Each call is an action in the ledger (stage first, repair or fallback) and the trace line
lists every call. The generator is never an Anthropic model. An answer of no edits is not
said again. The edit is tested as before.
No edits, JSON that does not read or is no edit, and no JSON were all 'holds no JSON'.
spelling() tells a refusal of how an answer is written from one of what it says, and
near() shows the file's own lines where an edit's old text was meant to stand.
A call that came back without text (finish length, 3,072 of 3,072 completion tokens
reasoning) was neither kept nor written to the spend file; now every call that reached
the model is kept with its request, finish reason and usage (reasoning tokens apart),
a call without text or ended by the server with an error is kept with its error, billed
and asked again. At Hanzo's API the answer is a strict json_schema with at least one
edit; the generator's answer room is 2,048 where 512 cut off its thinking aloud.
## Summary
- replace manual wiremock SSE mounts in the compact suite with the
shared response helpers
- simplify the exec auth_env integration test by using the
mount_sse_once_match helper
- rely on mount_sse_sequence plus server request collection to replace
the bespoke SeqResponder utility in tests
## Testing
- just fmt
------
https://chatgpt.com/codex/tasks/task_i_68e2e238f2a88320a337f0b9e4098093
## Summary
- add a reusable `ev_response_created` helper that builds
`response.created` SSE events for integration tests
- update the exec and core integration suites to use the new helper
instead of repeating manual JSON literals
- keep the streaming fixtures consistent by relying on the shared helper
in every touched test
## Testing
- `just fmt`
------
https://chatgpt.com/codex/tasks/task_i_68e1fe885bb883208aafffb94218da61
`codex-responses-api-proxy` is designed so that there should be exactly
one copy of the API key in memory (that is `mlock`'d on UNIX), but in
practice, I was seeing two when I dumped the process data from
`/proc/$PID/mem`.
It appears that `std::io::stdin()` maintains an internal `BufReader`
that we cannot zero out, so this PR changes the implementation on UNIX
so that we use a low-level `read(2)` instead.
Even though it seems like it would be incredibly unlikely, we also make
this logic tolerant of short reads. Either `\n` or `EOF` must be sent to
signal the end of the key written to stdin.
## Summary
- replace manual event polling loops in several core test suites with
the shared wait_for_event helpers
- keep prior assertions intact by using closure captures for stateful
expectations, including plan updates, patch lifecycles, and review flow
checks
- rely on wait_for_event_with_timeout where longer waits are required,
simplifying timeout handling
## Testing
- just fmt
------
https://chatgpt.com/codex/tasks/task_i_68e1d58582d483208febadc5f90dd95e
## Summary
This PR is an alternative approach to #4711, but instead of changing our
storage, parses out shell calls in the client and reserializes them on
the fly before we send them out as part of the request.
What this changes:
1. Adds additional serialization logic when the
ApplyPatchToolType::Freeform is in use.
2. Adds a --custom-apply-patch flag to enable this setting on a
session-by-session basis.
This change is delicate, but is not meant to be permanent. It is meant
to be the first step in a migration:
1. (This PR) Add in-flight serialization with config
2. Update model_family default
3. Update serialization logic to store turn outputs in a structured
format, with logic to serialize based on model_family setting.
4. Remove this rewrite in-flight logic.
## Test Plan
- [x] Additional unit tests added
- [x] Integration tests added
- [x] Tested locally
In the past, we were treating `input exceeded context window` as a
streaming error and retrying on it. Retrying on it has no point because
it won't change the behavior. In this PR, we surface the error to the
client without retry and also send a token count event to indicate that
the context window is full.
<img width="650" height="125" alt="image"
src="https://github.com/user-attachments/assets/c26b1213-4c27-4bfc-90f4-51a270a3efd5"
/>
We truncate the output of exec commands to not blow the context window.
However, some cases we weren't doing that. This caused reports of people
with 76% context window left facing `input exceeded context window`
which is weird.
# External (non-OpenAI) Pull Request Requirements
Before opening this Pull Request, please read the dedicated
"Contributing" markdown file or your PR may be closed:
https://github.com/openai/codex/blob/main/docs/contributing.md
If your PR conforms to our contribution guidelines, replace this text
with a detailed and high quality description of your changes.
