Compare commits

...

77 Commits

Author SHA1 Message Date
Michael Bolin
ef7208359f feat: show Config overview at start of exec (#1073)
Now the `exec` output starts with something like:

```
--------
workdir:  /Users/mbolin/code/codex/codex-rs
model:  o3
provider:  openai
approval:  Never
sandbox:  SandboxPolicy { permissions: [DiskFullReadAccess, DiskWritePlatformUserTempFolder, DiskWritePlatformGlobalTempFolder, DiskWriteCwd, DiskWriteFolder { folder: "/Users/mbolin/.pyenv/shims" }] }
--------
```

which makes it easier to reason about when looking at logs.
2025-05-21 22:53:02 -07:00
Michael Bolin
5746561428 chore: move types out of config.rs into config_types.rs (#1054)
`config.rs` is already quite long without these definitions. Since they
have no real dependencies of their own, let's move them to their own
file so `config.rs` can focus on the business logic of loading a config.
2025-05-20 11:55:25 -07:00
Michael Bolin
d766e845b3 feat: experimental --output-last-message flag to exec subcommand (#1037)
This introduces an experimental `--output-last-message` flag that can be
used to identify a file where the final message from the agent will be
written. Two use cases:

- Ultimately, we will likely add a `--quiet` option to `exec`, but even
if the user does not want any output written to the terminal, they
probably want to know what the agent did. Writing the output to a file
makes it possible to get that information in a clean way.
- Relatedly, when using `exec` in CI, it is easier to review the
transcript written "normally," (i.e., not as JSON or something with
extra escapes), but getting programmatic access to the last message is
likely helpful, so writing the last message to a file gets the best of
both worlds.

I am calling this "experimental" because it is possible that we are
overfitting and will want a more general solution to this problem that
would justify removing this flag.
2025-05-19 16:08:18 -07:00
Michael Bolin
a4bfdf6779 chore: produce .tar.gz versions of artifacts in addition to .zst (#1036)
For sparse containers/environments that do not have `zstd`, provide
`.tar.gz` as alternative archive format.
2025-05-19 15:17:45 -07:00
Fouad Matin
44022db8d0 bump(version): 0.1.2505172129 (#1008)
## `0.1.2505172129`

### 🪲 Bug Fixes

- Add node version check (#1007)
- Persist token after refresh (#1006)
2025-05-17 21:35:54 -07:00
Fouad Matin
a86270f581 fix: add node version check (#1007) 2025-05-17 21:27:41 -07:00
Fouad Matin
835eb77a7d fix: persist token after refresh (#1006)
After a token refresh/exchange, persist the new refresh and id token
2025-05-17 21:27:02 -07:00
Fouad Matin
dbc0ad348e bump(version): 0.1.2505171619 (#1001)
## `0.1.2505171619`

- `codex --login` + `codex --free` (#998)
2025-05-17 16:25:21 -07:00
Fouad Matin
9b4c2984d4 add: codex --login + codex --free (#998)
## Summary
- add `--login` and `--free` flags to cli help
- handle `--login` and `--free` logic in cli
- factor out redeem flow into `maybeRedeemCredits`
- call new helper from login callback
2025-05-17 16:13:12 -07:00
Michael Bolin
f3bde21759 chore: update install_native_deps.sh to use rust-v0.0.2505171051 (#995)
Use a more recent built of the Rust binaries to include with the Node
module.
2025-05-17 11:25:24 -07:00
Michael Bolin
1c6a3f1097 fix: artifacts from previous frames were bleeding through in TUI (#989)
Prior to this PR, I would frequently see glyphs from previous frames
"bleed" through like this:


![image](https://github.com/user-attachments/assets/8784b3d7-f691-4df6-8666-34e2f134ee85)

I think this was due to two issues (now addressed in this PR):

* We were not making use of `ratatui::widgets::Clear` to clear out the
buffer before drawing into it.
* To calculate the `width` used with `wrapped_line_count_for_cell()`, we
were not accounting for the scrollbar.
* Now we calculate `effective_width` using
`inner.width.saturating_sub(1)` where the `1` is for the scrollbar.
* We compute `text_area` using `effective_with` and pass the `text_area`
to `paragraph.render()`.
* We eliminate the conditional `needs_scrollbar` check and always call
`render(Scrollbar)`

I suspect this bug was introduced in
https://github.com/openai/codex/pull/937, though I did not try to
verify: I'm just happy that it appears to be fixed!
2025-05-17 10:51:11 -07:00
Michael Bolin
f8b6b1db81 fix: ensure the first user message always displays after the session info (#988)
Previously, if the first user message was sent with the command
invocation, e.g.:

```
$ cargo run --bin codex 'hello'
```

Then the user message was added as the first entry in the history and
then `is_first_event` would be `false` here:


031df77dfb/codex-rs/tui/src/conversation_history_widget.rs (L178-L179)

which would prevent the "welcome" message with things like the the model
version from displaying.

The fix in this PR is twofold:

* Reorganize the logic so the `ChatWidget` constructor stores
`initial_user_message` rather than sending it right away. Now inside
`handle_codex_event()`, it waits for the `SessionConfigured` event and
sends the `initial_user_message`, if it exists.
* In `conversation_history_widget.rs`, `add_session_info()` checks to
see whether a `WelcomeMessage` exists in the history when determining
the value of `has_welcome_message`. By construction, we expect that
`WelcomeMessage` is always the first message (in which case the existing
`let is_first_event = self.entries.is_empty();` logic would be sound),
but we decide to be extra defensive in case an `EventMsg::Error` is
processed before `EventMsg::SessionConfigured`.
2025-05-17 09:00:23 -07:00
Christoph K
031df77dfb Remove unnecessary console log from test (#970)
When running `npm test` on `codex-cli`, the test
`agent-cancel-prev-response.test.ts` logs a significant body of text to
console for no obvious reason.

This is not helpful, as it makes test logs messy and far longer.

This change deletes the `console.log(...)` that produces the behavior.
2025-05-16 19:48:11 -07:00
Michael Bolin
f9143d0361 fix: do not let Tab keypress flow through to composer when used to toggle focus (#977)
One line fix from Codex!
2025-05-16 19:27:49 -07:00
Fouad Matin
2880925a44 bump(version): 0.1.2505161800 (#978)
## `0.1.2505161800`

- Sign in with chatgpt credits (#974)
- Add support for OpenAI tool type, local_shell (#961)
2025-05-16 18:18:15 -07:00
Fouad Matin
3e19e8fd59 add: sign in with chatgpt credits (#974) 2025-05-16 17:55:08 -07:00
Trevor Creech
c7312c9d52 Fix CLA link in workflow (#964)
## Summary
- fix the CLA link posted by the bot
- docs suggest using an absolute URL:
https://github.com/marketplace/actions/cla-assistant-lite
2025-05-16 17:11:57 -07:00
Michael Bolin
1dc14cefa1 fix: make codex-mini-latest the default model in the Rust TUI (#972)
It's time to make `codex-mini-latest` the new default, as this should be
an "evergreen" model pointer.

* Equivalent change in TypeScript
https://github.com/openai/codex/pull/951
* See some notes about using `codex-mini-latest` with MCP in
https://github.com/openai/codex/pull/961
2025-05-16 17:08:18 -07:00
Michael Bolin
7ca84087e6 feat: make it possible to toggle mouse mode in the Rust TUI (#971)
I did a bit of research to understand why I could not use my mouse to
drag to select text to copy to the clipboard in iTerm.

Apparently https://github.com/openai/codex/pull/641 to enable mousewheel
scrolling broke this functionality. It seems that, unless we put in a
bit of effort, we can have drag-to-select or scrolling, but not both.
Though if you know the trick to hold down `Option` will dragging with
the mouse in iTerm, you can probably get by with this. (I did not know
about this option prior to researching this issue.)

Nevertheless, users may still prefer to disable mouse capture
altogether, so this PR introduces:

* the ability to set `tui.disable_mouse_capture = true` in `config.toml`
to disable mouse capture
* a new command, `/toggle-mouse-mode` to toggle mouse capture
2025-05-16 16:16:50 -07:00
Michael Bolin
67ac8ef605 fix: use text other than 'TODO' as test example (#969)
I casually `rg TODO` to look for TODOs, so the use of TODO in a sample
string in test output was throwing things off.
2025-05-16 14:51:03 -07:00
Michael Bolin
f48dd99f22 feat: add support for OpenAI tool type, local_shell (#961)
The new `codex-mini-latest` model expects a new tool with `{"type":
"local_shell"}`. Its contract is similar to the existing `function` tool
with `"name": "shell"`, so this takes the `local_shell` tool call into
`ExecParams` and sends it through the existing
`handle_container_exec_with_params()` code path.

This also adds the following logic when adding the default set of tools
to a request:

```rust
let default_tools = if self.model.starts_with("codex") {
    &DEFAULT_CODEX_MODEL_TOOLS
} else {
    &DEFAULT_TOOLS
};
```

That is, if the model name starts with `"codex"`, we add `{"type":
"local_shell"}` to the list of tools; otherwise, we add the
aforementioned `shell` tool.

To test this, I ran the TUI with `-m codex-mini-latest` and verified
that it used the `local_shell` tool. Though I also had some entries in
`[mcp_servers]` in my personal `config.toml`. The `codex-mini-latest`
model seemed eager to try the tools from the MCP servers first, so I
have personally commented them out for now, so keep an eye out if you're
testing `codex-mini-latest`!

Perhaps we should include more details with `{"type": "local_shell"}` or
update the following:


fd0b1b0208/codex-rs/core/prompt.md

For reference, the corresponding change in the TypeScript CLI is
https://github.com/openai/codex/pull/951.
2025-05-16 14:38:08 -07:00
Michael Bolin
dfd54e1433 chore: refactor handle_function_call() into smaller functions (#965)
Overall, `codex.rs` is still far too large, but at least there's less
indenting now that things have been moved into smaller functions.

This will also make it easier to introduce the `local_shell` tool in
https://github.com/openai/codex/pull/961.

---
[//]: # (BEGIN SAPLING FOOTER)
Stack created with [Sapling](https://sapling-scm.com). Best reviewed
with [ReviewStack](https://reviewstack.dev/openai/codex/pull/965).
* #961
* __->__ #965
2025-05-16 14:17:10 -07:00
Michael Bolin
9739820366 fix: remove file named ">" in the codex-cli folder (#968)
This file appears to have been accidentally added in
https://github.com/openai/codex/pull/912. Its presence makes the repo
impossible to clone on Windows.
2025-05-16 14:08:23 -07:00
Fouad Matin
fd0b1b0208 bump(version): 0.1.2505161243 (#967)
## `0.1.2505161243`

- Sign in with chatgpt (#963)
- Session history viewer (#912)
- Apply patch issue when using different cwd (#942)
- Diff command for filenames with special characters (#954)
2025-05-16 12:46:52 -07:00
Fouad Matin
c6e08ad8c1 add: sign in with chatgpt (#963)
Sign in with ChatGPT to get an API key (flow to grant API credits for Plus/Pro coming later today!)
2025-05-16 12:28:54 -07:00
Fouad Matin
cabf83f2ed add: session history viewer (#912)
- A new “/sessions” command is available for browsing previous sessions,
as shown in the updated slash command list

- The CLI now documents and parses a new “--history” flag to browse past
sessions from the command line

- A dedicated `SessionsOverlay` component loads session metadata and
allows toggling between viewing and resuming sessions

- When the sessions overlay is opened during a chat, selecting a session
can either show the saved rollout or resume it
2025-05-16 12:28:22 -07:00
Michael Bolin
1e39189393 feat: add support for file_opener option in Rust, similiar to #911 (#957)
This ports the enhancement introduced in
https://github.com/openai/codex/pull/911 (and the fixes in
https://github.com/openai/codex/pull/919) for the TypeScript CLI to the
Rust one.
2025-05-16 11:33:08 -07:00
Michael Bolin
3d9f4fcd8a fix: introduce ExtractHeredocError that implements PartialEq (#958) 2025-05-16 09:42:27 -07:00
Sebastian Lund
84e01f4b62 fix: apply patch issue when using different cwd (#942)
If you run a codex instance outside of the current working directory
from where you launched the codex binary it won't be able to apply
patches correctly, even if the sandbox policy allows it. This manifests
weird behaviours, such as

* Reading the same filename in the binary working directory, and
overwriting it in the session working directory. e.g. if you have a
`readme` in both folders it will overwrite the readme in the session
working directory with the readme in the binary working directory
*applied with the suggested patch*.
* The LLM ends up in weird loops trying to verify and debug why the
apply_patch won't work, and it can result in it applying patches by
manually writing python or javascript if it figures out that either is
supported by the system instead.

I added a test-case to ensure that the patch contents are based on the
cwd.

## Issue: mixing relative & absolute paths in apply_patch

1. The apply_patch tool use relative paths based on the session working
directory.
2. `unified_diff_from_chunks` eventually ends up [reading the source
file](https://github.com/reflectionai/codex/blob/main/codex-rs/apply-patch/src/lib.rs#L410)
to figure out what the diff is, by using the relative path.
3. The changes are targeted using an absolute path derived from the
current working directory.

The end-result in case session working directory differs from the binary
working directory: we get the diff for a file relative to the binary
working directory, and apply it on a file in the session working
directory.
2025-05-16 09:12:16 -07:00
hanson-openai
7edfbae062 fix: diff command for filenames with special characters (#954)
## Summary
- fix quoting issues in `/diff` to correctly handle files with special
characters
- add regression test for `getGitDiff` when filenames contain `$`
- relax timeout in raw-exec-process-group test

Fixes https://github.com/openai/codex/issues/943

## Testing
- `pnpm test`
2025-05-16 09:10:44 -07:00
Fouad Matin
316289d01d bump(version): 0.1.2505160811 codex-mini-latest (#953)
## `0.1.2505160811`

- `codex-mini-latest` (#951)
2025-05-16 08:18:20 -07:00
Michael Bolin
30cbfdfa87 chore: update exec crate to use std::time instead of chrono (#952)
When I originally wrote `elapsed.rs`, I realized we were using both
`std::time` and `chrono` with no real benefit of having both. We should
try to keep the `exec` subcommand trim (as it also buildable as a
standalone executable), so this helps tighten things up.
2025-05-16 08:14:50 -07:00
Fouad Matin
070499f534 add: codex-mini-latest (#951)
💽

---------

Co-authored-by: Trevor Creech <tcreech@openai.com>
2025-05-16 08:04:00 -07:00
Michael Bolin
ce2ecbe72f feat: record messages from user in ~/.codex/history.jsonl (#939)
This is a large change to support a "history" feature like you would
expect in a shell like Bash.

History events are recorded in `$CODEX_HOME/history.jsonl`. Because it
is a JSONL file, it is straightforward to append new entries (as opposed
to the TypeScript file that uses `$CODEX_HOME/history.json`, so to be
valid JSON, each new entry entails rewriting the entire file). Because
it is possible for there to be multiple instances of Codex CLI writing
to `history.jsonl` at once, we use advisory file locking when working
with `history.jsonl` in `codex-rs/core/src/message_history.rs`.

Because we believe history is a sufficiently useful feature, we enable
it by default. Though to provide some safety, we set the file
permissions of `history.jsonl` to be `o600` so that other users on the
system cannot read the user's history. We do not yet support a default
list of `SENSITIVE_PATTERNS` as the TypeScript CLI does:


3fdf9df133/codex-cli/src/utils/storage/command-history.ts (L10-L17)

We are going to take a more conservative approach to this list in the
Rust CLI. For example, while `/\b[A-Za-z0-9-_]{20,}\b/` might exclude
sensitive information like API tokens, it would also exclude valuable
information such as references to Git commits.

As noted in the updated documentation, users can opt-out of history by
adding the following to `config.toml`:

```toml
[history]
persistence = "none" 
```

Because `history.jsonl` could, in theory, be quite large, we take a[n
arguably overly pedantic] approach in reading history entries into
memory. Specifically, we start by telling the client the current number
of entries in the history file (`history_entry_count`) as well as the
inode (`history_log_id`) of `history.jsonl` (see the new fields on
`SessionConfiguredEvent`).

The client is responsible for keeping new entries in memory to create a
"local history," but if the user hits up enough times to go "past" the
end of local history, then the client should use the new
`GetHistoryEntryRequest` in the protocol to fetch older entries.
Specifically, it should pass the `history_log_id` it was given
originally and work backwards from `history_entry_count`. (It should
really fetch history in batches rather than one-at-a-time, but that is
something we can improve upon in subsequent PRs.)

The motivation behind this crazy scheme is that it is designed to defend
against:

* The `history.jsonl` being truncated during the session such that the
index into the history is no longer consistent with what had been read
up to that point. We do not yet have logic to enforce a `max_bytes` for
`history.jsonl`, but once we do, we will aspire to implement it in a way
that should result in a new inode for the file on most systems.
* New items from concurrent Codex CLI sessions amending to the history.
Because, in absence of truncation, `history.jsonl` is an append-only
log, so long as the client reads backwards from `history_entry_count`,
it should always get a consistent view of history. (That said, it will
not be able to read _new_ commands from concurrent sessions, but perhaps
we will introduce a `/` command to reload latest history or something
down the road.)

Admittedly, my testing of this feature thus far has been fairly light. I
expect we will find bugs and introduce enhancements/fixes going forward.
2025-05-15 16:26:23 -07:00
Michael Bolin
3fdf9df133 chore: introduce AppEventSender to help fix clippy warnings and update to Rust 1.87 (#948)
Moving to Rust 1.87 introduced a clippy warning that
`SendError<AppEvent>` was too large.

In practice, the only thing we ever did when we got this error was log
it (if the mspc channel is closed, then the app is likely shutting down
or something, so there's not much to do...), so this finally motivated
me to introduce `AppEventSender`, which wraps
`std::sync::mpsc::Sender<AppEvent>` with a `send()` method that invokes
`send()` on the underlying `Sender` and logs an `Err` if it gets one.

This greatly simplifies the code, as many functions that previously
returned `Result<(), SendError<AppEvent>>` now return `()`, so we don't
have to propagate an `Err` all over the place that we don't really
handle, anyway.

This also makes it so we can upgrade to Rust 1.87 in CI.
2025-05-15 14:50:30 -07:00
Michael Bolin
ec5e82b77c chore: pin Rust version to 1.86 and use io::Error::other to prepare for 1.87 (#947)
Previously, our GitHub actions specified the Rust toolchain as
`dtolnay/rust-toolchain@stable`, which meant the version could change
out from under us. In this case, the move from 1.86 to 1.87 introduced
new clippy warnings, causing build failures.

Because it will take a little time to fix all the new clippy warnings,
this PR pins things to 1.86 for now to unbreak the build.

It also replaces `io::Error::new(io::ErrorKind::Other)` with
`io::Error::other()` in preparation for 1.87.
2025-05-15 14:07:16 -07:00
Michael Bolin
5fc9fc3e3e chore: expose codex_home via Config (#941) 2025-05-15 00:30:13 -07:00
Michael Bolin
0b9ef93da5 fix: properly wrap lines in the Rust TUI (#937)
As discussed on
699ec5a87f (commitcomment-156776835),
to properly support scrolling long content in Ratatui for a sequence of
cells, we need to:

* take the `Vec<Line>` for each cell
* using the wrapping logic we want to use at render time, compute the
_effective line count_ using `Paragraph::line_count()` (see
`wrapped_line_count_for_cell()` in this PR)
* sum up the effective line count to compute the height of the area
being scrolled
* given a `scroll_position: usize`, index into the list of "effective
lines" and accumulate the appropriate `Vec<Line>` for the cells that
should be displayed
* take that `Vec<Line>` to create a `Paragraph` and use the same
line-wrapping policy that was used in `wrapped_line_count_for_cell()`
* display the resulting `Paragraph` and use the accounting to display a
scrollbar with the appropriate thumb size and offset without having to
render the `Vec<Line>` for the full history

With this change, lines wrap as I expect and everything appears to
redraw correctly as I resize my terminal!
2025-05-14 16:51:41 -07:00
Michael Bolin
34aa1991f1 chore: handle all cases for EventMsg (#936)
For now, this removes the `#[non_exhaustive]` directive on `EventMsg` so
that we are forced to handle all `EventMsg` by default. (We may revisit
this if/when we publish `core/` as a `lib` crate.) For now, it is
helpful to have this as a forcing function because we have effectively
two UIs (`tui` and `exec`) and usually when we add a new variant to
`EventMsg`, we want to be sure that we update both.
2025-05-14 13:36:43 -07:00
Michael Bolin
497c5396c0 feat: add mcp subcommand to CLI to run Codex as an MCP server (#934)
Previously, running Codex as an MCP server required a standalone binary
in our Cargo workspace, but this PR makes it available as a subcommand
(`mcp`) of the main CLI.

