos --log-denials /bin/zsh -c 'curl
https://www.example.com' ` gives 200 with config
```
[permissions.workspace.network.domains]
"www.example.com" = "allow"
```
and fails when set to deny: `curl: (56) CONNECT tunnel failed, response
403`.
Also tested backward compatibility path by verifying that adding the
following to `/etc/codex/requirements.toml` works:
```
[experimental_network]
allowed_domains = ["www.example.com"]
```
## Why
`codex-tools` already owns the shared tool input schema model and parser
from the first extraction step, but `core/src/tools/spec.rs` still owned
the MCP-specific adapter that normalizes `rmcp::model::Tool` schemas and
wraps `structuredContent` into the call result output schema.
Keeping that adapter in `codex-core` means the reusable MCP schema path
is still split across crates, and the unit tests for that logic stay
anchored in `codex-core` even though the runtime orchestration does not
need to move yet.
This change takes the next small step by moving the reusable MCP schema
adapter into `codex-tools` while leaving `ResponsesApiTool` assembly in
`codex-core`.
## What changed
- added `tools/src/mcp_tool.rs` and sibling
`tools/src/mcp_tool_tests.rs`
- introduced `ParsedMcpTool`, `parse_mcp_tool()`, and
`mcp_call_tool_result_output_schema()` in `codex-tools`
- updated `core/src/tools/spec.rs` to consume parsed MCP tool parts from
`codex-tools`
- removed the now-redundant MCP schema unit tests from
`core/src/tools/spec_tests.rs`
- expanded `codex-rs/tools/README.md` to describe this second migration
step
## Test plan
- `cargo test -p codex-tools`
- `cargo test -p codex-core --lib tools::spec::`
## Summary
This PR makes Windows sandbox proxying enforceable by routing proxy-only
runs through the existing `offline` sandbox user and reserving direct
network access for the existing `online` sandbox user.
In brief:
- if a Windows sandbox run should be proxy-enforced, we run it as the
`offline` user
- the `offline` user gets firewall rules that block direct outbound
traffic and only permit the configured localhost proxy path
- if a Windows sandbox run should have true direct network access, we
run it as the `online` user
- no new sandbox identity is introduced
This brings Windows in line with the intended model: proxy use is not
just env-based, it is backed by OS-level egress controls. Windows
already has two sandbox identities:
- `offline`: intended to have no direct network egress
- `online`: intended to have full network access
This PR makes proxy-enforced runs use that model directly.
### Proxy-enforced runs
When proxy enforcement is active:
- the run is assigned to the `offline` identity
- setup extracts the loopback proxy ports from the sandbox env
- Windows setup programs firewall rules for the `offline` user that:
- block all non-loopback outbound traffic
- block loopback UDP
- block loopback TCP except for the configured proxy ports
- optionally allow broader localhost access when `allow_local_binding=1`
So the sandboxed process can only talk to the local proxy. It cannot
open direct outbound sockets or do local UDP-based DNS on its own.The
proxy then performs the real outbound network access outside that
restricted sandbox identity.
### Direct-network runs
When proxy enforcement is not active and full network access is allowed:
- the run is assigned to the `online` identity
- no proxy-only firewall restrictions are applied
- the process gets normal direct network access
### Unelevated vs elevated
The restricted-token / unelevated path cannot enforce per-identity
firewall policy by itself.
So for Windows proxy-enforced runs, we transparently use the logon-user
sandbox path under the hood, even if the caller started from the
unelevated mode. That keeps enforcement real instead of best-effort.
---------
Co-authored-by: Codex <noreply@openai.com>
## Why
`PermissionProfile` should only describe the per-command permissions we
still want to grant dynamically. Keeping
`MacOsSeatbeltProfileExtensions` in that surface forced extra macOS-only
approval, protocol, schema, and TUI branches for a capability we no
longer want to expose.
## What changed
- Removed the macOS-specific permission-profile types from
`codex-protocol`, the app-server v2 API, and the generated
schema/TypeScript artifacts.
- Deleted the core and sandboxing plumbing that threaded
`MacOsSeatbeltProfileExtensions` through execution requests and seatbelt
construction.