Codex isn’t great yet on Windows outside of WSL, and while we’ve merged
https://github.com/openai/codex/pull/4269 to reduce the repetitive
manual approvals on readonly commands, we’ve noticed that users seem to
have more issues with GPT-5-Codex than with GPT-5 on Windows.
This change makes GPT-5 the default for Windows users while we continue
to improve the CLI harness and model for GPT-5-Codex on Windows.
Apparently we were not running our `pnpm run prettier` check in CI, so
many files that were covered by the existing Prettier check were not
well-formatted.
This updates CI and formats the files.
This issue was due to the fact that the timeout is not always sufficient
to have enough character for truncation + a race between synthetic
timeout and process kill
# External (non-OpenAI) Pull Request Requirements
Before opening this Pull Request, please read the dedicated
"Contributing" markdown file or your PR may be closed:
https://github.com/openai/codex/blob/main/docs/contributing.md
If your PR conforms to our contribution guidelines, replace this text
with a detailed and high quality description of your changes.
This updates `codex exec` so that, by default, most of the agent's
activity is written to stderr so that only the final agent message is
written to stdout. This makes it easier to pipe `codex exec` into
another tool without extra filtering.
I introduced `#![deny(clippy::print_stdout)]` to help enforce this
change and renamed the `ts_println!()` macro to `ts_msg()` because (1)
it no longer calls `println!()` and (2), `ts_eprintln!()` seemed too
long of a name.
While here, this also adds `-o` as an alias for `--output-last-message`.
Fixes https://github.com/openai/codex/issues/1670
# Tool System Refactor
- Centralizes tool definitions and execution in `core/src/tools/*`:
specs (`spec.rs`), handlers (`handlers/*`), router (`router.rs`),
registry/dispatch (`registry.rs`), and shared context (`context.rs`).
One registry now builds the model-visible tool list and binds handlers.
- Router converts model responses to tool calls; Registry dispatches
with consistent telemetry via `codex-rs/otel` and unified error
handling. Function, Local Shell, MCP, and experimental `unified_exec`
all flow through this path; legacy shell aliases still work.
- Rationale: reduce per‑tool boilerplate, keep spec/handler in sync, and
make adding tools predictable and testable.
Example: `read_file`
- Spec: `core/src/tools/spec.rs` (see `create_read_file_tool`,
registered by `build_specs`).
- Handler: `core/src/tools/handlers/read_file.rs` (absolute `file_path`,
1‑indexed `offset`, `limit`, `L#: ` prefixes, safe truncation).
- E2E test: `core/tests/suite/read_file.rs` validates the tool returns
the requested lines.
## Next steps:
- Decompose `handle_container_exec_with_params`
- Add parallel tool calls
Previously, users could supply their API key directly via:
```shell
codex login --api-key KEY
```
but this has the drawback that `KEY` is more likely to end up in shell
history, can be read from `/proc`, etc.
This PR removes support for `--api-key` and replaces it with
`--with-api-key`, which reads the key from stdin, so either of these are
better options:
```
printenv OPENAI_API_KEY | codex login --with-api-key
codex login --with-api-key < my_key.txt
```
Other CLIs, such as `gh auth login --with-token`, follow the same
practice.
Before when you would enter `/di`, hit tab on `/diff`, and then hit
enter, it would execute `/diff`. But now it's just sending it as a text.
This fixes the issue.
### Summary
* Updated fuzzy search result to include the file name.
* This should not affect CLI usage and the UI there will be addressed in
a separate PR.
### Testing
Tested locally and with the extension.
### Screenshot
<img width="431" height="244" alt="Screenshot 2025-10-02 at 11 08 44 AM"
src="https://github.com/user-attachments/assets/ba2ca299-a81d-4453-9242-1750e945aea2"
/>
---------
Co-authored-by: shijie.rao <[email protected]>
This PR fixes a bug that results in a hang in the oauth login flow if a
user logs in, then logs out, then logs in again without first closing
the browser window.
Root cause of problem: We use a local web server for the oauth flow, and
it's implemented using the `tiny_http` rust crate. During the first
login, a socket is created between the browser and the server. The
`tiny_http` library creates worker threads that persist for as long as
this socket remains open. Currently, there's no way to close the
connection on the server side — the library provides no API to do this.