Ran this with:

```
RUST_LOG=debug npx @modelcontextprotocol/inspector cargo run --bin codex -- mcp
```

and verified it worked as expected in the inspector at
`http://127.0.0.1:6274/`.
2025-05-14 13:15:41 -07:00
Michael Bolin
a12e4b0b31 feat: add support for commands in the Rust TUI (#935)
Introduces support for slash commands like in the TypeScript CLI. We do
not support the full set of commands yet, but the core abstraction is
there now.

In particular, we have a `SlashCommand` enum and due to thoughtful use
of the [strum](https://crates.io/crates/strum) crate, it requires
minimal boilerplate to add a new command to the list.

The key new piece of UI is `CommandPopup`, though the keyboard events
are still handled by `ChatComposer`. The behavior is roughly as follows:

* if the first character in the composer is `/`, the command popup is
displayed (if you really want to send a message to Codex that starts
with a `/`, simply put a space before the `/`)
* while the popup is displayed, up/down can be used to change the
selection of the popup
* if there is a selection, hitting tab completes the command, but does
not send it
* if there is a selection, hitting enter sends the command
* if the prefix of the composer matches a command, the command will be
visible in the popup so the user can see the description (commands could
take arguments, so additional text may appear after the command name
itself)


https://github.com/user-attachments/assets/39c3e6ee-eeb7-4ef7-a911-466d8184975f

Incidentally, Codex wrote almost all the code for this PR!
2025-05-14 12:55:49 -07:00
Michael Bolin
0402aef126 chore: move each view used in BottomPane into its own file (#928)
`BottomPane` was getting a bit unwieldy because it maintained a
`PaneState` enum with three variants and many of its methods had `match`
statements to handle each variant. To replace the enum, this PR:

* Introduces a `trait BottomPaneView` that has two implementations:
`StatusIndicatorView` and `ApprovalModalView`.
* Migrates `PaneState::TextInput` into its own struct, `ChatComposer`,
that does **not** implement `BottomPaneView`.
* Updates `BottomPane` so it has `composer: ChatComposer` and
`active_view: Option<Box<dyn BottomPaneView<'a> + 'a>>`. The idea is
that `active_view` takes priority and is displayed when it is `Some`;
otherwise, `ChatComposer` is displayed.
* While methods of `BottomPane` often have to check whether
`active_view` is present to decide which component to delegate to, the
code is more straightforward than before and introducing new
implementations of `BottomPaneView` should be less painful.

Because we want to retain the `TextArea` owned by `ChatComposer` even
when another view is displayed, to keep the ownership logic simple, it
seemed best to keep `ChatComposer` distinct from `BottomPaneView`.
2025-05-14 10:13:29 -07:00
Michael Bolin
399e819c9b fix: increase timeout for test_dev_null_write (#933)
After updating this test in https://github.com/openai/codex/pull/923, I
have been getting some timeouts with this test in CI, so increasing the
timeout to match that of `test_writable_root`:


327cf41f0f/codex-rs/core/src/landlock.rs (L211-L213)
2025-05-14 10:06:14 -07:00
Yaroslav Halchenko
327cf41f0f Add codespell support (config, workflow to detect/not fix) and make it fix some typos (#903)
More about codespell: https://github.com/codespell-project/codespell .

I personally introduced it to dozens if not hundreds of projects already
and so far only positive feedback.

CI workflow has 'permissions' set only to 'read' so also should be safe.

Let me know if just want to take typo fixes in and get rid of the CI

---------

Signed-off-by: Yaroslav O. Halchenko <debian@onerussian.com>
2025-05-14 09:39:49 -07:00
Fouad Matin
9e7cd2b25a bump(version): 0.1.2505140839 (#932)
## `0.1.2505140839`

### 🪲 Bug Fixes

- Gpt-4.1 apply_patch handling (#930)
- Add support for fileOpener in config.json (#911)
- Patch in #366 and #367 for marked-terminal (#916)
- Remember to set lastIndex = 0 on shared RegExp (#918)
- Always load version from package.json at runtime (#909)
- Tweak the label for citations for better rendering (#919)
- Tighten up some logic around session timestamps and ids (#922)
- Change EventMsg enum so every variant takes a single struct (#925)
- Reasoning default to medium, show workdir when supplied (#931)
- Test_dev_null_write() was not using echo as intended (#923)
2025-05-14 08:44:52 -07:00
Fouad Matin
73259351ff fix: reasoning default to medium, show workdir when supplied (#931) 2025-05-14 08:38:41 -07:00
Fouad Matin
77347d268d fix: gpt-4.1 apply_patch handling (#930) 2025-05-14 08:34:09 -07:00
Fouad Matin
678f0dbfec add: dynamic instructions (#927) 2025-05-14 01:27:46 -07:00
Michael Bolin
1bf00a3a95 feat: Ctrl+J for newline in Rust TUI, default to one line of height (#926)
While the `TextArea` used in the Rust TUI is "multiline," it is not like
an HTML `<textarea>` in that it does not wrap, so there was not much
benefit to setting `MIN_TEXTAREA_ROWS` to `3`, so this PR changes it to
`1`. Though there are now three ways to "increase" the height due to
actual linebreaks:

* paste in multiline content (this worked before this PR)
* pressing `Ctrl+J` will insert a newline
* if you have your terminal emulator set such that it is possible to
press something that `crossterm` interprets as "Enter plus some
modifier," then now that will also work

Now things look a bit more compact on startup:

<img width="745" alt="image"
src="https://github.com/user-attachments/assets/86e2857f-f31c-46f5-a80b-1ab2120b266e"
/>
2025-05-13 21:42:14 -07:00
Michael Bolin
5bf9445351 fix: test_dev_null_write() was not using echo as intended (#923)
I believe this test meant to verify that echoing content to `/dev/null`
succeeded, but instead, I believe it was testing the equivalent to `echo
'blah > /dev/null'`.
2025-05-13 21:40:26 -07:00
Michael Bolin
a5f3a34827 fix: change EventMsg enum so every variant takes a single struct (#925)
https://github.com/openai/codex/pull/922 did this for the
`SessionConfigured` enum variant, and I think it is generally helpful to
be able to work with the values as each enum variant as their own type,
so this converts the remaining variants and updates all of the
callsites.

Added a simple unit test to verify that the JSON-serialized version of
`Event` does not have any unexpected nesting.
2025-05-13 20:44:42 -07:00
Michael Bolin
e6c206d19d fix: tighten up some logic around session timestamps and ids (#922)
* update `SessionConfigured` event to include the UUID for the session
* show the UUID in the Rust TUI
* use local timestamps in log files instead of UTC
* include timestamps in log file names for easier discovery
2025-05-13 19:22:16 -07:00
Michael Bolin
3c03c25e56 feat: introduce --profile for Rust CLI (#921)
This introduces a much-needed "profile" concept where users can specify
a collection of options under one name and then pass that via
`--profile` to the CLI.

This PR introduces the `ConfigProfile` struct and makes it a field of
`CargoToml`. It further updates
`Config::load_from_base_config_with_overrides()` to respect
`ConfigProfile`, overriding default values where appropriate. A detailed
unit test is added at the end of `config.rs` to verify this behavior.

Details on how to use this feature have also been added to
`codex-rs/README.md`.
2025-05-13 16:52:52 -07:00
Adeeb
ae809f3721 restructure flake for codex-rs (#888)
Right now since the repo is having two different implementations of
codex, flake was updated to work with both typescript implementation and
rust implementation
2025-05-13 13:08:42 -07:00
Michael Bolin
a786c1d188 feat: auto-approve nl and support piping to sed (#920)
Auto-approved:

```
["nl", "-ba", "README.md"]
["sed", "-n", "1,200p", "filename.txt"]
["bash", "-lc", "sed -n '1,200p' filename.txt"]
["bash", "-lc", "nl -ba README.md | sed -n '1,200p'"]
```

Not auto approved:

```
["sed", "-n", "'1,200p'", "filename.txt"]
["sed", "-n", "1,200p", "file1.txt", "file2.txt"]
```
2025-05-13 13:06:35 -07:00
Michael Bolin
0ac7e8d55b fix: tweak the label for citations for better rendering (#919)
Adds a space so that sequential citations have some more breathing room.

As I had to update the tests for this change, I also introduced a
`toDiffableString()` helper to make the test easier to update as we make
formatting changes to the output.
2025-05-13 12:46:21 -07:00
Michael Bolin
1ff3e14d5a fix: patch in #366 and #367 for marked-terminal (#916)
This PR uses [`pnpm
patch`](https://www.petermekhaeil.com/til/pnpm-patch/) to pull in the
following proposed fixes for `marked-terminal`:

* https://github.com/mikaelbr/marked-terminal/pull/366
* https://github.com/mikaelbr/marked-terminal/pull/367

This adds a substantial test to `codex-cli/tests/markdown.test.tsx` to
verify the new behavior.

Note that one of the tests shows two citations being split across a line
even though the rendered version would fit comfortably on one line.
Changing this likely requires a subtle fix to `marked-terminal` to
account for "rendered length" when determining line breaks.
2025-05-13 12:29:17 -07:00
Michael Bolin
dd354e2134 fix: remember to set lastIndex = 0 on shared RegExp (#918)
I had not observed an issue in the wild because of this yet, but it
feels like it was only a matter of time...
2025-05-13 12:01:06 -07:00
Michael Bolin
557f608f25 fix: add support for fileOpener in config.json (#911)
This PR introduces the following type:

```typescript
export type FileOpenerScheme = "vscode" | "cursor" | "windsurf";
```

and uses it as the new type for a `fileOpener` option in `config.json`.
If set, this will be used to linkify file annotations in the output
using the URI-based file opener supported in VS Code-based IDEs.

Currently, this does not pass:

Updated `codex-cli/tests/markdown.test.tsx` to verify the new behavior.
Note it required mocking `supports-hyperlinks` and temporarily modifying
`chalk.level` to yield the desired output.
2025-05-13 09:45:46 -07:00
Michael Bolin
05bb5d7d46 fix: always load version from package.json at runtime (#909)
Note the high-level motivation behind this change is to avoid the need
to make temporary changes in the source tree in order to cut a release
build since that runs the risk of leaving things in an inconsistent
state in the event of a failure. The existing code:

```
import pkg from "../../package.json" assert { type: "json" };
```

did not work as intended because, as written, ESBuild would bake the
contents of the local `package.json` into the release build at build
time whereas we want it to read the contents at runtime so we can use
the `package.json` in the tree to build the code and later inject a
modified version into the release package with a timestamped build
version.

Changes:

* move `CLI_VERSION` out of `src/utils/session.ts` and into
`src/version.ts` so `../package.json` is a correct relative path both
from `src/version.ts` in the source tree and also in the final
`dist/cli.js` build output
* change `assert` to `with` in `import pkg` as apparently `with` became
standard in Node 22
* mark `"../package.json"` as external in `build.mjs` so the version is
not baked into the `.js` at build time

After using `pnpm stage-release` to build a release version, if I use
Node 22.0 to run Codex, I see the following printed to stderr at
startup:

```
(node:71308) ExperimentalWarning: Importing JSON modules is an experimental feature and might change at any time
(Use `node --trace-warnings ...` to show where the warning was created)
```

Note it is a warning and does not prevent Codex from running.

In Node 22.12, the warning goes away, but the warning still appears in
Node 22.11. For Node 22, 22.15.0 is the current LTS version, so LTS
users will not see this.

Also, something about moving the definition of `CLI_VERSION` caused a
problem with the mocks in `check-updates.test.ts`. I asked Codex to fix
it, and it came up with the change to the test configs. I don't know
enough about vitest to understand what it did, but the tests seem
healthy again, so I'm going with it.
2025-05-12 21:27:15 -07:00
Michael Bolin
61b881d4e5 fix: agent instructions were not being included when ~/.codex/instructions.md was empty (#908)
I had seen issues where `codex-rs` would not always write files without
me pressuring it to do so, and between that and the report of
https://github.com/openai/codex/issues/900, I decided to look into this
further. I found two serious issues with agent instructions:

(1) We were only sending agent instructions on the first turn, but
looking at the TypeScript code, we should be sending them on every turn.

(2) There was a serious issue where the agent instructions were
frequently lost:

* The TypeScript CLI appears to keep writing `~/.codex/instructions.md`:
55142e3e6c/codex-cli/src/utils/config.ts (L586)
* If `instructions.md` is present, the Rust CLI uses the contents of it
INSTEAD OF the default prompt, even if `instructions.md` is empty:
55142e3e6c/codex-rs/core/src/config.rs (L202-L203)

The combination of these two things means that I have been using
`codex-rs` without these key instructions:
https://github.com/openai/codex/blob/main/codex-rs/core/prompt.md

Looking at the TypeScript code, it appears we should be concatenating
these three items every time (if they exist):

* `prompt.md`
* `~/.codex/instructions.md`
* nearest `AGENTS.md`

This PR fixes things so that:

* `Config.instructions` is `None` if `instructions.md` is empty
* `Payload.instructions` is now `&'a str` instead of `Option<&'a
String>` because we should always have _something_ to send
* `Prompt` now has a `get_full_instructions()` helper that returns a
`Cow<str>` that will always include the agent instructions first.
2025-05-12 17:24:44 -07:00
Michael Bolin
55142e3e6c fix: use "thinking" instead of "codex reasoning" as the label for reasoning events in the TUI (#905) 2025-05-12 15:19:45 -07:00
Michael Bolin
115fb0b95d fix: navigate initialization phase before tools/list request in MCP client (#904)
Apparently the MCP server implemented in JavaScript did not require the
`initialize` handshake before responding to tool list/call, so I missed
this.
2025-05-12 15:15:26 -07:00
Avi Rosenberg
ab4cb94227 fix: Normalize paths in resolvePathAgainstWorkdir to prevent path traversal vulnerability (#895)
This PR fixes a potential path traversal vulnerability by ensuring all
paths are properly normalized in the `resolvePathAgainstWorkdir`
function.

## Changes
- Added path normalization for both absolute and relative paths
- Ensures normalized paths are used in all subsequent operations
- Prevents potential path traversal attacks through non-normalized paths

This minimal change addresses the security concern without adding
unnecessary complexity, while maintaining compatibility with existing
code.
2025-05-12 13:44:00 -07:00
Michael Bolin
73fe1381aa chore: introduce new --native flag to Node module release process (#844)
This PR introduces an optional build flag, `--native`, that will build a
version of the Codex npm module that:

- Includes both the Node.js and native Rust versions (for Mac and Linux)
- Will run the native version if `CODEX_RUST=1` is set
- Runs the TypeScript version otherwise

Note this PR also updates the workflow URL to
https://github.com/openai/codex/actions/runs/14872557396, as that is a
build from today that includes everything up through
https://github.com/openai/codex/pull/843.

Test Plan:

In `~/code/codex/codex-cli`, I ran:

```
pnpm stage-release --native
```

The end of the output was:

```
Staged version 0.1.2505121317 for release in /var/folders/wm/f209bc1n2bd_r0jncn9s6j_00000gp/T/tmp.xd2p5ETYGN
Test Node:
    node /var/folders/wm/f209bc1n2bd_r0jncn9s6j_00000gp/T/tmp.xd2p5ETYGN/bin/codex.js --help
Test Rust:
    CODEX_RUST=1 node /var/folders/wm/f209bc1n2bd_r0jncn9s6j_00000gp/T/tmp.xd2p5ETYGN/bin/codex.js --help
Next:  cd "/var/folders/wm/f209bc1n2bd_r0jncn9s6j_00000gp/T/tmp.xd2p5ETYGN" && npm publish --tag native
```

I verified that running each of these commands ran the expected version
of Codex.

While here, I also added `bin` to the `files` list in `package.json`,
which should have been done as part of
https://github.com/openai/codex/pull/757, as that added new entries to
`bin` that were matched by `.gitignore` but should have been included in
a release.
2025-05-12 13:38:10 -07:00
jcoens-openai
f3bd143867 Disallow expect via lints (#865)
Adds `expect()` as a denied lint. Same deal applies with `unwrap()`
where we now need to put `#[expect(...` on ones that we legit want. Took
care to enable `expect()` in test contexts.

# Tests

```
cargo fmt
cargo clippy --all-features --all-targets --no-deps -- -D warnings
cargo test
```
2025-05-12 08:45:46 -07:00
Michael Bolin
a1f51bf91b fix: fix border style for BottomPane (#893)
This PR fixes things so that:

* when the `BottomPane` is in the `StatusIndicator` state, the border
should be dim
* when the `BottomPane` does not have input focus, the border should be
dim

To make it easier to enforce this invariant, this PR introduces
`BottomPane::set_state()` that will:

* update `self.state`
* call `update_border_for_input_focus()`
* request a repaint

This should make it easier to enforce other updates for state changes
going forward.
2025-05-10 23:34:13 -07:00
Michael Bolin
b4785b5f88 feat: include "reasoning" messages in Rust TUI (#892)
As shown in the screenshot, we now include reasoning messages from the
model in the TUI under the heading "codex reasoning":


![image](https://github.com/user-attachments/assets/d8eb3dc3-2f9f-4e95-847e-d24b421249a8)

To ensure these are visible by default when using `o4-mini`, this also
changes the default value for `summary` (formerly `generate_summary`,
which is deprecated in favor of `summary` according to the docs) from
unset to `"auto"`.
2025-05-10 21:43:27 -07:00
Michael Bolin
2b122da087 feat: add support for AGENTS.md in Rust CLI (#885)
The TypeScript CLI already has support for including the contents of
`AGENTS.md` in the instructions sent with the first turn of a
conversation. This PR brings this functionality to the Rust CLI.

To be considered, `AGENTS.md` must be in the `cwd` of the session, or in
one of the parent folders up to a Git/filesystem root (whichever is
encountered first).

By default, a maximum of 32 KiB of `AGENTS.md` will be included, though
this is configurable using the new-in-this-PR `project_doc_max_bytes`
option in `config.toml`.
2025-05-10 17:52:59 -07:00
Corry Haines
b42ad670f1 fix: flex-mode via config/flag (#813)
* Add flexMode to stored config, and use it during config loading unless
the flag is explicitly passed.
* If the config asks for flexMode and the model doesn't support it,
silently disable flexMode.

Resolves #803
2025-05-10 16:18:20 -07:00
Pranav
646e7e9c11 feat: added arceeai as a provider (#818)
- Added ArceeAI as a provider  - https://conductor.arcee.ai/v1
- Compatible with ArceeAI SLMs (Virtuoso, Maestro)
- Works with ArceeAI's Conductor auto‑router models (auto, auto‑tool),
once #817 is merged
2025-05-10 16:16:28 -07:00
Pranav
19262f632f fix: guard against missing choices (#817)
- Fixes guard by using optional chaining to safely check
chunk.choices?.[0] before accessing.
- Currently, accessing chunk.choices[0] without checking could throw if
choices was missing from the chunk.
2025-05-10 16:16:19 -07:00
Corry Haines
fcc76cf3e7 Add reasoning effort option to CLI help text (#815)
Reasoning effort was already available, but not expressed into the help
text, so it was non-discoverable.

Other issues discovered, but will fix in separate PR since they are
larger:
* #816 reasoningEffort isn't displayed in the terminal-header, making it
rather hard to see the state of configuration
* I don't think the config file setting works, as the CLI option always
"wins" and overwrites it
2025-05-10 15:58:59 -07:00
Fouad Matin
3104d81b7b fix: migrate to AGENTS.md (#764)
Migrate from `codex.md` to `AGENTS.md`
2025-05-10 15:57:49 -07:00
Tomas Cupr
e307d007aa fix: retry on OpenAI server_error even without status code (#814)
Fix: retry on server_error responses that lack an HTTP status code

### What happened

1. An OpenAI endpoint returned a **5xx** (transient server-side
failure).
2. The SDK surfaced it as an `APIError` with

{ "type": "server_error", "message": "...", "status": undefined }

           (The SDK does not always populate `status` for these cases.)
3. Our retry logic in `src/utils/agent/agent-loop.ts` determined

isServerError = typeof status === "number" && status >= 500;

Because `status` was *undefined*, the error was **not** recognised as
retriable, the exception bubbled out, and the CLI crashed with a stack
           trace similar to:

               Error: An error occurred while processing the request.
                   at .../cli.js:474:1514

### Root cause

The transient-error detector ignored the semantic flag type ===
"server_error" that the SDK provides when the numeric status is missing.