- Simplified macOS seatbelt generation so it always includes the fixed
read-only preferences allowlist instead of carrying a configurable
profile extension.
- Removed the macOS additional-permissions UI/docs/test coverage and
deleted the obsolete macOS permission modules.
- Tightened `request_permissions` intersection handling so explicitly
empty requested read lists are preserved only when that field was
actually granted, avoiding zero-grant responses being stored as active
permissions.
## Why
`parse_tool_input_schema` and the supporting `JsonSchema` model were
living in `core/src/tools/spec.rs`, but they already serve callers
outside `codex-core`.
Keeping that shared schema parsing logic inside `codex-core` makes the
crate boundary harder to reason about and works against the guidance in
`AGENTS.md` to avoid growing `codex-core` when reusable code can live
elsewhere.
This change takes the first extraction step by moving the schema parsing
primitive into its own crate while keeping the rest of the tool-spec
assembly in `codex-core`.
## What changed
- added a new `codex-tools` crate under `codex-rs/tools`
- moved the shared tool input schema model and sanitizer/parser into
`tools/src/json_schema.rs`
- kept `tools/src/lib.rs` exports-only, with the module-level unit tests
split into `json_schema_tests.rs`
- updated `codex-core` to use `codex-tools::JsonSchema` and re-export
`parse_tool_input_schema`
- updated `codex-app-server` dynamic tool validation to depend on
`codex-tools` directly instead of reaching through `codex-core`
- wired the new crate into the Cargo workspace and Bazel build graph
## Why
This is effectively a follow-up to
[#15812](https://github.com/openai/codex/pull/15812). That change
removed the special skill-script exec path, but `skill_metadata` was
still being threaded through command-approval payloads even though the
approval flow no longer uses it to render prompts or resolve decisions.
Keeping it around added extra protocol, schema, and client surface area
without changing behavior.
Removing it keeps the command-approval contract smaller and avoids
carrying a dead field through app-server, TUI, and MCP boundaries.
## What changed
- removed `ExecApprovalRequestSkillMetadata` and the corresponding
`skillMetadata` field from core approval events and the v2 app-server
protocol
- removed the generated JSON and TypeScript schema output for that field
- updated app-server, MCP server, TUI, and TUI app-server approval
plumbing to stop forwarding the field
- cleaned up tests that previously constructed or asserted
`skillMetadata`
## Testing
- `cargo test -p codex-app-server-protocol`
- `cargo test -p codex-protocol`
- `cargo test -p codex-app-server-test-client`
- `cargo test -p codex-mcp-server`
- `just argument-comment-lint`
## Summary
- move the bwrap PATH lookup and warning helpers out of config/mod.rs
- move the related tests into a dedicated bwrap_tests.rs file
## Validation
- git diff --check
- skipped heavier local tests per request
Follow-up to #15791.
## Why
`SandboxCommand.program` represents an executable path, but keeping it
as `String` forced path-backed callers to run `to_string_lossy()` before
the sandbox layer ever touched the command. That loses fidelity earlier
than necessary and adds avoidable conversions in runtimes that already
have a `PathBuf`.
## What changed
- Changed `SandboxCommand.program` to `OsString`.
- Updated `SandboxManager::transform` to keep the program and argv in
`OsString` form until the `SandboxExecRequest` conversion boundary.
- Switched the path-backed `apply_patch` and `js_repl` runtimes to pass
`into_os_string()` instead of `to_string_lossy()`.
- Updated the remaining string-backed builders and tests to match the
new type while preserving the existing Linux helper `arg0` behavior.
## Verification
- `cargo test -p codex-sandboxing`
- `just argument-comment-lint -p codex-core -p codex-sandboxing`
- `cargo test -p codex-core` currently fails in unrelated existing
config tests: `config::tests::approvals_reviewer_*` and
`config::tests::smart_approvals_alias_*`
## Why
`token_data` is owned by `codex-login`, but `codex-core` was still
re-exporting it. That let callers pull auth token types through
`codex-core`, which keeps otherwise unrelated crates coupled to
`codex-core` and makes `codex-core` more of a build-graph bottleneck.