The library also filters all "Connect: close" headers, which makes it
difficult to tell the client browser to close the connection. On the
second login attempt, the browser uses the existing connection rather
than creating a new one. Since that connection is associated with a
server instance that no longer exists, it is effectively ignored.
I considered switching from `tiny_http` to a different web server
library, but that would have been a big change with significant
regression risk. This PR includes a more surgical fix that works around
the limitation of `tiny_http` and sends a "Connect: close" header on the
last "success" page of the oauth flow.
Before this PR:
```typescript
export type RequestId = string | bigint;
```
After:
```typescript
export type RequestId = string | number;
```
`bigint` introduces headaches in TypeScript without providing any real
value.
I just had to use this like so:
```
./codex-rs/scripts/create_github_release --publish-alpha --emergency-version-override 0.43.0-alpha.10
```
because the build for `0.43.0-alpha.9` failed:
https://github.com/openai/codex/actions/runs/18167317356
## Summary
- show the remaining context window percentage in `/status` alongside
existing token usage details
- replace the composer shortcut prompt with the context window
percentage (or an unavailable message) while a task is running
- update TUI snapshots to reflect the new context window line
## Testing
- cargo test -p codex-tui
------
https://chatgpt.com/codex/tasks/task_i_68dc6e7397ac8321909d7daff25a396c
## Summary
- show a dim “(no output)” placeholder when an executed command produces
no stdout or stderr so empty runs are visible
- update TUI snapshots to include the new placeholder in history
renderings
## Testing
- cargo test -p codex-tui
------
https://chatgpt.com/codex/tasks/task_i_68dc056c1d5883218fe8d9929e9b1657
**Summary**
This PR fixes an issue in the device code login flow where trailing
slashes in the issuer URL could cause malformed URLs during codex token
exchange step
**Test**
Before the changes
`Error logging in with device code: device code exchange failed: error
decoding response body`
After the changes
`Successfully logged in`
Implement command safety for PowerShell commands on Windows
This change adds a new Windows-specific command-safety module under
`codex-rs/core/src/command_safety/windows_safe_commands.rs` to strictly
sanitise PowerShell invocations. Key points:
- Introduce `is_safe_command_windows()` to only allow explicitly
read-only PowerShell calls.
- Parse and split PowerShell invocations (including inline `-Command`
scripts and pipelines).
- Block unsafe switches (`-File`, `-EncodedCommand`, `-ExecutionPolicy`,
unknown flags, call operators, redirections, separators).
- Whitelist only read-only cmdlets (`Get-ChildItem`, `Get-Content`,
`Select-Object`, etc.), safe Git subcommands (`status`, `log`, `show`,
`diff`, `cat-file`), and ripgrep without unsafe options.
- Add comprehensive unit tests covering allowed and rejected command
patterns (nested calls, side effects, chaining, redirections).
This ensures Codex on Windows can safely execute discover-only
PowerShell workflows without risking destructive operations.
There was a bit of copypasta I put up with when were publishing two
packages to npm, but now that it's three, I created some more scripts to
consolidate things.
With this change, I ran:
```shell
./scripts/stage_npm_packages.py --release-version 0.43.0-alpha.8 --package codex --package codex-responses-api-proxy --package codex-sdk
```
Indeed when it finished, I ended up with:
```shell
$ tree dist
dist
└── npm
├── codex-npm-0.43.0-alpha.8.tgz
├── codex-responses-api-proxy-npm-0.43.0-alpha.8.tgz
└── codex-sdk-npm-0.43.0-alpha.8.tgz
$ tar tzvf dist/npm/codex-sdk-npm-0.43.0-alpha.8.tgz
-rwxr-xr-x 0 0 0 25476720 Oct 26 1985 package/vendor/aarch64-apple-darwin/codex/codex
-rwxr-xr-x 0 0 0 29871400 Oct 26 1985 package/vendor/aarch64-unknown-linux-musl/codex/codex
-rwxr-xr-x 0 0 0 28368096 Oct 26 1985 package/vendor/x86_64-apple-darwin/codex/codex
-rwxr-xr-x 0 0 0 36029472 Oct 26 1985 package/vendor/x86_64-unknown-linux-musl/codex/codex
-rw-r--r-- 0 0 0 10926 Oct 26 1985 package/LICENSE
-rw-r--r-- 0 0 0 30187520 Oct 26 1985 package/vendor/aarch64-pc-windows-msvc/codex/codex.exe
-rw-r--r-- 0 0 0 35277824 Oct 26 1985 package/vendor/x86_64-pc-windows-msvc/codex/codex.exe
-rw-r--r-- 0 0 0 4842 Oct 26 1985 package/dist/index.js
-rw-r--r-- 0 0 0 1347 Oct 26 1985 package/package.json
-rw-r--r-- 0 0 0 9867 Oct 26 1985 package/dist/index.js.map
-rw-r--r-- 0 0 0 12 Oct 26 1985 package/README.md
-rw-r--r-- 0 0 0 4287 Oct 26 1985 package/dist/index.d.ts
```
# Extract and Centralize Sandboxing
- Goal: Improve safety and clarity by centralizing sandbox planning and
execution.