#### Fix (1 loc + comment)

Extend the check:

const status = errCtx?.status ?? errCtx?.httpStatus ??
errCtx?.statusCode;

const isServerError = (typeof status === "number" && status >= 500) ||
// classic 5xx
errCtx?.type === "server_error";                   // <-- NEW

Now the agent:

* Retries up to **5** times (existing logic) when the backend reports a
transient failure, even if `status` is absent.
* If all retries fail, surfaces the existing friendly system message
instead of an uncaught exception.

### Tests & validation

pnpm test # all suites green (17 agent-level tests now include this
path)
pnpm run lint    # 0 errors / warnings
pnpm run typecheck

A new unit-test file isn’t required—the behaviour is already covered by
tests/agent-server-retry.test.ts, which stubs type: "server_error" and
now passes with the updated logic.

### Impact

* No API-surface changes.
* Prevents CLI crashes on intermittent OpenAI outages.
* Adds robust handling for other providers that may follow the same
error-shape.
2025-05-10 15:43:03 -07:00
Michael Bolin
fde48aaa0d feat: experimental env var: CODEX_SANDBOX_NETWORK_DISABLED (#879)
When using Codex to develop Codex itself, I noticed that sometimes it
would try to add `#[ignore]` to the following tests:

```
keeps_previous_response_id_between_tasks()
retries_on_early_close()
```

Both of these tests start a `MockServer` that launches an HTTP server on
an ephemeral port and requires network access to hit it, which the
Seatbelt policy associated with `--full-auto` correctly denies. If I
wasn't paying attention to the code that Codex was generating, one of
these `#[ignore]` annotations could have slipped into the codebase,
effectively disabling the test for everyone.

To that end, this PR enables an experimental environment variable named
`CODEX_SANDBOX_NETWORK_DISABLED` that is set to `1` if the
`SandboxPolicy` used to spawn the process does not have full network
access. I say it is "experimental" because I'm not convinced this API is
quite right, but we need to start somewhere. (It might be more
appropriate to have an env var like `CODEX_SANDBOX=full-auto`, but the
challenge is that our newer `SandboxPolicy` abstraction does not map to
a simple set of enums like in the TypeScript CLI.)

We leverage this new functionality by adding the following code to the
aforementioned tests as a way to "dynamically disable" them:

```rust
if std::env::var(CODEX_SANDBOX_NETWORK_DISABLED_ENV_VAR).is_ok() {
    println!(
        "Skipping test because it cannot execute when network is disabled in a Codex sandbox."
    );
    return;
}
```

We can use the `debug seatbelt --full-auto` command to verify that
`cargo test` fails when run under Seatbelt prior to this change:

```
$ cargo run --bin codex -- debug seatbelt --full-auto -- cargo test
---- keeps_previous_response_id_between_tasks stdout ----

thread 'keeps_previous_response_id_between_tasks' panicked at /Users/mbolin/.cargo/registry/src/index.crates.io-1949cf8c6b5b557f/wiremock-0.6.3/src/mock_server/builder.rs:107:46:
Failed to bind an OS port for a mock server.: Os { code: 1, kind: PermissionDenied, message: "Operation not permitted" }
note: run with `RUST_BACKTRACE=1` environment variable to display a backtrace


failures:
    keeps_previous_response_id_between_tasks

test result: FAILED. 0 passed; 1 failed; 0 ignored; 0 measured; 0 filtered out; finished in 0.00s

error: test failed, to rerun pass `-p codex-core --test previous_response_id`
```

Though after this change, the above command succeeds! This means that,
going forward, when Codex operates on Codex itself, when it runs `cargo
test`, only "real failures" should cause the command to fail.

As part of this change, I decided to tighten up the codepaths for
running `exec()` for shell tool calls. In particular, we do it in `core`
for the main Codex business logic itself, but we also expose this logic
via `debug` subcommands in the CLI in the `cli` crate. The logic for the
`debug` subcommands was not quite as faithful to the true business logic
as I liked, so I:

* refactored a bit of the Linux code, splitting `linux.rs` into
`linux_exec.rs` and `landlock.rs` in the `core` crate.
* gating less code behind `#[cfg(target_os = "linux")]` because such
code does not get built by default when I develop on Mac, which means I
either have to build the code in Docker or wait for CI signal
* introduced `macro_rules! configure_command` in `exec.rs` so we can
have both sync and async versions of this code. The synchronous version
seems more appropriate for straight threads or potentially fork/exec.
2025-05-09 18:29:34 -07:00
Govind Kamtamneni
7795272282 Adds Azure OpenAI support (#769)
## Summary

This PR introduces support for Azure OpenAI as a provider within the
Codex CLI. Users can now configure the tool to leverage their Azure
OpenAI deployments by specifying `"azure"` as the provider in
`config.json` and setting the corresponding `AZURE_OPENAI_API_KEY` and
`AZURE_OPENAI_API_VERSION` environment variables. This functionality is
added alongside the existing provider options (OpenAI, OpenRouter,
etc.).

Related to #92

**Note:** This PR is currently in **Draft** status because tests on the
`main` branch are failing. It will be marked as ready for review once
the `main` branch is stable and tests are passing.

---

## What’s Changed

-   **Configuration (`config.ts`, `providers.ts`, `README.md`):**
- Added `"azure"` to the supported `providers` list in `providers.ts`,
specifying its name, default base URL structure, and environment
variable key (`AZURE_OPENAI_API_KEY`).
- Defined the `AZURE_OPENAI_API_VERSION` environment variable in
`config.ts` with a default value (`2025-03-01-preview`).
    -   Updated `README.md` to:
        -   Include "azure" in the list of providers.
- Add a configuration section for Azure OpenAI, detailing the required
environment variables (`AZURE_OPENAI_API_KEY`,
`AZURE_OPENAI_API_VERSION`) with examples.
- **Client Instantiation (`terminal-chat.tsx`, `singlepass-cli-app.tsx`,
`agent-loop.ts`, `compact-summary.ts`, `model-utils.ts`):**
- Modified various components and utility functions where the OpenAI
client is initialized.
- Added conditional logic to check if the configured `provider` is
`"azure"`.
- If the provider is Azure, the `AzureOpenAI` client from the `openai`
package is instantiated, using the configured `baseURL`, `apiKey` (from
`AZURE_OPENAI_API_KEY`), and `apiVersion` (from
`AZURE_OPENAI_API_VERSION`).
- Otherwise, the standard `OpenAI` client is instantiated as before.
-   **Dependencies:**
- Relies on the `openai` package's built-in support for `AzureOpenAI`.
No *new* external dependencies were added specifically for this Azure
implementation beyond the `openai` package itself.

---

## How to Test

*This has been tested locally and confirmed working with Azure OpenAI.*

1.  **Configure `config.json`:**
Ensure your `~/.codex/config.json` (or project-specific config) includes
Azure and sets it as the active provider:
    ```json
    {
      "providers": {
        // ... other providers
        "azure": {
          "name": "AzureOpenAI",
"baseURL": "https://YOUR_RESOURCE_NAME.openai.azure.com", // Replace
with your Azure endpoint
          "envKey": "AZURE_OPENAI_API_KEY"
        }
      },
      "provider": "azure", // Set Azure as the active provider
      "model": "o4-mini" // Use your Azure deployment name here
      // ... other config settings
    }
    ```
2.  **Set up Environment Variables:**
    ```bash
    # Set the API Key for your Azure OpenAI resource
    export AZURE_OPENAI_API_KEY="your-azure-api-key-here"

# Set the API Version (Optional - defaults to `2025-03-01-preview` if
not set)
# Ensure this version is supported by your Azure deployment and endpoint
    export AZURE_OPENAI_API_VERSION="2025-03-01-preview"
    ```
3.  **Get the Codex CLI by building from this PR branch:**
Clone your fork, checkout this branch (`feat/azure-openai`), navigate to
`codex-cli`, and build:
    ```bash
    # cd /path/to/your/fork/codex
    git checkout feat/azure-openai # Or your branch name
    cd codex-cli
    corepack enable
    pnpm install
    pnpm build
    ```
4.  **Invoke Codex:**
Run the locally built CLI using `node` from the `codex-cli` directory:
    ```bash
    node ./dist/cli.js "Explain the purpose of this PR"
    ```
*(Alternatively, if you ran `pnpm link` after building, you can use
`codex "Explain the purpose of this PR"` from anywhere)*.
5. **Verify:** Confirm that the command executes successfully and
interacts with your configured Azure OpenAI deployment.

---

## Tests

- [x] Tested locally against an Azure OpenAI deployment using API Key
authentication. Basic commands and interactions confirmed working.

---

## Checklist

- [x] Added Azure provider details to configuration files
(`providers.ts`, `config.ts`).
- [x] Implemented conditional `AzureOpenAI` client initialization based
on provider setting.
-   [x] Ensured `apiVersion` is passed correctly to the Azure client.
-   [x] Updated `README.md` with Azure OpenAI setup instructions.
- [x] Manually tested core functionality against a live Azure OpenAI
endpoint.
- [x] Add/update automated tests for the Azure code path (pending `main`
stability).

cc @theabhinavdas @nikodem-wrona @fouad-openai @tibo-openai (adjust as
needed)

---

I have read the CLA Document and I hereby sign the CLA
2025-05-09 18:11:32 -07:00
163 changed files with 8091 additions and 1906 deletions

1
.codespellignore Normal file
View File

@@ -0,0 +1 @@
iTerm

6
.codespellrc Normal file
View File

@@ -0,0 +1,6 @@
[codespell]
# Ref: https://github.com/codespell-project/codespell#using-a-config-file
skip = .git*,vendor,*-lock.yaml,*.lock,.codespellrc,*test.ts
check-hidden = true
ignore-regex = ^\s*"image/\S+": ".*|\b(afterAll)\b
ignore-words-list = ratatui,ser

View File

@@ -23,7 +23,7 @@ jobs:
env:
GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
with:
path-to-document: docs/CLA.md
path-to-document: https://github.com/openai/codex/blob/main/docs/CLA.md
path-to-signatures: signatures/cla.json
branch: cla-signatures
allowlist: dependabot[bot]

27
.github/workflows/codespell.yml vendored Normal file
View File

@@ -0,0 +1,27 @@
# Codespell configuration is within .codespellrc
---
name: Codespell
on:
push:
branches: [main]
pull_request:
branches: [main]
permissions:
contents: read
jobs:
codespell:
name: Check for spelling errors
runs-on: ubuntu-latest
steps:
- name: Checkout
uses: actions/checkout@v4
- name: Annotate locations with typos
uses: codespell-project/codespell-problem-matcher@b80729f885d32f78a716c2f107b4db1025001c42 # v1
- name: Codespell
uses: codespell-project/actions-codespell@406322ec52dd7b488e48c1c4b82e2a8b3a1bf630 # v2
with:
ignore_words_file: .codespellignore

View File

@@ -26,7 +26,9 @@ jobs:
steps:
- uses: actions/checkout@v4
- uses: dtolnay/rust-toolchain@stable
- uses: dtolnay/rust-toolchain@1.87
with:
components: rustfmt
- name: cargo fmt
run: cargo fmt -- --config imports_granularity=Item --check
@@ -58,9 +60,10 @@ jobs:
steps:
- uses: actions/checkout@v4
- uses: dtolnay/rust-toolchain@stable
- uses: dtolnay/rust-toolchain@1.87
with:
targets: ${{ matrix.target }}
components: clippy
- uses: actions/cache@v4
with:

View File

@@ -74,7 +74,7 @@ jobs:
steps:
- uses: actions/checkout@v4
- uses: dtolnay/rust-toolchain@stable
- uses: dtolnay/rust-toolchain@1.87
with:
targets: ${{ matrix.target }}
@@ -115,13 +115,42 @@ jobs:
- name: Compress artifacts
shell: bash
run: |
# Path that contains the uncompressed binaries for the current
# ${{ matrix.target }}
dest="dist/${{ matrix.target }}"
zstd -T0 -19 --rm "$dest"/*
# For compatibility with environments that lack the `zstd` tool we
# additionally create a `.tar.gz` alongside every single binary that
# we publish. The end result is:
# codex-<target>.zst (existing)
# codex-<target>.tar.gz (new)
# ...same naming for codex-exec-* and codex-linux-sandbox-*
# 1. Produce a .tar.gz for every file in the directory *before* we
# run `zstd --rm`, because that flag deletes the original files.
for f in "$dest"/*; do
base="$(basename "$f")"
# Skip files that are already archives (shouldn't happen, but be
# safe).
if [[ "$base" == *.tar.gz ]]; then
continue
fi
# Create per-binary tar.gz
tar -C "$dest" -czf "$dest/${base}.tar.gz" "$base"
# Also create .zst (existing behaviour) *and* remove the original
# uncompressed binary to keep the directory small.
zstd -T0 -19 --rm "$dest/$base"
done
- uses: actions/upload-artifact@v4
with:
name: ${{ matrix.target }}
path: codex-rs/dist/${{ matrix.target }}/*
# Upload the per-binary .zst files as well as the new .tar.gz
# equivalents we generated in the previous step.
path: |
codex-rs/dist/${{ matrix.target }}/*
release:
needs: build

4
.gitignore vendored
View File

@@ -77,3 +77,7 @@ yarn.lock
package.json-e
session.ts-e
CHANGELOG.ignore.md
# nix related
.direnv
.envrc

5
AGENTS.md Normal file
View File

@@ -0,0 +1,5 @@
# Rust/codex-rs
In the codex-rs folder where the rust code lives:
- Never add or modify any code related to `CODEX_SANDBOX_NETWORK_DISABLED_ENV_VAR`. You operate in a sandbox where `CODEX_SANDBOX_NETWORK_DISABLED=1` will be set whenever you use the `shell` tool. Any existing code that uses `CODEX_SANDBOX_NETWORK_DISABLED_ENV_VAR` was authored with this fact in mind. It is often used to early exit out of tests that the author knew you would not be able to run given your sandbox limitations.

View File

@@ -2,6 +2,48 @@
You can install any of these versions: `npm install -g codex@version`
## `0.1.2505172129`
### 🪲 Bug Fixes
- Add node version check (#1007)
- Persist token after refresh (#1006)
## `0.1.2505171619`
- `codex --login` + `codex --free` (#998)
## `0.1.2505161800`
- Sign in with chatgpt credits (#974)
- Add support for OpenAI tool type, local_shell (#961)
## `0.1.2505161243`
- Sign in with chatgpt (#963)
- Session history viewer (#912)
- Apply patch issue when using different cwd (#942)
- Diff command for filenames with special characters (#954)
## `0.1.2505160811`
- `codex-mini-latest` (#951)
## `0.1.2505140839`
### 🪲 Bug Fixes
- Gpt-4.1 apply_patch handling (#930)
- Add support for fileOpener in config.json (#911)
- Patch in #366 and #367 for marked-terminal (#916)
- Remember to set lastIndex = 0 on shared RegExp (#918)
- Always load version from package.json at runtime (#909)
- Tweak the label for citations for better rendering (#919)
- Tighten up some logic around session timestamps and ids (#922)
- Change EventMsg enum so every variant takes a single struct (#925)
- Reasoning default to medium, show workdir when supplied (#931)
- Test_dev_null_write() was not using echo as intended (#923)
## `0.1.2504301751`
### 🚀 Features

View File

@@ -98,12 +98,14 @@ export OPENAI_API_KEY="your-api-key-here"
>
> - openai (default)
> - openrouter
> - azure
> - gemini
> - ollama
> - mistral
> - deepseek
> - xai
> - groq
> - arceeai
> - any other provider that is compatible with the OpenAI API
>
> If you use a provider other than OpenAI, you will need to set the API key for the provider in the config file or in the environment variable as:
@@ -226,13 +228,13 @@ Key flags: `--model/-m`, `--approval-mode/-a`, `--quiet/-q`, and `--notify`.
## Memory & project docs
Codex merges Markdown instructions in this order:
You can give Codex extra instructions and guidance using `AGENTS.md` files. Codex looks for `AGENTS.md` files in the following places, and merges them top-down:
1. `~/.codex/instructions.md` - personal global guidance
2. `codex.md` at repo root - shared project notes
3. `codex.md` in cwd - sub-package specifics
1. `~/.codex/AGENTS.md` - personal global guidance
2. `AGENTS.md` at repo root - shared project notes
3. `AGENTS.md` in the current working directory - sub-folder/feature specifics
Disable with `--no-project-doc` or `CODEX_DISABLE_PROJECT_DOC=1`.
Disable loading of these files with `--no-project-doc` or the environment variable `CODEX_DISABLE_PROJECT_DOC=1`.
---
@@ -394,6 +396,11 @@ Below is a comprehensive example of `config.json` with multiple custom providers
"baseURL": "https://api.openai.com/v1",
"envKey": "OPENAI_API_KEY"
},
"azure": {
"name": "AzureOpenAI",
"baseURL": "https://YOUR_PROJECT_NAME.openai.azure.com/openai",
"envKey": "AZURE_OPENAI_API_KEY"
},
"openrouter": {
"name": "OpenRouter",
"baseURL": "https://openrouter.ai/api/v1",
@@ -428,6 +435,11 @@ Below is a comprehensive example of `config.json` with multiple custom providers
"name": "Groq",
"baseURL": "https://api.groq.com/openai/v1",
"envKey": "GROQ_API_KEY"
},
"arceeai": {
"name": "ArceeAI",
"baseURL": "https://conductor.arcee.ai/v1",
"envKey": "ARCEEAI_API_KEY"
}
},
"history": {
@@ -440,7 +452,7 @@ Below is a comprehensive example of `config.json` with multiple custom providers
### Custom instructions
You can create a `~/.codex/instructions.md` file to define custom instructions:
You can create a `~/.codex/AGENTS.md` file to define custom guidance for the agent:
```markdown
- Always respond with emojis
@@ -455,6 +467,10 @@ For each AI provider, you need to set the corresponding API key in your environm
# OpenAI
export OPENAI_API_KEY="your-api-key-here"
# Azure OpenAI
export AZURE_OPENAI_API_KEY="your-azure-api-key-here"
export AZURE_OPENAI_API_VERSION="2025-03-01-preview" (Optional)
# OpenRouter
export OPENROUTER_API_KEY="your-openrouter-key-here"
@@ -636,17 +652,21 @@ The **DCO check** blocks merges until every commit in the PR carries the footer
### Releasing `codex`
To publish a new version of the CLI, run the following in the `codex-cli` folder to stage the release in a temporary directory:
To publish a new version of the CLI you first need to stage the npm package. A
helper script in `codex-cli/scripts/` does all the heavy lifting. Inside the
`codex-cli` folder run:
```
```bash
# Classic, JS implementation that includes small, native binaries for Linux sandboxing.
pnpm stage-release
```
Note you can specify the folder for the staged release:
```
# Optionally specify the temp directory to reuse between runs.
RELEASE_DIR=$(mktemp -d)
pnpm stage-release "$RELEASE_DIR"
pnpm stage-release --tmp "$RELEASE_DIR"
# "Fat" package that additionally bundles the native Rust CLI binaries for
# Linux. End-users can then opt-in at runtime by setting CODEX_RUST=1.
pnpm stage-release --native
```
Go to the folder where the release is staged and verify that it works as intended. If so, run the following from the temp folder:
@@ -665,7 +685,9 @@ Prerequisite: Nix >= 2.4 with flakes enabled (`experimental-features = nix-comma
Enter a Nix development shell:
```bash
nix develop
# Use either one of the commands according to which implementation you want to work with
nix develop .#codex-cli # For entering codex-cli specific shell
nix develop .#codex-rs # For entering codex-rs specific shell
```
This shell includes Node.js, installs dependencies, builds the CLI, and provides a `codex` command alias.
@@ -673,14 +695,29 @@ This shell includes Node.js, installs dependencies, builds the CLI, and provides
Build and run the CLI directly:
```bash
nix build
# Use either one of the commands according to which implementation you want to work with
nix build .#codex-cli # For building codex-cli
nix build .#codex-rs # For building codex-rs
./result/bin/codex --help
```
Run the CLI via the flake app:
```bash
nix run .#codex
# Use either one of the commands according to which implementation you want to work with
nix run .#codex-cli # For running codex-cli
nix run .#codex-rs # For running codex-rs
```
Use direnv with flakes
If you have direnv installed, you can use the following `.envrc` to automatically enter the Nix shell when you `cd` into the project directory:
```bash
cd codex-rs
echo "use flake ../flake.nix#codex-cli" >> .envrc && direnv allow
cd codex-cli
echo "use flake ../flake.nix#codex-rs" >> .envrc && direnv allow
```
---