## What changed
- remove the `codex-core` re-export of `codex_login::token_data`
- update the remaining `codex-core` internals that used
`crate::token_data` to import `codex_login::token_data` directly
- update downstream callers in `codex-rs/chatgpt`,
`codex-rs/tui_app_server`, `codex-rs/app-server/tests/common`, and
`codex-rs/core/tests` to import `codex_login::token_data` directly
- add explicit `codex-login` workspace dependencies and refresh lock
metadata for crates that now depend on it directly
## Validation
- `cargo test -p codex-chatgpt --locked`
- `just argument-comment-lint`
- `just bazel-lock-update`
- `just bazel-lock-check`
## Notes
- attempted `cargo test -p codex-core --locked` and `cargo test -p
codex-core auth_refresh --locked`, but both ran out of disk while
linking `codex-core` test binaries in the local environment
## Problem
Codex already treated an existing top-level project `./.codex` directory
as protected, but there was a gap on first creation.
If `./.codex` did not exist yet, a turn could create files under it,
such as `./.codex/config.toml`, without going through the same approval
path as later modifications. That meant the initial write could bypass
the intended protection for project-local Codex state.
## What this changes
This PR closes that first-creation gap in the Unix enforcement layers:
- `codex-protocol`
- treat the top-level project `./.codex` path as a protected carveout
even when it does not exist yet
- avoid injecting the default carveout when the user already has an
explicit rule for that exact path
- macOS Seatbelt
- deny writes to both the exact protected path and anything beneath it,
so creating `./.codex` itself is blocked in addition to writes inside it
- Linux bubblewrap
- preserve the same protected-path behavior for first-time creation
under `./.codex`
- tests
- add protocol regressions for missing `./.codex` and explicit-rule
collisions
- add Unix sandbox coverage for blocking first-time `./.codex` creation
- tighten Seatbelt policy assertions around excluded subpaths
## Scope
This change is intentionally scoped to protecting the top-level project
`.codex` subtree from agent writes.
It does not make `.codex` unreadable, and it does not change the product
behavior around loading project skills from `.codex` when project config
is untrusted.
## Why this shape
The fix is pointed rather than broad:
- it preserves the current model of “project `.codex` is protected from
writes”
- it closes the security-relevant first-write hole
- it avoids folding a larger permissions-model redesign into this PR
## Validation
- `cargo test -p codex-protocol`
- `cargo test -p codex-sandboxing seatbelt`
- `cargo test -p codex-exec --test all
sandbox_blocks_first_time_dot_codex_creation -- --nocapture`
---------
Co-authored-by: Michael Bolin <mbolin@openai.com>
## Why
Skill metadata accepted a `permissions` block and stored the result on
`SkillMetadata`, but that data was never consumed by runtime behavior.
Leaving the dead parsing path in place makes it look like skills can
widen or otherwise influence execution permissions when, in practice,
declared skill permissions are ignored.
This change removes that misleading surface area so the skill metadata
model matches what the system actually uses.
## What changed
- removed `permission_profile` and `managed_network_override` from
`core-skills::SkillMetadata`
- stopped parsing `permissions` from skill metadata in
`core-skills/src/loader.rs`
- deleted the loader tests that only exercised the removed permissions
parsing path
- cleaned up dependent `SkillMetadata` constructors in tests and TUI
code that were only carrying `None` for those fields
## Testing
- `cargo test -p codex-core-skills`
- `cargo test -p codex-tui
submission_prefers_selected_duplicate_skill_path`
- `just argument-comment-lint`
## Summary
- resolve system bwrap from PATH instead of hardcoding /usr/bin/bwrap
- skip PATH entries that resolve inside the current workspace before
launching the sandbox helper
- keep the vendored bubblewrap fallback when no trusted system bwrap is
found
## Validation
- cargo test -p codex-core bwrap --lib
- cargo test -p codex-linux-sandbox
- just fix -p codex-core
- just fix -p codex-linux-sandbox
- just fmt
- just argument-comment-lint
- cargo clean
- [x] Polish tool suggest prompts to distinguish between missing
connectors and discoverable plugins, and be very precise about the
triggering conditions.