- Approach:
- Add planner (ExecPlan) and backend registry (Direct/Seatbelt/Linux)
with run_with_plan.
- Refactor codex.rs to plan-then-execute; handle failures/escalation via
the plan.
- Delegate apply_patch to the codex binary and run it with an empty env
for determinism.
# External (non-OpenAI) Pull Request Requirements
Before opening this Pull Request, please read the dedicated
"Contributing" markdown file or your PR may be closed:
https://github.com/openai/codex/blob/main/docs/contributing.md
If your PR conforms to our contribution guidelines, replace this text
with a detailed and high quality description of your changes.
We continue the separation between `codex app-server` and `codex
mcp-server`.
In particular, we introduce a new crate, `codex-app-server-protocol`,
and migrate `codex-rs/protocol/src/mcp_protocol.rs` into it, renaming it
`codex-rs/app-server-protocol/src/protocol.rs`.
Because `ConversationId` was defined in `mcp_protocol.rs`, we move it
into its own file, `codex-rs/protocol/src/conversation_id.rs`, and
because it is referenced in a ton of places, we have to touch a lot of
files as part of this PR.
We also decide to get away from proper JSON-RPC 2.0 semantics, so we
also introduce `codex-rs/app-server-protocol/src/jsonrpc_lite.rs`, which
is basically the same `JSONRPCMessage` type defined in `mcp-types`
except with all of the `"jsonrpc": "2.0"` removed.
Getting rid of `"jsonrpc": "2.0"` makes our serialization logic
considerably simpler, as we can lean heavier on serde to serialize
directly into the wire format that we use now.
Manually curating `protocol-ts/src/lib.rs` was error-prone, as expected.
I finally asked Codex to write some Rust macros so we can ensure that:
- For every variant of `ClientRequest` and `ServerRequest`, there is an
associated `params` and `response` type.
- All response types are included automatically in the output of `codex
generate-ts`.
I don't believe there is any upside in making process hardening opt-in
for Codex CLI releases. If you want to tinker with Codex CLI, then build
from source (or run as `root`)?
Fixes:
- Removed overdeclaration of types that were unnecessary because they
were already included by induction.
- Reordered list of response types to match the enum order, making it
easier to identify what was missing.
- Added `ExecArbitraryCommandResponse` because it was missing.
- Leveraged `use codex_protocol::mcp_protocol::*;` to make the file more
readable.
- Removed crate dependency on `mcp-types` now that we have separate the
app server from the MCP server:
https://github.com/openai/codex/pull/4471
My next move is to come up with some scheme that ensures request types
always have a response type and that the response type is automatically
included with the output of `codex generate-ts`.
This ensures changes the generated TypeScript type for `ClientRequest`
so that instead of this:
```typescript
/**
* Request from the client to the server.
*/
export type ClientRequest =
| { method: "initialize"; id: RequestId; params: InitializeParams }
| { method: "newConversation"; id: RequestId; params: NewConversationParams }
// ...
| { method: "getUserAgent"; id: RequestId }
| { method: "userInfo"; id: RequestId }
// ...
```
we have this:
```typescript
/**
* Request from the client to the server.
*/
export type ClientRequest =
| { method: "initialize"; id: RequestId; params: InitializeParams }
| { method: "newConversation"; id: RequestId; params: NewConversationParams }
// ...
| { method: "getUserAgent"; id: RequestId; params: undefined }
| { method: "userInfo"; id: RequestId; params: undefined }
// ...