View File

@@ -1,6 +1,6 @@
module.exports = {
root: true,
env: { browser: true, es2020: true },
env: { browser: true, node: true, es2020: true },
extends: [
"eslint:recommended",
"plugin:@typescript-eslint/recommended",

View File

@@ -1,17 +1,89 @@
#!/usr/bin/env node
// Unified entry point for the Codex CLI.
/*
* Behavior
* =========
* 1. By default we import the JavaScript implementation located in
* dist/cli.js.
*
* 2. Developers can opt-in to a pre-compiled Rust binary by setting the
* environment variable CODEX_RUST to a truthy value (`1`, `true`, etc.).
* When that variable is present we resolve the correct binary for the
* current platform / architecture and execute it via child_process.
*
* If the CODEX_RUST=1 is specified and there is no native binary for the
* current platform / architecture, an error is thrown.
*/
// Unified entry point for Codex CLI on all platforms
// Dynamically loads the compiled ESM bundle in dist/cli.js
import { spawnSync } from "child_process";
import path from "path";
import { fileURLToPath, pathToFileURL } from "url";
import path from 'path';
import { fileURLToPath, pathToFileURL } from 'url';
// Determine whether the user explicitly wants the Rust CLI.
const wantsNative =
process.env.CODEX_RUST != null
? ["1", "true", "yes"].includes(process.env.CODEX_RUST.toLowerCase())
: false;
// Try native binary if requested.
if (wantsNative) {
const { platform, arch } = process;
let targetTriple = null;
switch (platform) {
case "linux":
switch (arch) {
case "x64":
targetTriple = "x86_64-unknown-linux-musl";
break;
case "arm64":
targetTriple = "aarch64-unknown-linux-gnu";
break;
default:
break;
}
break;
case "darwin":
switch (arch) {
case "x64":
targetTriple = "x86_64-apple-darwin";
break;
case "arm64":
targetTriple = "aarch64-apple-darwin";
break;
default:
break;
}
break;
default:
break;
}
if (!targetTriple) {
throw new Error(`Unsupported platform: ${platform} (${arch})`);
}
// __dirname equivalent in ESM
const __filename = fileURLToPath(import.meta.url);
const __dirname = path.dirname(__filename);
const binaryPath = path.join(__dirname, "..", "bin", `codex-${targetTriple}`);
const result = spawnSync(binaryPath, process.argv.slice(2), {
stdio: "inherit",
});
const exitCode = typeof result.status === "number" ? result.status : 1;
process.exit(exitCode);
}
// Fallback: execute the original JavaScript CLI.
// Determine this script's directory
const __filename = fileURLToPath(import.meta.url);
const __dirname = path.dirname(__filename);
// Resolve the path to the compiled CLI bundle
const cliPath = path.resolve(__dirname, '../dist/cli.js');
const cliPath = path.resolve(__dirname, "../dist/cli.js");
const cliUrl = pathToFileURL(cliPath).href;
// Load and execute the CLI
@@ -21,7 +93,6 @@ const cliUrl = pathToFileURL(cliPath).href;
} catch (err) {
// eslint-disable-next-line no-console
console.error(err);
// eslint-disable-next-line no-undef
process.exit(1);
}
})();

View File

@@ -72,6 +72,9 @@ if (isDevBuild) {
esbuild
.build({
entryPoints: ["src/cli.tsx"],
// Do not bundle the contents of package.json at build time: always read it
// at runtime.
external: ["../package.json"],
bundle: true,
format: "esm",
platform: "node",

43
codex-cli/default.nix Normal file
View File

@@ -0,0 +1,43 @@
{ pkgs, monorep-deps ? [], ... }:
let
node = pkgs.nodejs_22;
in
rec {
package = pkgs.buildNpmPackage {
pname = "codex-cli";
version = "0.1.0";
src = ./.;
npmDepsHash = "sha256-3tAalmh50I0fhhd7XreM+jvl0n4zcRhqygFNB1Olst8";
nodejs = node;
npmInstallFlags = [ "--frozen-lockfile" ];
meta = with pkgs.lib; {
description = "OpenAI Codex commandline interface";
license = licenses.asl20;
homepage = "https://github.com/openai/codex";
};
};
devShell = pkgs.mkShell {
name = "codex-cli-dev";
buildInputs = monorep-deps ++ [
node
pkgs.pnpm
];
shellHook = ''
echo "Entering development shell for codex-cli"
# cd codex-cli
if [ -f package-lock.json ]; then
pnpm ci || echo "npm ci failed"
else
pnpm install || echo "npm install failed"
fi
npm run build || echo "npm build failed"
export PATH=$PWD/node_modules/.bin:$PATH
alias codex="node $PWD/dist/cli.js"
'';
};
app = {
type = "app";
program = "${package}/bin/codex";
};
}

View File

@@ -1,7 +1,7 @@
name: "impossible-pong"
description: |
Update index.html with the following features:
- Add an overlayed styled popup to start the game on first load
- Add an overlaid styled popup to start the game on first load
- Between each point, show a 3 second countdown (this should be skipped if a player wins)
- After each game the AI wins, display text at the bottom of the screen with lighthearted insults for the player
- Add a leaderboard to the right of the court that shows how many games each player has won.

View File

@@ -13,7 +13,7 @@ act,prompt,for_devs
"Advertiser","I want you to act as an advertiser. You will create a campaign to promote a product or service of your choice. You will choose a target audience, develop key messages and slogans, select the media channels for promotion, and decide on any additional activities needed to reach your goals. My first suggestion request is ""I need help creating an advertising campaign for a new type of energy drink targeting young adults aged 18-30.""",FALSE
"Storyteller","I want you to act as a storyteller. You will come up with entertaining stories that are engaging, imaginative and captivating for the audience. It can be fairy tales, educational stories or any other type of stories which has the potential to capture people's attention and imagination. Depending on the target audience, you may choose specific themes or topics for your storytelling session e.g., if it's children then you can talk about animals; If it's adults then history-based tales might engage them better etc. My first request is ""I need an interesting story on perseverance.""",FALSE
"Football Commentator","I want you to act as a football commentator. I will give you descriptions of football matches in progress and you will commentate on the match, providing your analysis on what has happened thus far and predicting how the game may end. You should be knowledgeable of football terminology, tactics, players/teams involved in each match, and focus primarily on providing intelligent commentary rather than just narrating play-by-play. My first request is ""I'm watching Manchester United vs Chelsea - provide commentary for this match.""",FALSE
"Stand-up Comedian","I want you to act as a stand-up comedian. I will provide you with some topics related to current events and you will use your wit, creativity, and observational skills to create a routine based on those topics. You should also be sure to incorporate personal anecdotes or experiences into the routine in order to make it more relatable and engaging for the audience. My first request is ""I want an humorous take on politics.""",FALSE
"Stand-up Comedian","I want you to act as a stand-up comedian. I will provide you with some topics related to current events and you will use your with, creativity, and observational skills to create a routine based on those topics. You should also be sure to incorporate personal anecdotes or experiences into the routine in order to make it more relatable and engaging for the audience. My first request is ""I want an humorous take on politics.""",FALSE
"Motivational Coach","I want you to act as a motivational coach. I will provide you with some information about someone's goals and challenges, and it will be your job to come up with strategies that can help this person achieve their goals. This could involve providing positive affirmations, giving helpful advice or suggesting activities they can do to reach their end goal. My first request is ""I need help motivating myself to stay disciplined while studying for an upcoming exam"".",FALSE
"Composer","I want you to act as a composer. I will provide the lyrics to a song and you will create music for it. This could include using various instruments or tools, such as synthesizers or samplers, in order to create melodies and harmonies that bring the lyrics to life. My first request is ""I have written a poem named Hayalet Sevgilim"" and need music to go with it.""""""",FALSE
"Debater","I want you to act as a debater. I will provide you with some topics related to current events and your task is to research both sides of the debates, present valid arguments for each side, refute opposing points of view, and draw persuasive conclusions based on evidence. Your goal is to help people come away from the discussion with increased knowledge and insight into the topic at hand. My first request is ""I want an opinion piece about Deno.""",FALSE
@@ -23,7 +23,7 @@ act,prompt,for_devs
"Movie Critic","I want you to act as a movie critic. You will develop an engaging and creative movie review. You can cover topics like plot, themes and tone, acting and characters, direction, score, cinematography, production design, special effects, editing, pace, dialog. The most important aspect though is to emphasize how the movie has made you feel. What has really resonated with you. You can also be critical about the movie. Please avoid spoilers. My first request is ""I need to write a movie review for the movie Interstellar""",FALSE
"Relationship Coach","I want you to act as a relationship coach. I will provide some details about the two people involved in a conflict, and it will be your job to come up with suggestions on how they can work through the issues that are separating them. This could include advice on communication techniques or different strategies for improving their understanding of one another's perspectives. My first request is ""I need help solving conflicts between my spouse and myself.""",FALSE
"Poet","I want you to act as a poet. You will create poems that evoke emotions and have the power to stir people's soul. Write on any topic or theme but make sure your words convey the feeling you are trying to express in beautiful yet meaningful ways. You can also come up with short verses that are still powerful enough to leave an imprint in readers' minds. My first request is ""I need a poem about love.""",FALSE
"Rapper","I want you to act as a rapper. You will come up with powerful and meaningful lyrics, beats and rhythm that can 'wow' the audience. Your lyrics should have an intriguing meaning and message which people can relate too. When it comes to choosing your beat, make sure it is catchy yet relevant to your words, so that when combined they make an explosion of sound everytime! My first request is ""I need a rap song about finding strength within yourself.""",FALSE
"Rapper","I want you to act as a rapper. You will come up with powerful and meaningful lyrics, beats and rhythm that can 'wow' the audience. Your lyrics should have an intriguing meaning and message which people can relate too. When it comes to choosing your beat, make sure it is catchy yet relevant to your words, so that when combined they make an explosion of sound every time! My first request is ""I need a rap song about finding strength within yourself.""",FALSE
"Motivational Speaker","I want you to act as a motivational speaker. Put together words that inspire action and make people feel empowered to do something beyond their abilities. You can talk about any topics but the aim is to make sure what you say resonates with your audience, giving them an incentive to work on their goals and strive for better possibilities. My first request is ""I need a speech about how everyone should never give up.""",FALSE
"Philosophy Teacher","I want you to act as a philosophy teacher. I will provide some topics related to the study of philosophy, and it will be your job to explain these concepts in an easy-to-understand manner. This could include providing examples, posing questions or breaking down complex ideas into smaller pieces that are easier to comprehend. My first request is ""I need help understanding how different philosophical theories can be applied in everyday life.""",FALSE
"Philosopher","I want you to act as a philosopher. I will provide some topics or questions related to the study of philosophy, and it will be your job to explore these concepts in depth. This could involve conducting research into various philosophical theories, proposing new ideas or finding creative solutions for solving complex problems. My first request is ""I need help developing an ethical framework for decision making.""",FALSE
1 act prompt for_devs
13 Advertiser I want you to act as an advertiser. You will create a campaign to promote a product or service of your choice. You will choose a target audience, develop key messages and slogans, select the media channels for promotion, and decide on any additional activities needed to reach your goals. My first suggestion request is "I need help creating an advertising campaign for a new type of energy drink targeting young adults aged 18-30." FALSE
14 Storyteller I want you to act as a storyteller. You will come up with entertaining stories that are engaging, imaginative and captivating for the audience. It can be fairy tales, educational stories or any other type of stories which has the potential to capture people's attention and imagination. Depending on the target audience, you may choose specific themes or topics for your storytelling session e.g., if it's children then you can talk about animals; If it's adults then history-based tales might engage them better etc. My first request is "I need an interesting story on perseverance." FALSE
15 Football Commentator I want you to act as a football commentator. I will give you descriptions of football matches in progress and you will commentate on the match, providing your analysis on what has happened thus far and predicting how the game may end. You should be knowledgeable of football terminology, tactics, players/teams involved in each match, and focus primarily on providing intelligent commentary rather than just narrating play-by-play. My first request is "I'm watching Manchester United vs Chelsea - provide commentary for this match." FALSE
16 Stand-up Comedian I want you to act as a stand-up comedian. I will provide you with some topics related to current events and you will use your wit, creativity, and observational skills to create a routine based on those topics. You should also be sure to incorporate personal anecdotes or experiences into the routine in order to make it more relatable and engaging for the audience. My first request is "I want an humorous take on politics." I want you to act as a stand-up comedian. I will provide you with some topics related to current events and you will use your with, creativity, and observational skills to create a routine based on those topics. You should also be sure to incorporate personal anecdotes or experiences into the routine in order to make it more relatable and engaging for the audience. My first request is "I want an humorous take on politics." FALSE
17 Motivational Coach I want you to act as a motivational coach. I will provide you with some information about someone's goals and challenges, and it will be your job to come up with strategies that can help this person achieve their goals. This could involve providing positive affirmations, giving helpful advice or suggesting activities they can do to reach their end goal. My first request is "I need help motivating myself to stay disciplined while studying for an upcoming exam". FALSE
18 Composer I want you to act as a composer. I will provide the lyrics to a song and you will create music for it. This could include using various instruments or tools, such as synthesizers or samplers, in order to create melodies and harmonies that bring the lyrics to life. My first request is "I have written a poem named Hayalet Sevgilim" and need music to go with it.""" FALSE
19 Debater I want you to act as a debater. I will provide you with some topics related to current events and your task is to research both sides of the debates, present valid arguments for each side, refute opposing points of view, and draw persuasive conclusions based on evidence. Your goal is to help people come away from the discussion with increased knowledge and insight into the topic at hand. My first request is "I want an opinion piece about Deno." FALSE
23 Movie Critic I want you to act as a movie critic. You will develop an engaging and creative movie review. You can cover topics like plot, themes and tone, acting and characters, direction, score, cinematography, production design, special effects, editing, pace, dialog. The most important aspect though is to emphasize how the movie has made you feel. What has really resonated with you. You can also be critical about the movie. Please avoid spoilers. My first request is "I need to write a movie review for the movie Interstellar" FALSE
24 Relationship Coach I want you to act as a relationship coach. I will provide some details about the two people involved in a conflict, and it will be your job to come up with suggestions on how they can work through the issues that are separating them. This could include advice on communication techniques or different strategies for improving their understanding of one another's perspectives. My first request is "I need help solving conflicts between my spouse and myself." FALSE
25 Poet I want you to act as a poet. You will create poems that evoke emotions and have the power to stir people's soul. Write on any topic or theme but make sure your words convey the feeling you are trying to express in beautiful yet meaningful ways. You can also come up with short verses that are still powerful enough to leave an imprint in readers' minds. My first request is "I need a poem about love." FALSE
26 Rapper I want you to act as a rapper. You will come up with powerful and meaningful lyrics, beats and rhythm that can 'wow' the audience. Your lyrics should have an intriguing meaning and message which people can relate too. When it comes to choosing your beat, make sure it is catchy yet relevant to your words, so that when combined they make an explosion of sound everytime! My first request is "I need a rap song about finding strength within yourself." I want you to act as a rapper. You will come up with powerful and meaningful lyrics, beats and rhythm that can 'wow' the audience. Your lyrics should have an intriguing meaning and message which people can relate too. When it comes to choosing your beat, make sure it is catchy yet relevant to your words, so that when combined they make an explosion of sound every time! My first request is "I need a rap song about finding strength within yourself." FALSE
27 Motivational Speaker I want you to act as a motivational speaker. Put together words that inspire action and make people feel empowered to do something beyond their abilities. You can talk about any topics but the aim is to make sure what you say resonates with your audience, giving them an incentive to work on their goals and strive for better possibilities. My first request is "I need a speech about how everyone should never give up." FALSE
28 Philosophy Teacher I want you to act as a philosophy teacher. I will provide some topics related to the study of philosophy, and it will be your job to explain these concepts in an easy-to-understand manner. This could include providing examples, posing questions or breaking down complex ideas into smaller pieces that are easier to comprehend. My first request is "I need help understanding how different philosophical theories can be applied in everyday life." FALSE
29 Philosopher I want you to act as a philosopher. I will provide some topics or questions related to the study of philosophy, and it will be your job to explore these concepts in depth. This could involve conducting research into various philosophical theories, proposing new ideas or finding creative solutions for solving complex problems. My first request is "I need help developing an ethical framework for decision making." FALSE

View File

@@ -1,6 +1,6 @@
{
"name": "@openai/codex",
"version": "0.1.2504301751",
"version": "0.0.0-dev",
"license": "Apache-2.0",
"bin": {
"codex": "bin/codex.js"
@@ -23,6 +23,7 @@
"stage-release": "./scripts/stage_release.sh"
},
"files": [
"bin",
"dist"
],
"dependencies": {
@@ -30,6 +31,7 @@
"chalk": "^5.2.0",
"diff": "^7.0.0",
"dotenv": "^16.1.4",
"express": "^5.1.0",
"fast-deep-equal": "^3.1.3",
"fast-npm-meta": "^0.4.2",
"figures": "^6.1.0",
@@ -53,6 +55,7 @@
"devDependencies": {
"@eslint/js": "^9.22.0",
"@types/diff": "^7.0.2",
"@types/express": "^5.0.1",
"@types/js-yaml": "^4.0.9",
"@types/marked-terminal": "^6.1.1",
"@types/react": "^18.0.32",

View File

@@ -1,20 +1,44 @@
#!/bin/bash
#!/usr/bin/env bash
# Copy the Linux sandbox native binaries into the bin/ subfolder of codex-cli/.
# Install native runtime dependencies for codex-cli.
#
# Usage:
# ./scripts/install_native_deps.sh [CODEX_CLI_ROOT]
# By default the script copies the sandbox binaries that are required at
# runtime. When called with the --full-native flag, it additionally
# bundles pre-built Rust CLI binaries so that the resulting npm package can run
# the native implementation when users set CODEX_RUST=1.
#
# Arguments
# [CODEX_CLI_ROOT] Optional. If supplied, it should be the codex-cli
# folder that contains the package.json for @openai/codex.
# Usage
# install_native_deps.sh [RELEASE_ROOT] [--full-native]
#
# When no argument is given we assume the script is being run directly from a
# development checkout. In that case we install the binaries into the
# repositorys own `bin/` directory so that the CLI can run locally.
# The optional RELEASE_ROOT is the path that contains package.json. Omitting
# it installs the binaries into the repository's own bin/ folder to support
# local development.
set -euo pipefail
# ------------------
# Parse arguments
# ------------------
DEST_DIR=""
INCLUDE_RUST=0
for arg in "$@"; do
case "$arg" in
--full-native)
INCLUDE_RUST=1
;;
*)
if [[ -z "$DEST_DIR" ]]; then
DEST_DIR="$arg"
else
echo "Unexpected argument: $arg" >&2
exit 1
fi
;;
esac
done
# ----------------------------------------------------------------------------
# Determine where the binaries should be installed.
# ----------------------------------------------------------------------------
@@ -41,7 +65,7 @@ mkdir -p "$BIN_DIR"
# Until we start publishing stable GitHub releases, we have to grab the binaries
# from the GitHub Action that created them. Update the URL below to point to the
# appropriate workflow run:
WORKFLOW_URL="https://github.com/openai/codex/actions/runs/14763725716"
WORKFLOW_URL="https://github.com/openai/codex/actions/runs/15087655786"
WORKFLOW_ID="${WORKFLOW_URL##*/}"
ARTIFACTS_DIR="$(mktemp -d)"
@@ -50,12 +74,26 @@ trap 'rm -rf "$ARTIFACTS_DIR"' EXIT
# NB: The GitHub CLI `gh` must be installed and authenticated.
gh run download --dir "$ARTIFACTS_DIR" --repo openai/codex "$WORKFLOW_ID"
# Decompress the two target architectures.
# Decompress the artifacts for Linux sandboxing.
zstd -d "$ARTIFACTS_DIR/x86_64-unknown-linux-musl/codex-linux-sandbox-x86_64-unknown-linux-musl.zst" \
-o "$BIN_DIR/codex-linux-sandbox-x64"
zstd -d "$ARTIFACTS_DIR/aarch64-unknown-linux-gnu/codex-linux-sandbox-aarch64-unknown-linux-gnu.zst" \
-o "$BIN_DIR/codex-linux-sandbox-arm64"
echo "Installed native dependencies into $BIN_DIR"
if [[ "$INCLUDE_RUST" -eq 1 ]]; then
# x64 Linux
zstd -d "$ARTIFACTS_DIR/x86_64-unknown-linux-musl/codex-x86_64-unknown-linux-musl.zst" \
-o "$BIN_DIR/codex-x86_64-unknown-linux-musl"
# ARM64 Linux
zstd -d "$ARTIFACTS_DIR/aarch64-unknown-linux-gnu/codex-aarch64-unknown-linux-gnu.zst" \
-o "$BIN_DIR/codex-aarch64-unknown-linux-gnu"
# x64 macOS
zstd -d "$ARTIFACTS_DIR/x86_64-apple-darwin/codex-x86_64-apple-darwin.zst" \
-o "$BIN_DIR/codex-x86_64-apple-darwin"
# ARM64 macOS
zstd -d "$ARTIFACTS_DIR/aarch64-apple-darwin/codex-aarch64-apple-darwin.zst" \
-o "$BIN_DIR/codex-aarch64-apple-darwin"
fi
echo "Installed native dependencies into $BIN_DIR"

View File

@@ -1,28 +1,145 @@
#!/bin/bash
#!/usr/bin/env bash
# -----------------------------------------------------------------------------
# stage_release.sh
# -----------------------------------------------------------------------------
# Stages an npm release for @openai/codex.
#
# The script used to accept a single optional positional argument that indicated
# the temporary directory in which to stage the package. We now support a
# flag-based interface so that we can extend the command with further options
# without breaking the call-site contract.
#
# --tmp <dir> : Use <dir> instead of a freshly created temp directory.
# --native : Bundle the pre-built Rust CLI binaries for Linux alongside
# the JavaScript implementation (a so-called "fat" package).
# -h|--help : Print usage.
#
# When --native is supplied we copy the linux-sandbox binaries (as before) and
# additionally fetch / unpack the two Rust targets that we currently support:
# - x86_64-unknown-linux-musl
# - aarch64-unknown-linux-gnu
#
# NOTE: This script is intended to be run from the repository root via
# `pnpm --filter codex-cli stage-release ...` or inside codex-cli with the
# helper script entry in package.json (`pnpm stage-release ...`).
# -----------------------------------------------------------------------------
set -euo pipefail
# Change to the codex-cli directory.
cd "$(dirname "${BASH_SOURCE[0]}")/.."
# Helper - usage / flag parsing
# First argument is where to stage the release. Creates a temporary directory
# if not provided.
RELEASE_DIR="${1:-$(mktemp -d)}"
[ -n "${1-}" ] && shift
usage() {
cat <<EOF
Usage: $(basename "$0") [--tmp DIR] [--native]
Options
--tmp DIR Use DIR to stage the release (defaults to a fresh mktemp dir)
--native Bundle Rust binaries for Linux (fat package)
-h, --help Show this help
Legacy positional argument: the first non-flag argument is still interpreted
as the temporary directory (for backwards compatibility) but is deprecated.
EOF
exit "${1:-0}"
}
TMPDIR=""
INCLUDE_NATIVE=0
# Manual flag parser - Bash getopts does not handle GNU long options well.
while [[ $# -gt 0 ]]; do
case "$1" in
--tmp)
shift || { echo "--tmp requires an argument"; usage 1; }
TMPDIR="$1"
;;
--tmp=*)
TMPDIR="${1#*=}"
;;
--native)
INCLUDE_NATIVE=1
;;
-h|--help)
usage 0
;;
--*)
echo "Unknown option: $1" >&2
usage 1
;;
*)
echo "Unexpected extra argument: $1" >&2
usage 1
;;
esac
shift
done
# Fallback when the caller did not specify a directory.
# If no directory was specified create a fresh temporary one.
if [[ -z "$TMPDIR" ]]; then
TMPDIR="$(mktemp -d)"
fi
# Ensure the directory exists, then resolve to an absolute path.
mkdir -p "$TMPDIR"
TMPDIR="$(cd "$TMPDIR" && pwd)"
# Main build logic
echo "Staging release in $TMPDIR"
# The script lives in codex-cli/scripts/ - change into codex-cli root so that
# relative paths keep working.
SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
CODEX_CLI_ROOT="$(cd "$SCRIPT_DIR/.." && pwd)"
pushd "$CODEX_CLI_ROOT" >/dev/null
# 1. Build the JS artifacts ---------------------------------------------------
# Compile the JavaScript.
pnpm install
pnpm build
mkdir "$RELEASE_DIR/bin"
cp -r bin/codex.js "$RELEASE_DIR/bin/codex.js"
cp -r dist "$RELEASE_DIR/dist"
cp -r src "$RELEASE_DIR/src" # important if we want sourcemaps to continue to work
cp ../README.md "$RELEASE_DIR"
# TODO: Derive version from Git tag.
VERSION=$(printf '0.1.%d' "$(date +%y%m%d%H%M)")
jq --arg version "$VERSION" '.version = $version' package.json > "$RELEASE_DIR/package.json"
# Copy the native dependencies.
./scripts/install_native_deps.sh "$RELEASE_DIR"
# Paths inside the staged package
mkdir -p "$TMPDIR/bin"
echo "Staged version $VERSION for release in $RELEASE_DIR"
cp -r bin/codex.js "$TMPDIR/bin/codex.js"
cp -r dist "$TMPDIR/dist"
cp -r src "$TMPDIR/src" # keep source for TS sourcemaps
cp ../README.md "$TMPDIR" || true # README is one level up - ignore if missing
# Derive a timestamp-based version (keep same scheme as before)
VERSION="$(printf '0.1.%d' "$(date +%y%m%d%H%M)")"
# Modify package.json - bump version and optionally add the native directory to
# the files array so that the binaries are published to npm.
jq --arg version "$VERSION" \
'.version = $version' \
package.json > "$TMPDIR/package.json"
# 2. Native runtime deps (sandbox plus optional Rust binaries)
if [[ "$INCLUDE_NATIVE" -eq 1 ]]; then
./scripts/install_native_deps.sh "$TMPDIR" --full-native
else
./scripts/install_native_deps.sh "$TMPDIR"
fi
popd >/dev/null
echo "Staged version $VERSION for release in $TMPDIR"
echo "Test Node:"
echo " node ${TMPDIR}/bin/codex.js --help"
if [[ "$INCLUDE_NATIVE" -eq 1 ]]; then
echo "Test Rust:"
echo " CODEX_RUST=1 node ${TMPDIR}/bin/codex.js --help"
fi
# Print final hint for convenience
if [[ "$INCLUDE_NATIVE" -eq 1 ]]; then
echo "Next: cd \"$TMPDIR\" && npm publish --tag native"
else
echo "Next: cd \"$TMPDIR\" && npm publish"
fi

View File

@@ -1,12 +1,13 @@
import type { ApprovalPolicy } from "./approvals";
import type { AppConfig } from "./utils/config";
import type { TerminalChatSession } from "./utils/session.js";
import type { ResponseItem } from "openai/resources/responses/responses";
import TerminalChat from "./components/chat/terminal-chat";
import TerminalChatPastRollout from "./components/chat/terminal-chat-past-rollout";
import { checkInGit } from "./utils/check-in-git";
import { CLI_VERSION, type TerminalChatSession } from "./utils/session.js";
import { onExit } from "./utils/terminal";
import { CLI_VERSION } from "./version";
import { ConfirmInput } from "@inkjs/ui";
import { Box, Text, useApp, useStdin } from "ink";
import React, { useMemo, useState } from "react";
@@ -49,6 +50,7 @@ export default function App({
<TerminalChatPastRollout
session={rollout.session}
items={rollout.items}
fileOpener={config.fileOpener}
/>
);
}

View File

@@ -281,12 +281,14 @@ export function resolvePathAgainstWorkdir(
candidatePath: string,
workdir: string | undefined,
): string {
if (path.isAbsolute(candidatePath)) {
return candidatePath;
// Normalize candidatePath to prevent path traversal attacks
const normalizedCandidatePath = path.normalize(candidatePath);
if (path.isAbsolute(normalizedCandidatePath)) {
return normalizedCandidatePath;
} else if (workdir != null) {
return path.resolve(workdir, candidatePath);
return path.resolve(workdir, normalizedCandidatePath);
} else {
return path.resolve(candidatePath);
return path.resolve(normalizedCandidatePath);
}
}
@@ -363,6 +365,11 @@ export function isSafeCommand(
reason: "View file contents",
group: "Reading files",
};
case "nl":
return {
reason: "View file with line numbers",
group: "Reading files",
};
case "rg":
return {
reason: "Ripgrep search",
@@ -446,11 +453,15 @@ export function isSafeCommand(
}
break;
case "sed":
// We allow two types of sed invocations:
// 1. `sed -n 1,200p FILE`
// 2. `sed -n 1,200p` because the file is passed via stdin, e.g.,
// `nl -ba README.md | sed -n '1,200p'`
if (
cmd1 === "-n" &&
isValidSedNArg(cmd2) &&
typeof cmd3 === "string" &&
command.length === 4
(command.length === 3 ||
(typeof cmd3 === "string" && command.length === 4))
) {
return {
reason: "Sed print subset",

View File

@@ -1,6 +1,19 @@
#!/usr/bin/env node
import "dotenv/config";
// Exit early if on an older version of Node.js (< 22)
const major = process.versions.node.split(".").map(Number)[0]!;
if (major < 22) {
// eslint-disable-next-line no-console
console.error(
"\n" +
"Codex CLI requires Node.js version 22 or newer.\n" +
`You are running Node.js v${process.versions.node}.\n` +
"Please upgrade Node.js: https://nodejs.org/en/download/\n",
);
process.exit(1);
}
// Hack to suppress deprecation warnings (punycode)
// eslint-disable-next-line @typescript-eslint/no-explicit-any
(process as any).noDeprecation = true;
@@ -14,16 +27,20 @@ import type { ReasoningEffort } from "openai/resources.mjs";
import App from "./app";
import { runSinglePass } from "./cli-singlepass";
import SessionsOverlay from "./components/sessions-overlay.js";
import { AgentLoop } from "./utils/agent/agent-loop";
import { ReviewDecision } from "./utils/agent/review";
import { AutoApprovalMode } from "./utils/auto-approval-mode";
import { checkForUpdates } from "./utils/check-updates";
import {
getApiKey,
loadConfig,
PRETTY_PRINT,
INSTRUCTIONS_FILEPATH,
} from "./utils/config";
import {
getApiKey as fetchApiKey,
maybeRedeemCredits,
} from "./utils/get-api-key";
import { createInputItem } from "./utils/input-utils";
import { initLogger } from "./utils/logger/log";
import { isModelSupportedForResponses } from "./utils/model-utils.js";
@@ -34,6 +51,7 @@ import { spawnSync } from "child_process";
import fs from "fs";
import { render } from "ink";
import meow from "meow";
import os from "os";
import path from "path";
import React from "react";
@@ -56,10 +74,13 @@ const cli = meow(
--version Print version and exit
-h, --help Show usage and exit
-m, --model <model> Model to use for completions (default: o4-mini)
-m, --model <model> Model to use for completions (default: codex-mini-latest)
-p, --provider <provider> Provider to use for completions (default: openai)
-i, --image <path> Path(s) to image files to include as input
-v, --view <rollout> Inspect a previously saved rollout instead of starting a session
--history Browse previous sessions
--login Start a new sign in flow
--free Retry redeeming free credits
-q, --quiet Non-interactive mode that only prints the assistant's final output
-c, --config Open the instructions file in your editor
-w, --writable-root <path> Writable folder for sandbox in full-auto mode (can be specified multiple times)
@@ -68,7 +89,7 @@ const cli = meow(
--auto-edit Automatically approve file edits; still prompt for commands
--full-auto Automatically approve edits and commands when executed in the sandbox
--no-project-doc Do not automatically include the repository's 'codex.md'
--no-project-doc Do not automatically include the repository's 'AGENTS.md'
--project-doc <file> Include an additional markdown file at <file> as context
--full-stdout Do not truncate stdout/stderr from command outputs
--notify Enable desktop notifications for responses
@@ -79,6 +100,8 @@ const cli = meow(
--flex-mode Use "flex-mode" processing mode for the request (only supported
with models o3 and o4-mini)
--reasoning <effort> Set the reasoning effort level (low, medium, high) (default: high)
Dangerous options
--dangerously-auto-approve-everything
Skip all confirmation prompts and execute commands without
@@ -102,6 +125,9 @@ const cli = meow(
help: { type: "boolean", aliases: ["h"] },
version: { type: "boolean", description: "Print version and exit" },
view: { type: "string" },
history: { type: "boolean", description: "Browse previous sessions" },
login: { type: "boolean", description: "Force a new sign in flow" },
free: { type: "boolean", description: "Retry redeeming free credits" },
model: { type: "string", aliases: ["m"] },
provider: { type: "string", aliases: ["p"] },
image: { type: "string", isMultiple: true, aliases: ["i"] },
@@ -144,7 +170,7 @@ const cli = meow(
},
noProjectDoc: {
type: "boolean",
description: "Disable automatic inclusion of project-level codex.md",
description: "Disable automatic inclusion of project-level AGENTS.md",
},
projectDoc: {
type: "string",
@@ -259,11 +285,82 @@ let config = loadConfig(undefined, undefined, {
isFullContext: fullContextMode,
});
const prompt = cli.input[0];
// `prompt` can be updated later when the user resumes a previous session
// via the `--history` flag. Therefore it must be declared with `let` rather
// than `const`.
let prompt = cli.input[0];
const model = cli.flags.model ?? config.model;
const imagePaths = cli.flags.image;
const provider = cli.flags.provider ?? config.provider ?? "openai";
const apiKey = getApiKey(provider);
const client = {
issuer: "https://auth.openai.com",
client_id: "app_EMoamEEZ73f0CkXaXp7hrann",
};
let apiKey = "";
let savedTokens:
| {
id_token?: string;
access_token?: string;
refresh_token: string;
}
| undefined;
// Try to load existing auth file if present
try {
const home = os.homedir();
const authDir = path.join(home, ".codex");
const authFile = path.join(authDir, "auth.json");
if (fs.existsSync(authFile)) {
const data = JSON.parse(fs.readFileSync(authFile, "utf-8"));
savedTokens = data.tokens;
const lastRefreshTime = data.last_refresh
? new Date(data.last_refresh).getTime()
: 0;
const expired = Date.now() - lastRefreshTime > 28 * 24 * 60 * 60 * 1000;
if (data.OPENAI_API_KEY && !expired) {
apiKey = data.OPENAI_API_KEY;
}
}
} catch {
// ignore errors
}
if (cli.flags.login) {
apiKey = await fetchApiKey(client.issuer, client.client_id);
try {
const home = os.homedir();
const authDir = path.join(home, ".codex");
const authFile = path.join(authDir, "auth.json");
if (fs.existsSync(authFile)) {
const data = JSON.parse(fs.readFileSync(authFile, "utf-8"));
savedTokens = data.tokens;
}
} catch {
/* ignore */
}
} else if (!apiKey) {
apiKey = await fetchApiKey(client.issuer, client.client_id);
}
// Ensure the API key is available as an environment variable for legacy code
process.env["OPENAI_API_KEY"] = apiKey;
if (cli.flags.free) {
// eslint-disable-next-line no-console
console.log(`${chalk.bold("codex --free")} attempting to redeem credits...`);
if (!savedTokens?.refresh_token) {
apiKey = await fetchApiKey(client.issuer, client.client_id, true);
// fetchApiKey includes credit redemption as the end of the flow
} else {
await maybeRedeemCredits(
client.issuer,
client.client_id,
savedTokens.refresh_token,
savedTokens.id_token,
);
}
}
// Set of providers that don't require API keys
const NO_API_KEY_REQUIRED = new Set(["ollama"]);
@@ -306,8 +403,8 @@ config = {
model: model ?? config.model,
notify: Boolean(cli.flags.notify),
reasoningEffort:
(cli.flags.reasoning as ReasoningEffort | undefined) ?? "high",
flexMode: Boolean(cli.flags.flexMode),
(cli.flags.reasoning as ReasoningEffort | undefined) ?? "medium",
flexMode: cli.flags.flexMode || (config.flexMode ?? false),
provider,
disableResponseStorage,
};
@@ -321,15 +418,19 @@ try {
}
// For --flex-mode, validate and exit if incorrect.
if (cli.flags.flexMode) {
if (config.flexMode) {
const allowedFlexModels = new Set(["o3", "o4-mini"]);
if (!allowedFlexModels.has(config.model)) {
// eslint-disable-next-line no-console
console.error(
`The --flex-mode option is only supported when using the 'o3' or 'o4-mini' models. ` +
`Current model: '${config.model}'.`,
);
process.exit(1);
if (cli.flags.flexMode) {
// eslint-disable-next-line no-console
console.error(
`The --flex-mode option is only supported when using the 'o3' or 'o4-mini' models. ` +
`Current model: '${config.model}'.`,
);
process.exit(1);
} else {
config.flexMode = false;
}
}
}
@@ -349,6 +450,46 @@ if (
let rollout: AppRollout | undefined;
// For --history, show session selector and optionally update prompt or rollout.
if (cli.flags.history) {
const result: { path: string; mode: "view" | "resume" } | null =
await new Promise((resolve) => {
const instance = render(
React.createElement(SessionsOverlay, {
onView: (p: string) => {
instance.unmount();
resolve({ path: p, mode: "view" });
},
onResume: (p: string) => {
instance.unmount();
resolve({ path: p, mode: "resume" });
},
onExit: () => {
instance.unmount();
resolve(null);
},
}),
);
});
if (!result) {
process.exit(0);
}
if (result.mode === "view") {
try {
const content = fs.readFileSync(result.path, "utf-8");
rollout = JSON.parse(content) as AppRollout;
} catch (error) {
// eslint-disable-next-line no-console
console.error("Error reading session file:", error);
process.exit(1);
}
} else {
prompt = `Resume this session: ${result.path}`;
}
}
// For --view, optionally load an existing rollout from disk, display it and exit.
if (cli.flags.view) {
const viewPath = cli.flags.view;

View File

@@ -1,6 +1,7 @@
import type { TerminalHeaderProps } from "./terminal-header.js";
import type { GroupedResponseItem } from "./use-message-grouping.js";
import type { ResponseItem } from "openai/resources/responses/responses.mjs";
import type { FileOpenerScheme } from "src/utils/config.js";
import TerminalChatResponseItem from "./terminal-chat-response-item.js";
import TerminalHeader from "./terminal-header.js";
@@ -19,11 +20,13 @@ type MessageHistoryProps = {
confirmationPrompt: React.ReactNode;
loading: boolean;
headerProps: TerminalHeaderProps;
fileOpener: FileOpenerScheme | undefined;
};
const MessageHistory: React.FC<MessageHistoryProps> = ({
batch,
headerProps,
fileOpener,
}) => {
const messages = batch.map(({ item }) => item!);
@@ -68,7 +71,10 @@ const MessageHistory: React.FC<MessageHistoryProps> = ({
message.type === "message" && message.role === "user" ? 0 : 1
}
>
<TerminalChatResponseItem item={message} />
<TerminalChatResponseItem
item={message}
fileOpener={fileOpener}
/>
</Box>
);
}}

View File

@@ -54,6 +54,7 @@ export default function TerminalChatInput({
openApprovalOverlay,
openHelpOverlay,
openDiffOverlay,
openSessionsOverlay,
onCompact,
interruptAgent,
active,
@@ -77,6 +78,7 @@ export default function TerminalChatInput({
openApprovalOverlay: () => void;
openHelpOverlay: () => void;
openDiffOverlay: () => void;
openSessionsOverlay: () => void;
onCompact: () => void;
interruptAgent: () => void;
active: boolean;
@@ -280,6 +282,9 @@ export default function TerminalChatInput({
case "/history":
openOverlay();
break;
case "/sessions":
openSessionsOverlay();
break;
case "/help":
openHelpOverlay();
break;
@@ -484,6 +489,10 @@ export default function TerminalChatInput({
setInput("");
openOverlay();
return;
} else if (inputValue === "/sessions") {
setInput("");
openSessionsOverlay();
return;
} else if (inputValue === "/help") {
setInput("");
openHelpOverlay();
@@ -584,7 +593,7 @@ export default function TerminalChatInput({
try {
const os = await import("node:os");
const { CLI_VERSION } = await import("../../utils/session.js");
const { CLI_VERSION } = await import("../../version.js");
const { buildBugReportUrl } = await import(
"../../utils/bug-report.js"
);
@@ -728,6 +737,7 @@ export default function TerminalChatInput({
openModelOverlay,
openHelpOverlay,
openDiffOverlay,
openSessionsOverlay,
history,
onCompact,
skipNextSubmit,