```
which makes TypeScript happier when it comes to destructuring instances
of `ClientRequest` because it does not complain about `params` not being
guaranteed to exist anymore.
Update prompt to prevent codex to use Python script or fancy commands to
edit files.
## Testing:
3 scenarios have been considered:
1. Rename codex to meca_code. Proceed to the whole refactor file by
file. Don't ask for approval at each step
2. Add a description to every single function you can find in the repo
3. Rewrite codex.rs in a more idiomatic way. Make sure to touch ONLY
this file and that clippy does not complain at the end
Before this update, 22% (estimation as it's sometimes hard to find all
the creative way the model find to edit files) of the file editions
where made using something else than a raw `apply_patch`
After this update, not a single edition without `apply_patch` was found
[EDIT]
I managed to have a few `["bash", "-lc", "apply_path"]` when reaching <
10% context left
Here's the logic:
1. If text is empty and selector is open:
- Enter on a prompt without args should autosubmit the prompt
- Enter on a prompt with numeric args should add `/prompts:name ` to the
text input
- Enter on a prompt with named args should add `/prompts:name ARG1=""
ARG2=""` to the text input
2. If text is not empty but no args are passed:
- For prompts with numeric args -> we allow it to submit (params are
optional)
- For prompts with named args -> we throw an error (all params should
have values)
<img width="454" height="246" alt="Screenshot 2025-09-23 at 2 23 21 PM"
src="https://github.com/user-attachments/assets/fd180a1b-7d17-42ec-b231-8da48828b811"
/>
This is a very large PR with some non-backwards-compatible changes.
Historically, `codex mcp` (or `codex mcp serve`) started a JSON-RPC-ish
server that had two overlapping responsibilities:
- Running an MCP server, providing some basic tool calls.
- Running the app server used to power experiences such as the VS Code
extension.
This PR aims to separate these into distinct concepts:
- `codex mcp-server` for the MCP server
- `codex app-server` for the "application server"
Note `codex mcp` still exists because it already has its own subcommands
for MCP management (`list`, `add`, etc.)
The MCP logic continues to live in `codex-rs/mcp-server` whereas the
refactored app server logic is in the new `codex-rs/app-server` folder.
Note that most of the existing integration tests in
`codex-rs/mcp-server/tests/suite` were actually for the app server, so
all the tests have been moved with the exception of
`codex-rs/mcp-server/tests/suite/mod.rs`.
Because this is already a large diff, I tried not to change more than I
had to, so `codex-rs/app-server/tests/common/mcp_process.rs` still uses
the name `McpProcess` for now, but I will do some mechanical renamings
to things like `AppServer` in subsequent PRs.
While `mcp-server` and `app-server` share some overlapping functionality
(like reading streams of JSONL and dispatching based on message types)
and some differences (completely different message types), I ended up
doing a bit of copypasta between the two crates, as both have somewhat
similar `message_processor.rs` and `outgoing_message.rs` files for now,
though I expect them to diverge more in the near future.
One material change is that of the initialize handshake for `codex
app-server`, as we no longer use the MCP types for that handshake.
Instead, we update `codex-rs/protocol/src/mcp_protocol.rs` to add an
`Initialize` variant to `ClientRequest`, which takes the `ClientInfo`
object we need to update the `USER_AGENT_SUFFIX` in
`codex-rs/app-server/src/message_processor.rs`.
One other material change is in
`codex-rs/app-server/src/codex_message_processor.rs` where I eliminated
a use of the `send_event_as_notification()` method I am generally trying
to deprecate (because it blindly maps an `EventMsg` into a
`JSONNotification`) in favor of `send_server_notification()`, which
takes a `ServerNotification`, as that is intended to be a custom enum of
all notification types supported by the app server. So to make this
update, I had to introduce a new variant of `ServerNotification`,
`SessionConfigured`, which is a non-backwards compatible change with the
old `codex mcp`, and clients will have to be updated after the next
release that contains this PR. Note that
`codex-rs/app-server/tests/suite/list_resume.rs` also had to be update
to reflect this change.
I introduced `codex-rs/utils/json-to-toml/src/lib.rs` as a small utility
crate to avoid some of the copying between `mcp-server` and
`app-server`.
hanzo-dev