View File

@@ -1,5 +1,6 @@
import type { TerminalChatSession } from "../../utils/session.js";
import type { ResponseItem } from "openai/resources/responses/responses";
import type { FileOpenerScheme } from "src/utils/config.js";
import TerminalChatResponseItem from "./terminal-chat-response-item";
import { Box, Text } from "ink";
@@ -8,9 +9,11 @@ import React from "react";
export default function TerminalChatPastRollout({
session,
items,
fileOpener,
}: {
session: TerminalChatSession;
items: Array<ResponseItem>;
fileOpener: FileOpenerScheme | undefined;
}): React.ReactElement {
const { version, id: sessionId, model } = session;
return (
@@ -51,9 +54,13 @@ export default function TerminalChatPastRollout({
{React.useMemo(
() =>
items.map((item, key) => (
<TerminalChatResponseItem key={key} item={item} />
<TerminalChatResponseItem
key={key}
item={item}
fileOpener={fileOpener}
/>
)),
[items],
[items, fileOpener],
)}
</Box>
</Box>

View File

@@ -8,6 +8,7 @@ import type {
ResponseOutputMessage,
ResponseReasoningItem,
} from "openai/resources/responses/responses";
import type { FileOpenerScheme } from "src/utils/config";
import { useTerminalSize } from "../../hooks/use-terminal-size";
import { collapseXmlBlocks } from "../../utils/file-tag-utils";
@@ -16,16 +17,21 @@ import chalk, { type ForegroundColorName } from "chalk";
import { Box, Text } from "ink";
import { parse, setOptions } from "marked";
import TerminalRenderer from "marked-terminal";
import path from "path";
import React, { useEffect, useMemo } from "react";
import { formatCommandForDisplay } from "src/format-command.js";
import supportsHyperlinks from "supports-hyperlinks";
export default function TerminalChatResponseItem({
item,
fullStdout = false,
setOverlayMode,
fileOpener,
}: {
item: ResponseItem;
fullStdout?: boolean;
setOverlayMode?: React.Dispatch<React.SetStateAction<OverlayModeType>>;
fileOpener: FileOpenerScheme | undefined;
}): React.ReactElement {
switch (item.type) {
case "message":
@@ -33,10 +39,15 @@ export default function TerminalChatResponseItem({
<TerminalChatResponseMessage
setOverlayMode={setOverlayMode}
message={item}
fileOpener={fileOpener}
/>
);
// @ts-expect-error new item types aren't in SDK yet
case "local_shell_call":
case "function_call":
return <TerminalChatResponseToolCall message={item} />;
// @ts-expect-error new item types aren't in SDK yet
case "local_shell_call_output":
case "function_call_output":
return (
<TerminalChatResponseToolCallOutput
@@ -50,7 +61,9 @@ export default function TerminalChatResponseItem({
// @ts-expect-error `reasoning` is not in the responses API yet
if (item.type === "reasoning") {
return <TerminalChatResponseReasoning message={item} />;
return (
<TerminalChatResponseReasoning message={item} fileOpener={fileOpener} />
);
}
return <TerminalChatResponseGenericMessage message={item} />;
@@ -78,8 +91,10 @@ export default function TerminalChatResponseItem({
export function TerminalChatResponseReasoning({
message,
fileOpener,
}: {
message: ResponseReasoningItem & { duration_ms?: number };
fileOpener: FileOpenerScheme | undefined;
}): React.ReactElement | null {
// Only render when there is a reasoning summary
if (!message.summary || message.summary.length === 0) {
@@ -92,7 +107,7 @@ export function TerminalChatResponseReasoning({
return (
<Box key={key} flexDirection="column">
{s.headline && <Text bold>{s.headline}</Text>}
<Markdown>{s.text}</Markdown>
<Markdown fileOpener={fileOpener}>{s.text}</Markdown>
</Box>
);
})}
@@ -108,9 +123,11 @@ const colorsByRole: Record<string, ForegroundColorName> = {
function TerminalChatResponseMessage({
message,
setOverlayMode,
fileOpener,
}: {
message: ResponseInputMessageItem | ResponseOutputMessage;
setOverlayMode?: React.Dispatch<React.SetStateAction<OverlayModeType>>;
fileOpener: FileOpenerScheme | undefined;
}) {
// auto switch to model mode if the system message contains "has been deprecated"
useEffect(() => {
@@ -129,7 +146,7 @@ function TerminalChatResponseMessage({
<Text bold color={colorsByRole[message.role] || "gray"}>
{message.role === "assistant" ? "codex" : message.role}
</Text>
<Markdown>
<Markdown fileOpener={fileOpener}>
{message.content
.map(
(c) =>
@@ -154,16 +171,28 @@ function TerminalChatResponseMessage({
function TerminalChatResponseToolCall({
message,
}: {
message: ResponseFunctionToolCallItem;
// eslint-disable-next-line @typescript-eslint/no-explicit-any
message: ResponseFunctionToolCallItem | any;
}) {
const details = parseToolCall(message);
let workdir: string | undefined;
let cmdReadableText: string | undefined;
if (message.type === "function_call") {
const details = parseToolCall(message);
workdir = details?.workdir;
cmdReadableText = details?.cmdReadableText;
} else if (message.type === "local_shell_call") {
const action = message.action;
workdir = action.working_directory;
cmdReadableText = formatCommandForDisplay(action.command);
}
return (
<Box flexDirection="column" gap={1}>
<Text color="magentaBright" bold>
command
{workdir ? <Text dimColor>{` (${workdir})`}</Text> : ""}
</Text>
<Text>
<Text dimColor>$</Text> {details?.cmdReadableText}
<Text dimColor>$</Text> {cmdReadableText}
</Text>
</Box>
);
@@ -173,7 +202,8 @@ function TerminalChatResponseToolCallOutput({
message,
fullStdout,
}: {
message: ResponseFunctionToolCallOutputItem;
// eslint-disable-next-line @typescript-eslint/no-explicit-any
message: ResponseFunctionToolCallOutputItem | any;
fullStdout: boolean;
}) {
const { output, metadata } = parseToolCallOutput(message.output);
@@ -240,26 +270,91 @@ export function TerminalChatResponseGenericMessage({
export type MarkdownProps = TerminalRendererOptions & {
children: string;
fileOpener: FileOpenerScheme | undefined;
/** Base path for resolving relative file citation paths. */
cwd?: string;
};
export function Markdown({
children,
fileOpener,
cwd,
...options
}: MarkdownProps): React.ReactElement {
const size = useTerminalSize();
const rendered = React.useMemo(() => {
const linkifiedMarkdown = rewriteFileCitations(children, fileOpener, cwd);
// Configure marked for this specific render
setOptions({
// @ts-expect-error missing parser, space props
renderer: new TerminalRenderer({ ...options, width: size.columns }),
});
const parsed = parse(children, { async: false }).trim();
const parsed = parse(linkifiedMarkdown, { async: false }).trim();
// Remove the truncation logic
return parsed;
// eslint-disable-next-line react-hooks/exhaustive-deps -- options is an object of primitives
}, [children, size.columns, size.rows]);
}, [
children,
size.columns,
size.rows,
fileOpener,
supportsHyperlinks.stdout,
chalk.level,
]);
return <Text>{rendered}</Text>;
}
/** Regex to match citations for source files (hence the `F:` prefix). */
const citationRegex = new RegExp(
[
// Opening marker
"【",
// Capture group 1: file ID or name (anything except '†')
"F:([^†]+)",
// Field separator
"†",
// Capture group 2: start line (digits)
"L(\\d+)",
// Non-capturing group for optional end line
"(?:",
// Capture group 3: end line (digits or '?')
"-L(\\d+|\\?)",
// End of optional group (may not be present)
")?",
// Closing marker
"】",
].join(""),
"g", // Global flag
);
function rewriteFileCitations(
markdown: string,
fileOpener: FileOpenerScheme | undefined,
cwd: string = process.cwd(),
): string {
citationRegex.lastIndex = 0;
return markdown.replace(citationRegex, (_match, file, start, _end) => {
const absPath = path.resolve(cwd, file);
if (!fileOpener) {
return `[${file}](${absPath})`;
}
const uri = `${fileOpener}://file${absPath}:${start}`;
const label = `${file}:${start}`;
// In practice, sometimes multiple citations for the same file, but with a
// different line number, are shown sequentially, so we:
// - include the line number in the label to disambiguate them
// - add a space after the link to make it easier to read
return `[${label}](${uri}) `;
});
}

View File

@@ -1,3 +1,4 @@
import type { AppRollout } from "../../app.js";
import type { ApplyPatchCommand, ApprovalPolicy } from "../../approvals.js";
import type { CommandConfirmation } from "../../utils/agent/agent-loop.js";
import type { AppConfig } from "../../utils/config.js";
@@ -5,6 +6,7 @@ import type { ColorName } from "chalk";
import type { ResponseItem } from "openai/resources/responses/responses.mjs";
import TerminalChatInput from "./terminal-chat-input.js";
import TerminalChatPastRollout from "./terminal-chat-past-rollout.js";
import { TerminalChatToolCallCommand } from "./terminal-chat-tool-call-command.js";
import TerminalMessageHistory from "./terminal-message-history.js";
import { formatCommandForDisplay } from "../../format-command.js";
@@ -13,7 +15,7 @@ import { useTerminalSize } from "../../hooks/use-terminal-size.js";
import { AgentLoop } from "../../utils/agent/agent-loop.js";
import { ReviewDecision } from "../../utils/agent/review.js";
import { generateCompactSummary } from "../../utils/compact-summary.js";
import { getBaseUrl, getApiKey, saveConfig } from "../../utils/config.js";
import { saveConfig } from "../../utils/config.js";
import { extractAppliedPatches as _extractAppliedPatches } from "../../utils/extract-applied-patches.js";
import { getGitDiff } from "../../utils/get-diff.js";
import { createInputItem } from "../../utils/input-utils.js";
@@ -23,24 +25,27 @@ import {
calculateContextPercentRemaining,
uniqueById,
} from "../../utils/model-utils.js";
import { CLI_VERSION } from "../../utils/session.js";
import { createOpenAIClient } from "../../utils/openai-client.js";
import { shortCwd } from "../../utils/short-path.js";
import { saveRollout } from "../../utils/storage/save-rollout.js";
import { CLI_VERSION } from "../../version.js";
import ApprovalModeOverlay from "../approval-mode-overlay.js";
import DiffOverlay from "../diff-overlay.js";
import HelpOverlay from "../help-overlay.js";
import HistoryOverlay from "../history-overlay.js";
import ModelOverlay from "../model-overlay.js";
import SessionsOverlay from "../sessions-overlay.js";
import chalk from "chalk";
import fs from "fs/promises";
import { Box, Text } from "ink";
import { spawn } from "node:child_process";
import OpenAI from "openai";
import React, { useEffect, useMemo, useRef, useState } from "react";
import { inspect } from "util";
export type OverlayModeType =
| "none"
| "history"
| "sessions"
| "model"
| "approval"
| "help"
@@ -78,10 +83,7 @@ async function generateCommandExplanation(
): Promise<string> {
try {
// Create a temporary OpenAI client
const oai = new OpenAI({
apiKey: getApiKey(config.provider),
baseURL: getBaseUrl(config.provider),
});
const oai = createOpenAIClient(config);
// Format the command for display
const commandForDisplay = formatCommandForDisplay(command);
@@ -194,6 +196,7 @@ export default function TerminalChat({
submitConfirmation,
} = useConfirmation();
const [overlayMode, setOverlayMode] = useState<OverlayModeType>("none");
const [viewRollout, setViewRollout] = useState<AppRollout | null>(null);
// Store the diff text when opening the diff overlay so the view isnt
// recomputed on every rerender while it is open.
@@ -457,6 +460,16 @@ export default function TerminalChat({
[items, model],
);
if (viewRollout) {
return (
<TerminalChatPastRollout
fileOpener={config.fileOpener}
session={viewRollout.session}
items={viewRollout.items}
/>
);
}
return (
<Box flexDirection="column">
<Box flexDirection="column">
@@ -483,6 +496,7 @@ export default function TerminalChat({
initialImagePaths,
flexModeEnabled: Boolean(config.flexMode),
}}
fileOpener={config.fileOpener}
/>
) : (
<Box>
@@ -511,6 +525,7 @@ export default function TerminalChat({
openModelOverlay={() => setOverlayMode("model")}
openApprovalOverlay={() => setOverlayMode("approval")}
openHelpOverlay={() => setOverlayMode("help")}
openSessionsOverlay={() => setOverlayMode("sessions")}
openDiffOverlay={() => {
const { isGitRepo, diff } = getGitDiff();
let text: string;
@@ -570,6 +585,25 @@ export default function TerminalChat({
{overlayMode === "history" && (
<HistoryOverlay items={items} onExit={() => setOverlayMode("none")} />
)}
{overlayMode === "sessions" && (
<SessionsOverlay
onView={async (p) => {
try {
const txt = await fs.readFile(p, "utf-8");
const data = JSON.parse(txt) as AppRollout;
setViewRollout(data);
setOverlayMode("none");
} catch {
setOverlayMode("none");
}
}}
onResume={(p) => {
setOverlayMode("none");
setInitialPrompt(`Resume this session: ${p}`);
}}
onExit={() => setOverlayMode("none")}
/>
)}
{overlayMode === "model" && (
<ModelOverlay
currentModel={model}

View File

@@ -2,6 +2,7 @@ import type { OverlayModeType } from "./terminal-chat.js";
import type { TerminalHeaderProps } from "./terminal-header.js";
import type { GroupedResponseItem } from "./use-message-grouping.js";
import type { ResponseItem } from "openai/resources/responses/responses.mjs";
import type { FileOpenerScheme } from "src/utils/config.js";
import TerminalChatResponseItem from "./terminal-chat-response-item.js";
import TerminalHeader from "./terminal-header.js";
@@ -23,6 +24,7 @@ type TerminalMessageHistoryProps = {
headerProps: TerminalHeaderProps;
fullStdout: boolean;
setOverlayMode: React.Dispatch<React.SetStateAction<OverlayModeType>>;
fileOpener: FileOpenerScheme | undefined;
};
const TerminalMessageHistory: React.FC<TerminalMessageHistoryProps> = ({
@@ -33,6 +35,7 @@ const TerminalMessageHistory: React.FC<TerminalMessageHistoryProps> = ({
thinkingSeconds: _thinkingSeconds,
fullStdout,
setOverlayMode,
fileOpener,
}) => {
// Flatten batch entries to response items.
const messages = useMemo(() => batch.map(({ item }) => item!), [batch]);
@@ -59,16 +62,25 @@ const TerminalMessageHistory: React.FC<TerminalMessageHistoryProps> = ({
key={`${message.id}-${index}`}
flexDirection="column"
marginLeft={
message.type === "message" && message.role === "user" ? 0 : 4
message.type === "message" &&
(message.role === "user" || message.role === "assistant")
? 0
: 4
}
marginTop={
message.type === "message" && message.role === "user" ? 0 : 1
}
marginBottom={
message.type === "message" && message.role === "assistant"
? 1
: 0
}
>
<TerminalChatResponseItem
item={message}
fullStdout={fullStdout}
setOverlayMode={setOverlayMode}
fileOpener={fileOpener}
/>
</Box>
);

View File

@@ -0,0 +1,130 @@
import type { TypeaheadItem } from "./typeahead-overlay.js";
import TypeaheadOverlay from "./typeahead-overlay.js";
import fs from "fs/promises";
import { Box, Text, useInput } from "ink";
import os from "os";
import path from "path";
import React, { useEffect, useState } from "react";
const SESSIONS_ROOT = path.join(os.homedir(), ".codex", "sessions");
export type SessionMeta = {
path: string;
timestamp: string;
userMessages: number;
toolCalls: number;
firstMessage: string;
};
async function loadSessions(): Promise<Array<SessionMeta>> {
try {
const entries = await fs.readdir(SESSIONS_ROOT);
const sessions: Array<SessionMeta> = [];
for (const entry of entries) {
if (!entry.endsWith(".json")) {
continue;
}
const filePath = path.join(SESSIONS_ROOT, entry);
try {
// eslint-disable-next-line no-await-in-loop
const content = await fs.readFile(filePath, "utf-8");
const data = JSON.parse(content) as {
session?: { timestamp?: string };
items?: Array<{
type: string;
role: string;
content: Array<{ text: string }>;
}>;
};
const items = Array.isArray(data.items) ? data.items : [];
const firstUser = items.find(
(i) => i?.type === "message" && i.role === "user",
);
const firstText =
firstUser?.content?.[0]?.text?.replace(/\n/g, " ").slice(0, 16) ?? "";
const userMessages = items.filter(
(i) => i?.type === "message" && i.role === "user",
).length;
const toolCalls = items.filter(
(i) => i?.type === "function_call",
).length;
sessions.push({
path: filePath,
timestamp: data.session?.timestamp || "",
userMessages,
toolCalls,
firstMessage: firstText,
});
} catch {
/* ignore invalid session */
}
}
sessions.sort((a, b) => b.timestamp.localeCompare(a.timestamp));
return sessions;
} catch {
return [];
}
}
type Props = {
onView: (sessionPath: string) => void;
onResume: (sessionPath: string) => void;
onExit: () => void;
};
export default function SessionsOverlay({
onView,
onResume,
onExit,
}: Props): JSX.Element {
const [items, setItems] = useState<Array<TypeaheadItem>>([]);
const [mode, setMode] = useState<"view" | "resume">("view");
useEffect(() => {
(async () => {
const sessions = await loadSessions();
const formatted = sessions.map((s) => {
const ts = s.timestamp
? new Date(s.timestamp).toLocaleString(undefined, {
dateStyle: "short",
timeStyle: "short",
})
: "";
const first = s.firstMessage?.slice(0, 50);
const label = `${ts} · ${s.userMessages} msgs/${s.toolCalls} tools · ${first}`;
return { label, value: s.path } as TypeaheadItem;
});
setItems(formatted);
})();
}, []);
useInput((_input, key) => {
if (key.tab) {
setMode((m) => (m === "view" ? "resume" : "view"));
}
});
return (
<TypeaheadOverlay
title={mode === "view" ? "View session" : "Resume session"}
description={
<Box flexDirection="column">
<Text>
{mode === "view" ? "press enter to view" : "press enter to resume"}
</Text>
<Text dimColor>tab to toggle mode · esc to cancel</Text>
</Box>
}
initialItems={items}
onSelect={(value) => {
if (mode === "view") {
onView(value);
} else {
onResume(value);
}
}}
onExit={onExit}
/>
);
}

View File

@@ -5,13 +5,7 @@ import type { FileOperation } from "../utils/singlepass/file_ops";
import Spinner from "./vendor/ink-spinner"; // Thirdparty / vendor components
import TextInput from "./vendor/ink-text-input";
import {
OPENAI_TIMEOUT_MS,
OPENAI_ORGANIZATION,
OPENAI_PROJECT,
getBaseUrl,
getApiKey,
} from "../utils/config";
import { createOpenAIClient } from "../utils/openai-client";
import {
generateDiffSummary,
generateEditSummary,
@@ -26,7 +20,6 @@ import { EditedFilesSchema } from "../utils/singlepass/file_ops";
import * as fsSync from "fs";
import * as fsPromises from "fs/promises";
import { Box, Text, useApp, useInput } from "ink";
import OpenAI from "openai";
import { zodResponseFormat } from "openai/helpers/zod";
import path from "path";
import React, { useEffect, useState, useRef } from "react";
@@ -399,20 +392,7 @@ export function SinglePassApp({
files,
});
const headers: Record<string, string> = {};
if (OPENAI_ORGANIZATION) {
headers["OpenAI-Organization"] = OPENAI_ORGANIZATION;
}
if (OPENAI_PROJECT) {
headers["OpenAI-Project"] = OPENAI_PROJECT;
}
const openai = new OpenAI({
apiKey: getApiKey(config.provider),
baseURL: getBaseUrl(config.provider),
timeout: OPENAI_TIMEOUT_MS,
defaultHeaders: headers,
});
const openai = createOpenAIClient(config);
const chatResp = await openai.beta.chat.completions.parse({
model: config.model,
...(config.flexMode ? { service_tier: "flex" } : {}),

View File

@@ -8,30 +8,34 @@ import type {
ResponseItem,
ResponseCreateParams,
FunctionTool,
Tool,
} from "openai/resources/responses/responses.mjs";
import type { Reasoning } from "openai/resources.mjs";
import { CLI_VERSION } from "../../version.js";
import {
OPENAI_TIMEOUT_MS,
OPENAI_ORGANIZATION,
OPENAI_PROJECT,
getApiKey,
getBaseUrl,
AZURE_OPENAI_API_VERSION,
} from "../config.js";
import { log } from "../logger/log.js";
import { parseToolCallArguments } from "../parsers.js";
import { responsesCreateViaChatCompletions } from "../responses.js";
import {
ORIGIN,
CLI_VERSION,
getSessionId,
setCurrentModel,
setSessionId,
} from "../session.js";
import { applyPatchToolInstructions } from "./apply-patch.js";
import { handleExecCommand } from "./handle-exec-command.js";
import { HttpsProxyAgent } from "https-proxy-agent";
import { spawnSync } from "node:child_process";
import { randomUUID } from "node:crypto";
import OpenAI, { APIConnectionTimeoutError } from "openai";
import OpenAI, { APIConnectionTimeoutError, AzureOpenAI } from "openai";
import os from "os";
// Wait time before retrying after rate limit errors (ms).
const RATE_LIMIT_RETRY_WAIT_MS = parseInt(
@@ -80,7 +84,7 @@ type AgentLoopParams = {
onLastResponseId: (lastResponseId: string) => void;
};
const shellTool: FunctionTool = {
const shellFunctionTool: FunctionTool = {
type: "function",
name: "shell",
description: "Runs a shell command, and returns its output.",
@@ -104,6 +108,11 @@ const shellTool: FunctionTool = {
},
};
const localShellTool: Tool = {
//@ts-expect-error - waiting on sdk
type: "local_shell",
};
export class AgentLoop {
private model: string;
private provider: string;
@@ -297,7 +306,7 @@ export class AgentLoop {
this.sessionId = getSessionId() || randomUUID().replaceAll("-", "");
// Configure OpenAI client with optional timeout (ms) from environment
const timeoutMs = OPENAI_TIMEOUT_MS;
const apiKey = getApiKey(this.provider);
const apiKey = this.config.apiKey ?? process.env["OPENAI_API_KEY"] ?? "";
const baseURL = getBaseUrl(this.provider);
this.oai = new OpenAI({
@@ -322,6 +331,25 @@ export class AgentLoop {
...(timeoutMs !== undefined ? { timeout: timeoutMs } : {}),
});
if (this.provider.toLowerCase() === "azure") {
this.oai = new AzureOpenAI({
apiKey,
baseURL,
apiVersion: AZURE_OPENAI_API_VERSION,
defaultHeaders: {
originator: ORIGIN,
version: CLI_VERSION,
session_id: this.sessionId,
...(OPENAI_ORGANIZATION
? { "OpenAI-Organization": OPENAI_ORGANIZATION }
: {}),
...(OPENAI_PROJECT ? { "OpenAI-Project": OPENAI_PROJECT } : {}),
},
httpAgent: PROXY_URL ? new HttpsProxyAgent(PROXY_URL) : undefined,
...(timeoutMs !== undefined ? { timeout: timeoutMs } : {}),
});
}
setSessionId(this.sessionId);
setCurrentModel(this.model);
@@ -438,6 +466,73 @@ export class AgentLoop {
return [outputItem, ...additionalItems];
}
private async handleLocalShellCall(
// eslint-disable-next-line @typescript-eslint/no-explicit-any
item: any,
): Promise<Array<ResponseInputItem>> {
// If the agent has been canceled in the meantime we should not perform any
// additional work. Returning an empty array ensures that we neither execute
// the requested tool call nor enqueue any followup input items. This keeps
// the cancellation semantics intuitive for users once they interrupt a
// task no further actions related to that task should be taken.
if (this.canceled) {
return [];
}
// eslint-disable-next-line @typescript-eslint/no-explicit-any
const outputItem: any = {
type: "local_shell_call_output",
// `call_id` is mandatory ensure we never send `undefined` which would
// trigger the "No tool output found…" 400 from the API.
call_id: item.call_id,
output: "no function found",
};
// We intentionally *do not* remove this `callId` from the `pendingAborts`
// set right away. The output produced below is only queued up for the
// *next* request to the OpenAI API it has not been delivered yet. If
// the user presses ESCESC (i.e. invokes `cancel()`) in the small window
// between queuing the result and the actual network call, we need to be
// able to surface a synthetic `function_call_output` marked as
// "aborted". Keeping the ID in the set until the run concludes
// successfully lets the next `run()` differentiate between an aborted
// tool call (needs the synthetic output) and a completed one (cleared
// below in the `flush()` helper).
// used to tell model to stop if needed
const additionalItems: Array<ResponseInputItem> = [];
if (item.action.type !== "exec") {
throw new Error("Invalid action type");
}
const args = {
cmd: item.action.command,
workdir: item.action.working_directory,
timeoutInMillis: item.action.timeout_ms,
};
const {
outputText,
metadata,
additionalItems: additionalItemsFromExec,
} = await handleExecCommand(
args,
this.config,
this.approvalPolicy,
this.additionalWritableRoots,
this.getCommandConfirmation,
this.execAbortController?.signal,
);
outputItem.output = JSON.stringify({ output: outputText, metadata });
if (additionalItemsFromExec) {
additionalItems.push(...additionalItemsFromExec);
}
return [outputItem, ...additionalItems];
}
public async run(
input: Array<ResponseInputItem>,
previousResponseId: string = "",
@@ -522,6 +617,11 @@ export class AgentLoop {
// `disableResponseStorage === true`.
let transcriptPrefixLen = 0;
let tools: Array<Tool> = [shellFunctionTool];
if (this.model.startsWith("codex")) {
tools = [localShellTool];
}
const stripInternalFields = (
item: ResponseInputItem,
): ResponseInputItem => {
@@ -625,6 +725,8 @@ export class AgentLoop {
if (
(item as ResponseInputItem).type === "function_call" ||
(item as ResponseInputItem).type === "reasoning" ||
//@ts-expect-error - waiting on sdk
(item as ResponseInputItem).type === "local_shell_call" ||
((item as ResponseInputItem).type === "message" &&
// eslint-disable-next-line @typescript-eslint/no-explicit-any
(item as any).role === "user")
@@ -663,7 +765,7 @@ export class AgentLoop {
// prompts) and so that freshly generated `function_call_output`s are
// shown immediately.
// Figure out what subset of `turnInput` constitutes *new* information
// for the UI so that we dont spam the interface with repeats of the
// for the UI so that we don't spam the interface with repeats of the
// entire transcript on every iteration when response storage is
// disabled.
const deltaInput = this.disableResponseStorage
@@ -680,13 +782,19 @@ export class AgentLoop {
for (let attempt = 1; attempt <= MAX_RETRIES; attempt++) {
try {
let reasoning: Reasoning | undefined;
if (this.model.startsWith("o")) {
reasoning = { effort: this.config.reasoningEffort ?? "high" };
if (this.model === "o3" || this.model === "o4-mini") {
reasoning.summary = "auto";
}
let modelSpecificInstructions: string | undefined;
if (this.model.startsWith("o") || this.model.startsWith("codex")) {
reasoning = { effort: this.config.reasoningEffort ?? "medium" };
reasoning.summary = "auto";
}
const mergedInstructions = [prefix, this.instructions]
if (this.model.startsWith("gpt-4.1")) {
modelSpecificInstructions = applyPatchToolInstructions;
}
const mergedInstructions = [
prefix,
modelSpecificInstructions,
this.instructions,
]
.filter(Boolean)
.join("\n");
@@ -719,7 +827,7 @@ export class AgentLoop {
store: true,
previous_response_id: lastResponseId || undefined,
}),
tools: [shellTool],
tools: tools,
// Explicitly tell the model it is allowed to pick whatever
// tool it deems appropriate. Omitting this sometimes leads to
// the model ignoring the available tools and responding with
@@ -744,7 +852,13 @@ export class AgentLoop {
const errCtx = error as any;
const status =
errCtx?.status ?? errCtx?.httpStatus ?? errCtx?.statusCode;
const isServerError = typeof status === "number" && status >= 500;
// Treat classical 5xx *and* explicit OpenAI `server_error` types
// as transient server-side failures that qualify for a retry. The
// SDK often omits the numeric status for these, reporting only
// the `type` field.
const isServerError =
(typeof status === "number" && status >= 500) ||
errCtx?.type === "server_error";
if (
(isTimeout || isServerError || isConnectionError) &&
attempt < MAX_RETRIES
@@ -933,7 +1047,10 @@ export class AgentLoop {
if (maybeReasoning.type === "reasoning") {
maybeReasoning.duration_ms = Date.now() - thinkingStart;
}
if (item.type === "function_call") {
if (
item.type === "function_call" ||
item.type === "local_shell_call"
) {
// Track outstanding tool call so we can abort later if needed.
// The item comes from the streaming response, therefore it has
// either `id` (chat) or `call_id` (responses) we normalise
@@ -1056,7 +1173,11 @@ export class AgentLoop {
let reasoning: Reasoning | undefined;
if (this.model.startsWith("o")) {
reasoning = { effort: "high" };
if (this.model === "o3" || this.model === "o4-mini") {
if (
this.model === "o3" ||
this.model === "o4-mini" ||
this.model === "codex-mini-latest"
) {
reasoning.summary = "auto";
}
}
@@ -1095,7 +1216,7 @@ export class AgentLoop {
store: true,
previous_response_id: lastResponseId || undefined,
}),
tools: [shellTool],
tools: tools,
tool_choice: "auto",
});
@@ -1457,6 +1578,17 @@ export class AgentLoop {
// eslint-disable-next-line no-await-in-loop
const result = await this.handleFunctionCall(item);
turnInput.push(...result);
//@ts-expect-error - waiting on sdk
} else if (item.type === "local_shell_call") {
//@ts-expect-error - waiting on sdk
if (alreadyProcessedResponses.has(item.id)) {
continue;
}
//@ts-expect-error - waiting on sdk
alreadyProcessedResponses.add(item.id);
// eslint-disable-next-line no-await-in-loop
const result = await this.handleLocalShellCall(item);
turnInput.push(...result);
}
emitItem(item as ResponseItem);
}
@@ -1464,6 +1596,19 @@ export class AgentLoop {
}
}
// Dynamic developer message prefix: includes user, workdir, and rg suggestion.
const userName = os.userInfo().username;
const workdir = process.cwd();
const dynamicLines: Array<string> = [
`User: ${userName}`,
`Workdir: ${workdir}`,
];
if (spawnSync("rg", ["--version"], { stdio: "ignore" }).status === 0) {
dynamicLines.push(
"- Always use rg instead of grep/ls -R because it is much faster and respects gitignore",
);
}
const dynamicPrefix = dynamicLines.join("\n");
const prefix = `You are operating as and within the Codex CLI, a terminal-based agentic coding assistant built by OpenAI. It wraps OpenAI models to enable natural language interaction with a local codebase. You are expected to be precise, safe, and helpful.
You can:
@@ -1499,7 +1644,6 @@ You MUST adhere to the following criteria when executing the task:
- If there is a .pre-commit-config.yaml, use \`pre-commit run --files ...\` to check that your changes pass the pre-commit checks. However, do not fix pre-existing errors on lines you didn't touch.
- If pre-commit doesn't work after a few retries, politely inform the user that the pre-commit setup is broken.
- Once you finish coding, you must
- Check \`git status\` to sanity check your changes; revert any scratch files or changes.
- Remove all inline comments you added as much as possible, even if they look normal. Check using \`git diff\`. Inline comments must be generally avoided, unless active maintainers of the repo, after long careful study of the code and the issue, will still misinterpret the code without the comments.
- Check if you accidentally add copyright or license headers. If so, remove them.
- Try to run pre-commit if it is available.
@@ -1509,7 +1653,9 @@ You MUST adhere to the following criteria when executing the task:
- Respond in a friendly tone as a remote teammate, who is knowledgeable, capable and eager to help with coding.
- When your task involves writing or modifying files:
- Do NOT tell the user to "save the file" or "copy the code into a file" if you already created or modified the file using \`apply_patch\`. Instead, reference the file as already saved.
- Do NOT show the full contents of large files you have already written, unless the user explicitly asks for them.`;
- Do NOT show the full contents of large files you have already written, unless the user explicitly asks for them.
${dynamicPrefix}`;
function filterToApiMessages(
items: Array<ResponseInputItem>,

View File

@@ -550,7 +550,15 @@ export function text_to_patch(
!(lines[0] ?? "").startsWith(PATCH_PREFIX.trim()) ||
lines[lines.length - 1] !== PATCH_SUFFIX.trim()
) {
throw new DiffError("Invalid patch text");
let reason = "Invalid patch text: ";
if (lines.length < 2) {
reason += "Patch text must have at least two lines.";
} else if (!(lines[0] ?? "").startsWith(PATCH_PREFIX.trim())) {
reason += "Patch text must start with the correct patch prefix.";
} else if (lines[lines.length - 1] !== PATCH_SUFFIX.trim()) {
reason += "Patch text must end with the correct patch suffix.";
}
throw new DiffError(reason);
}
const parser = new Parser(orig, lines);
parser.index = 1;
@@ -762,3 +770,46 @@ if (import.meta.url === `file://${process.argv[1]}`) {
}
});
}
export const applyPatchToolInstructions = `
To edit files, ALWAYS use the \`shell\` tool with \`apply_patch\` CLI. \`apply_patch\` effectively allows you to execute a diff/patch against a file, but the format of the diff specification is unique to this task, so pay careful attention to these instructions. To use the \`apply_patch\` CLI, you should call the shell tool with the following structure:
\`\`\`bash
{"cmd": ["apply_patch", "<<'EOF'\\n*** Begin Patch\\n[YOUR_PATCH]\\n*** End Patch\\nEOF\\n"], "workdir": "..."}
\`\`\`
Where [YOUR_PATCH] is the actual content of your patch, specified in the following V4A diff format.
*** [ACTION] File: [path/to/file] -> ACTION can be one of Add, Update, or Delete.
For each snippet of code that needs to be changed, repeat the following:
[context_before] -> See below for further instructions on context.
- [old_code] -> Precede the old code with a minus sign.
+ [new_code] -> Precede the new, replacement code with a plus sign.
[context_after] -> See below for further instructions on context.
For instructions on [context_before] and [context_after]:
- By default, show 3 lines of code immediately above and 3 lines immediately below each change. If a change is within 3 lines of a previous change, do NOT duplicate the first changes [context_after] lines in the second changes [context_before] lines.
- If 3 lines of context is insufficient to uniquely identify the snippet of code within the file, use the @@ operator to indicate the class or function to which the snippet belongs. For instance, we might have:
@@ class BaseClass
[3 lines of pre-context]
- [old_code]
+ [new_code]
[3 lines of post-context]
- If a code block is repeated so many times in a class or function such that even a single \`@@\` statement and 3 lines of context cannot uniquely identify the snippet of code, you can use multiple \`@@\` statements to jump to the right context. For instance:
@@ class BaseClass
@@ def method():
[3 lines of pre-context]
- [old_code]
+ [new_code]
[3 lines of post-context]
Note, then, that we do not use line numbers in this diff format, as the context is enough to uniquely identify code. An example of a message that you might pass as "input" to this function, in order to apply a patch, is shown below.
\`\`\`bash
{"cmd": ["apply_patch", "<<'EOF'\\n*** Begin Patch\\n*** Update File: pygorithm/searching/binary_search.py\\n@@ class BaseClass\\n@@ def search():\\n- pass\\n+ raise NotImplementedError()\\n@@ class Subclass\\n@@ def search():\\n- pass\\n+ raise NotImplementedError()\\n*** End Patch\\nEOF\\n"], "workdir": "..."}
\`\`\`
File references can only be relative, NEVER ABSOLUTE. After the apply_patch command is run, it will always say "Done!", regardless of whether the patch was successfully applied or not. However, you can determine if there are issue and errors by looking at any warnings or logging lines printed BEFORE the "Done!" is output.
`;

View File

@@ -9,11 +9,13 @@ import { execWithLandlock } from "./sandbox/landlock.js";
import { execWithSeatbelt } from "./sandbox/macos-seatbelt.js";
import { exec as rawExec } from "./sandbox/raw-exec.js";
import { formatCommandForDisplay } from "../../format-command.js";
import { log } from "../logger/log.js";
import fs from "fs";
import os from "os";
import path from "path";
import { parse } from "shell-quote";
import { resolvePathAgainstWorkdir } from "src/approvals.js";
import { PATCH_SUFFIX } from "src/parse-apply-patch.js";
const DEFAULT_TIMEOUT_MS = 10_000; // 10 seconds
@@ -81,12 +83,22 @@ export function execApplyPatch(
patchText: string,
workdir: string | undefined = undefined,
): ExecResult {
// This is a temporary measure to understand what are the common base commands
// until we start persisting and uploading rollouts
// This find/replace is required from some models like 4.1 where the patch
// text is wrapped in quotes that breaks the apply_patch command.
let applyPatchInput = patchText
.replace(/('|")?<<('|")EOF('|")/, "")
.replace(/\*\*\* End Patch\nEOF('|")?/, "*** End Patch")
.trim();
if (!applyPatchInput.endsWith(PATCH_SUFFIX)) {
applyPatchInput += "\n" + PATCH_SUFFIX;
}
log(`Applying patch: \`\`\`${applyPatchInput}\`\`\`\n\n`);
try {
const result = process_patch(
patchText,
applyPatchInput,
(p) => fs.readFileSync(resolvePathAgainstWorkdir(p, workdir), "utf8"),
(p, c) => {
const resolvedPath = resolvePathAgainstWorkdir(p, workdir);

View File

@@ -1,7 +1,7 @@
import type { AgentName } from "package-manager-detector";
import { detectInstallerByPath } from "./package-manager-detector";
import { CLI_VERSION } from "./session";
import { CLI_VERSION } from "../version";
import boxen from "boxen";
import chalk from "chalk";
import { getLatestVersion } from "fast-npm-meta";

View File

@@ -1,12 +1,14 @@
import type { AppConfig } from "./config.js";
import type { ResponseItem } from "openai/resources/responses/responses.mjs";
import { getBaseUrl, getApiKey } from "./config.js";
import OpenAI from "openai";
import { createOpenAIClient } from "./openai-client.js";
/**
* Generate a condensed summary of the conversation items.
* @param items The list of conversation items to summarize
* @param model The model to use for generating the summary
* @param flexMode Whether to use the flex-mode service tier
* @param config The configuration object
* @returns A concise structured summary string
*/
/**
@@ -23,10 +25,7 @@ export async function generateCompactSummary(
flexMode = false,
config: AppConfig,
): Promise<string> {
const oai = new OpenAI({
apiKey: getApiKey(config.provider),
baseURL: getBaseUrl(config.provider),
});
const oai = createOpenAIClient(config);
const conversationText = items
.filter(

View File

@@ -43,7 +43,7 @@ if (!isVitest) {
loadDotenv({ path: USER_WIDE_CONFIG_PATH });
}
export const DEFAULT_AGENTIC_MODEL = "o4-mini";
export const DEFAULT_AGENTIC_MODEL = "codex-mini-latest";
export const DEFAULT_FULL_CONTEXT_MODEL = "gpt-4.1";
export const DEFAULT_APPROVAL_MODE = AutoApprovalMode.SUGGEST;
export const DEFAULT_INSTRUCTIONS = "";
@@ -68,12 +68,15 @@ export const OPENAI_TIMEOUT_MS =
export const OPENAI_BASE_URL = process.env["OPENAI_BASE_URL"] || "";
export let OPENAI_API_KEY = process.env["OPENAI_API_KEY"] || "";
export const AZURE_OPENAI_API_VERSION =
process.env["AZURE_OPENAI_API_VERSION"] || "2025-03-01-preview";
export const DEFAULT_REASONING_EFFORT = "high";
export const OPENAI_ORGANIZATION = process.env["OPENAI_ORGANIZATION"] || "";
export const OPENAI_PROJECT = process.env["OPENAI_PROJECT"] || "";
// Can be set `true` when Codex is running in an environment that is marked as already
// considered sufficiently locked-down so that we allow running wihtout an explicit sandbox.
// considered sufficiently locked-down so that we allow running without an explicit sandbox.
export const CODEX_UNSAFE_ALLOW_NO_SANDBOX = Boolean(
process.env["CODEX_UNSAFE_ALLOW_NO_SANDBOX"] || "",
);
@@ -117,7 +120,7 @@ export function getApiKey(provider: string = "openai"): string | undefined {
return process.env[providerInfo.envKey];
}
// Checking `PROVIDER_API_KEY feels more intuitive with a custom provider.
// Checking `PROVIDER_API_KEY` feels more intuitive with a custom provider.
const customApiKey = process.env[`${provider.toUpperCase()}_API_KEY`];
if (customApiKey) {
return customApiKey;
@@ -132,6 +135,8 @@ export function getApiKey(provider: string = "openai"): string | undefined {
return undefined;
}
export type FileOpenerScheme = "vscode" | "cursor" | "windsurf";
// Represents config as persisted in config.json.
export type StoredConfig = {
model?: string;
@@ -143,6 +148,7 @@ export type StoredConfig = {
notify?: boolean;
/** Disable server-side response storage (send full transcript each request) */
disableResponseStorage?: boolean;
flexMode?: boolean;
providers?: Record<string, { name: string; baseURL: string; envKey: string }>;
history?: {
maxSize?: number;
@@ -158,6 +164,12 @@ export type StoredConfig = {
/** User-defined safe commands */
safeCommands?: Array<string>;
reasoningEffort?: ReasoningEffort;
/**
* URI-based file opener. This is used when linking code references in
* terminal output.
*/
fileOpener?: FileOpenerScheme;
};
// Minimal config written on first run. An *empty* model string ensures that
@@ -202,18 +214,29 @@ export type AppConfig = {
maxLines: number;
};
};
fileOpener?: FileOpenerScheme;
};
// Formatting (quiet mode-only).
export const PRETTY_PRINT = Boolean(process.env["PRETTY_PRINT"] || "");
// ---------------------------------------------------------------------------
// Project doc support (codex.md)
// Project doc support (AGENTS.md / codex.md)
// ---------------------------------------------------------------------------
export const PROJECT_DOC_MAX_BYTES = 32 * 1024; // 32 kB
const PROJECT_DOC_FILENAMES = ["codex.md", ".codex.md", "CODEX.md"];
// We support multiple filenames for project-level agent instructions. As of
// 2025 the recommended convention is to use `AGENTS.md`, however we keep
// the legacy `codex.md` variants for backwards-compatibility so that existing
// repositories continue to work without changes. The list is ordered so that
// the first match wins newer conventions first, older fallbacks later.
const PROJECT_DOC_FILENAMES = [
"AGENTS.md", // preferred
"codex.md", // legacy
".codex.md",
"CODEX.md",
];
const PROJECT_DOC_SEPARATOR = "\n\n--- project-doc ---\n\n";
export function discoverProjectDocPath(startDir: string): string | null {
@@ -254,7 +277,8 @@ export function discoverProjectDocPath(startDir: string): string | null {
}
/**
* Load the project documentation markdown (codex.md) if present. If the file
* Load the project documentation markdown (`AGENTS.md` or the legacy
* `codex.md`) if present. If the file
* exceeds {@link PROJECT_DOC_MAX_BYTES} it will be truncated and a warning is
* logged.
*
@@ -414,6 +438,7 @@ export const loadConfig = (
},
disableResponseStorage: storedConfig.disableResponseStorage === true,
reasoningEffort: storedConfig.reasoningEffort,
fileOpener: storedConfig.fileOpener,
};
// -----------------------------------------------------------------------
@@ -475,6 +500,10 @@ export const loadConfig = (
}
// Notification setting: enable desktop notifications when set in config
config.notify = storedConfig.notify === true;
// Flex-mode setting: enable the flex-mode service tier when set in config
if (storedConfig.flexMode !== undefined) {
config.flexMode = storedConfig.flexMode;
}
// Add default history config if not provided
if (storedConfig.history !== undefined) {
@@ -529,6 +558,7 @@ export const saveConfig = (
providers: config.providers,
approvalMode: config.approvalMode,
disableResponseStorage: config.disableResponseStorage,
flexMode: config.flexMode,
reasoningEffort: config.reasoningEffort,
};

View File

@@ -0,0 +1,75 @@
import SelectInput from "../components/select-input/select-input.js";
import Spinner from "../components/vendor/ink-spinner.js";
import TextInput from "../components/vendor/ink-text-input.js";
import { Box, Text } from "ink";
import React, { useState } from "react";
export type Choice = { type: "signin" } | { type: "apikey"; key: string };
export function ApiKeyPrompt({
onDone,
}: {
onDone: (choice: Choice) => void;
}): JSX.Element {
const [step, setStep] = useState<"select" | "paste">("select");
const [apiKey, setApiKey] = useState("");
if (step === "select") {
return (
<Box flexDirection="column" gap={1}>
<Box flexDirection="column">
<Text>
Sign in with ChatGPT to generate an API key or paste one you already
have.
</Text>
<Text dimColor>[use arrows to move, enter to select]</Text>
</Box>
<SelectInput
items={[
{ label: "Sign in with ChatGPT", value: "signin" },
{
label: "Paste an API key (or set as OPENAI_API_KEY)",
value: "paste",
},
]}
onSelect={(item: { value: string }) => {
if (item.value === "signin") {
onDone({ type: "signin" });
} else {
setStep("paste");
}
}}
/>
</Box>
);
}
return (
<Box flexDirection="column">
<Text>Paste your OpenAI API key and press &lt;Enter&gt;:</Text>
<TextInput
value={apiKey}
onChange={setApiKey}
onSubmit={(value: string) => {
if (value.trim() !== "") {
onDone({ type: "apikey", key: value.trim() });
}
}}
placeholder="sk-..."
mask="*"
/>
</Box>
);
}
export function WaitingForAuth(): JSX.Element {
return (
<Box flexDirection="row" marginTop={1}>
<Spinner type="ball" />
<Text>
{" "}
Waiting for authentication <Text dimColor>ctrl + c to quit</Text>
</Text>
</Box>
);
}

View File

@@ -0,0 +1,764 @@
import type { Choice } from "./get-api-key-components";
import type { Request, Response } from "express";
import { ApiKeyPrompt, WaitingForAuth } from "./get-api-key-components";
import chalk from "chalk";
import express from "express";
import fs from "fs/promises";
import { render } from "ink";
import crypto from "node:crypto";
import { URL } from "node:url";
import open from "open";
import os from "os";
import path from "path";
import React from "react";
function promptUserForChoice(): Promise<Choice> {
return new Promise<Choice>((resolve) => {
const instance = render(
<ApiKeyPrompt
onDone={(choice: Choice) => {
resolve(choice);
instance.unmount();
}}
/>,
);
});
}
interface OidcConfiguration {
issuer: string;
authorization_endpoint: string;
token_endpoint: string;
}
async function getOidcConfiguration(
issuer: string,
): Promise<OidcConfiguration> {
const discoveryUrl = new URL(issuer);
discoveryUrl.pathname = "/.well-known/openid-configuration";
if (issuer === "https://auth.openai.com") {
// Account for legacy quirk in production tenant
discoveryUrl.pathname = "/v2.0" + discoveryUrl.pathname;
}
const res = await fetch(discoveryUrl.toString());
if (!res.ok) {
throw new Error("Failed to fetch OIDC configuration");
}
return (await res.json()) as OidcConfiguration;
}
interface IDTokenClaims {
"exp": number;
"https://api.openai.com/auth": {
organization_id: string;
project_id: string;
completed_platform_onboarding: boolean;
is_org_owner: boolean;
chatgpt_subscription_active_start: string;
chatgpt_subscription_active_until: string;
chatgpt_plan_type: string;
};
}
interface AccessTokenClaims {
"https://api.openai.com/auth": {
chatgpt_plan_type: string;
};
}
function generatePKCECodes(): {
code_verifier: string;
code_challenge: string;
} {
const code_verifier = crypto.randomBytes(64).toString("hex");
const code_challenge = crypto
.createHash("sha256")
.update(code_verifier)
.digest("base64url");
return { code_verifier, code_challenge };
}
async function maybeRedeemCredits(
issuer: string,
clientId: string,
refreshToken: string,
idToken?: string,
): Promise<void> {
try {
let currentIdToken = idToken;
let idClaims: IDTokenClaims | undefined;
if (
currentIdToken &&
typeof currentIdToken === "string" &&
currentIdToken.split(".")[1]
) {
idClaims = JSON.parse(
Buffer.from(currentIdToken.split(".")[1]!, "base64url").toString(
"utf8",
),
) as IDTokenClaims;
} else {
currentIdToken = "";
}
// Validate idToken expiration
// if expired, attempt token-exchange for a fresh idToken
if (!idClaims || !idClaims.exp || Date.now() >= idClaims.exp * 1000) {
// eslint-disable-next-line no-console
console.log(chalk.dim("Refreshing credentials..."));
try {
const refreshRes = await fetch("https://auth.openai.com/oauth/token", {
method: "POST",
headers: { "Content-Type": "application/json" },
body: JSON.stringify({
client_id: clientId,
grant_type: "refresh_token",
refresh_token: refreshToken,
scope: "openid profile email",
}),
});
if (!refreshRes.ok) {
// eslint-disable-next-line no-console
console.warn(
`Failed to refresh credentials: ${refreshRes.status} ${refreshRes.statusText}\n${chalk.dim(await refreshRes.text())}`,
);
// eslint-disable-next-line no-console
console.warn(
`Please sign in again to redeem credits: ${chalk.bold("codex --login")}`,
);
return;
}
const refreshData = (await refreshRes.json()) as {
id_token: string;
refresh_token?: string;
};
currentIdToken = refreshData.id_token;
idClaims = JSON.parse(
Buffer.from(currentIdToken.split(".")[1]!, "base64url").toString(
"utf8",
),
) as IDTokenClaims;
if (refreshData.refresh_token) {
try {
const home = os.homedir();
const authDir = path.join(home, ".codex");
const authFile = path.join(authDir, "auth.json");
const existingJson = JSON.parse(
await fs.readFile(authFile, "utf-8"),
);
existingJson.tokens.id_token = currentIdToken;
existingJson.tokens.refresh_token = refreshData.refresh_token;
existingJson.last_refresh = new Date().toISOString();
await fs.writeFile(
authFile,
JSON.stringify(existingJson, null, 2),
{ mode: 0o600 },
);
} catch (err) {
// eslint-disable-next-line no-console
console.warn("Unable to update refresh token in auth file:", err);
}
}
} catch (err) {
// eslint-disable-next-line no-console
console.warn("Unable to refresh ID token via token-exchange:", err);
return;
}
}
// Confirm the subscription is active for more than 7 days
const subStart =
idClaims["https://api.openai.com/auth"]
?.chatgpt_subscription_active_start;
if (
typeof subStart === "string" &&
Date.now() - new Date(subStart).getTime() < 7 * 24 * 60 * 60 * 1000
) {
// eslint-disable-next-line no-console
console.warn(
"Sorry, your subscription must be active for more than 7 days to redeem credits.\nMore info: " +
chalk.dim("https://help.openai.com/en/articles/11381614") +
chalk.bold(
"\nPlease try again on " +
new Date(
new Date(subStart).getTime() + 7 * 24 * 60 * 60 * 1000,
).toLocaleDateString() +
" " +
new Date(
new Date(subStart).getTime() + 7 * 24 * 60 * 60 * 1000,
).toLocaleTimeString(),
),
);
return;
}
const completed = Boolean(
idClaims["https://api.openai.com/auth"]?.completed_platform_onboarding,
);
const isOwner = Boolean(
idClaims["https://api.openai.com/auth"]?.is_org_owner,
);
const needsSetup = !completed && isOwner;
const planType = idClaims["https://api.openai.com/auth"]
?.chatgpt_plan_type as string | undefined;
if (needsSetup || !(planType === "plus" || planType === "pro")) {
// eslint-disable-next-line no-console
console.warn(
"Users with Plus or Pro subscriptions can redeem free API credits.\nMore info: " +
chalk.dim("https://help.openai.com/en/articles/11381614"),
);
return;
}
const apiHost =
issuer === "https://auth.openai.com"
? "https://api.openai.com"
: "https://api.openai.org";
const redeemRes = await fetch(`${apiHost}/v1/billing/redeem_credits`, {
method: "POST",
headers: { "Content-Type": "application/json" },
body: JSON.stringify({ id_token: currentIdToken }),
});
if (!redeemRes.ok) {
// eslint-disable-next-line no-console
console.warn(
`Credit redemption request failed: ${redeemRes.status} ${redeemRes.statusText}`,
);
return;
}
try {
const redeemData = (await redeemRes.json()) as {
granted_chatgpt_subscriber_api_credits?: number;
};
const granted = redeemData?.granted_chatgpt_subscriber_api_credits ?? 0;
if (granted > 0) {
// eslint-disable-next-line no-console
console.log(
chalk.green(
`${chalk.bold(
`Thanks for being a ChatGPT ${
planType === "plus" ? "Plus" : "Pro"
} subscriber!`,
)}\nIf you haven't already redeemed, you should receive ${
planType === "plus" ? "$5" : "$50"
} in API credits\nCredits: ${chalk.dim(chalk.underline("https://platform.openai.com/settings/organization/billing/credit-grants"))}\nMore info: ${chalk.dim(chalk.underline("https://help.openai.com/en/articles/11381614"))}`,
),
);
} else {
// eslint-disable-next-line no-console
console.log(
chalk.green(
`It looks like no credits were granted:\n${JSON.stringify(
redeemData,
null,
2,
)}\nCredits: ${chalk.dim(
chalk.underline(
"https://platform.openai.com/settings/organization/billing/credit-grants",
),
)}\nMore info: ${chalk.dim(
chalk.underline("https://help.openai.com/en/articles/11381614"),
)}`,
),
);
}
} catch (parseErr) {
// eslint-disable-next-line no-console
console.warn("Unable to parse credit redemption response:", parseErr);
}
} catch (err) {
// eslint-disable-next-line no-console
console.warn("Unable to redeem ChatGPT subscriber API credits:", err);
}
}
async function handleCallback(
req: Request,
issuer: string,
oidcConfig: OidcConfiguration,
codeVerifier: string,
clientId: string,
redirectUri: string,
expectedState: string,
): Promise<{ access_token: string; success_url: string }> {
const state = (req.query as Record<string, string>)["state"] as
| string
| undefined;
if (!state || state !== expectedState) {
throw new Error("Invalid state parameter");
}
const code = (req.query as Record<string, string>)["code"] as
| string
| undefined;
if (!code) {
throw new Error("Missing authorization code");
}
const params = new URLSearchParams();
params.append("grant_type", "authorization_code");
params.append("code", code);
params.append("redirect_uri", redirectUri);
params.append("client_id", clientId);
params.append("code_verifier", codeVerifier);
oidcConfig.token_endpoint = `${issuer}/oauth/token`;
const tokenRes = await fetch(oidcConfig.token_endpoint, {
method: "POST",
headers: {
"Content-Type": "application/x-www-form-urlencoded",
},
body: params.toString(),
});
if (!tokenRes.ok) {
throw new Error("Failed to exchange authorization code for tokens");
}
const tokenData = (await tokenRes.json()) as {
id_token: string;
access_token: string;
refresh_token: string;
};
const idTokenParts = tokenData.id_token.split(".");
if (idTokenParts.length !== 3) {
throw new Error("Invalid ID token");
}
const accessTokenParts = tokenData.access_token.split(".");
if (accessTokenParts.length !== 3) {
throw new Error("Invalid access token");
}
const idTokenClaims = JSON.parse(
Buffer.from(idTokenParts[1]!, "base64url").toString("utf8"),
) as IDTokenClaims;
const accessTokenClaims = JSON.parse(
Buffer.from(accessTokenParts[1]!, "base64url").toString("utf8"),
) as AccessTokenClaims;
const org_id = idTokenClaims["https://api.openai.com/auth"]?.organization_id;
if (!org_id) {
throw new Error("Missing organization in id_token claims");
}
const project_id = idTokenClaims["https://api.openai.com/auth"]?.project_id;
if (!project_id) {
throw new Error("Missing project in id_token claims");
}
const randomId = crypto.randomBytes(6).toString("hex");
const exchangeParams = new URLSearchParams({
grant_type: "urn:ietf:params:oauth:grant-type:token-exchange",
client_id: clientId,
requested_token: "openai-api-key",
subject_token: tokenData.id_token,
subject_token_type: "urn:ietf:params:oauth:token-type:id_token",
name: `Codex CLI [auto-generated] (${new Date().toISOString().slice(0, 10)}) [${
randomId
}]`,
});
const exchangeRes = await fetch(oidcConfig.token_endpoint, {
method: "POST",
headers: {
"Content-Type": "application/x-www-form-urlencoded",
},
body: exchangeParams.toString(),
});
if (!exchangeRes.ok) {
throw new Error(`Failed to create API key: ${await exchangeRes.text()}`);
}
const exchanged = (await exchangeRes.json()) as {
access_token: string;
key: string;
};
// Determine whether the organization still requires additional
// setup (e.g., adding a payment method) based on the ID-token
// claim provided by the auth service.
const completedOnboarding = Boolean(
idTokenClaims["https://api.openai.com/auth"]?.completed_platform_onboarding,
);
const chatgptPlanType =
accessTokenClaims["https://api.openai.com/auth"]?.chatgpt_plan_type;
const isOrgOwner = Boolean(
idTokenClaims["https://api.openai.com/auth"]?.is_org_owner,
);
const needsSetup = !completedOnboarding && isOrgOwner;
// Build the success URL on the same host/port as the callback and
// include the required query parameters for the front-end page.
// console.log("Redirecting to success page");
const successUrl = new URL("/success", redirectUri);
if (issuer === "https://auth.openai.com") {
successUrl.searchParams.set("platform_url", "https://platform.openai.com");
} else {
successUrl.searchParams.set(
"platform_url",
"https://platform.api.openai.org",
);
}
successUrl.searchParams.set("id_token", tokenData.id_token);
successUrl.searchParams.set("needs_setup", needsSetup ? "true" : "false");
successUrl.searchParams.set("org_id", org_id);
successUrl.searchParams.set("project_id", project_id);
successUrl.searchParams.set("plan_type", chatgptPlanType);
try {
const home = os.homedir();
const authDir = path.join(home, ".codex");
await fs.mkdir(authDir, { recursive: true });
const authFile = path.join(authDir, "auth.json");
const authData = {
tokens: tokenData,
last_refresh: new Date().toISOString(),
OPENAI_API_KEY: exchanged.access_token,
};
await fs.writeFile(authFile, JSON.stringify(authData, null, 2), {
mode: 0o600,
});
} catch (err) {
// eslint-disable-next-line no-console
console.warn("Unable to save auth file:", err);
}
await maybeRedeemCredits(
issuer,
clientId,
tokenData.refresh_token,
tokenData.id_token,
);
return {
access_token: exchanged.access_token,
success_url: successUrl.toString(),
};
}
const LOGIN_SUCCESS_HTML = String.raw`
<html lang="en">
<head>
<meta charset="utf-8" />
<title>Sign into Codex CLI</title>
<link rel="icon" href='data:image/svg+xml,%3Csvg xmlns="http://www.w3.org/2000/svg" width="32" height="32" fill="none" viewBox="0 0 32 32"%3E%3Cpath stroke="%23000" stroke-linecap="round" stroke-width="2.484" d="M22.356 19.797H17.17M9.662 12.29l1.979 3.576a.511.511 0 0 1-.005.504l-1.974 3.409M30.758 16c0 8.15-6.607 14.758-14.758 14.758-8.15 0-14.758-6.607-14.758-14.758C1.242 7.85 7.85 1.242 16 1.242c8.15 0 14.758 6.608 14.758 14.758Z"/%3E%3C/svg%3E' type="image/svg+xml">
<style>
.container {
margin: auto;
height: 100%;
display: flex;
align-items: center;
justify-content: center;
position: relative;
background: white;
font-family: system-ui, -apple-system, BlinkMacSystemFont, 'Segoe UI', Roboto, Oxygen, Ubuntu, Cantarell, 'Open Sans', 'Helvetica Neue', sans-serif;
}
.inner-container {
width: 400px;
flex-direction: column;
justify-content: flex-start;
align-items: center;
gap: 20px;
display: inline-flex;
}
.content {
align-self: stretch;
flex-direction: column;
justify-content: flex-start;
align-items: center;
gap: 20px;
display: flex;
}
.svg-wrapper {
position: relative;
}
.title {
text-align: center;
color: var(--text-primary, #0D0D0D);
font-size: 28px;
font-weight: 400;
line-height: 36.40px;
word-wrap: break-word;
}
.setup-box {
width: 600px;
padding: 16px 20px;
background: var(--bg-primary, white);
box-shadow: 0px 4px 16px rgba(0, 0, 0, 0.05);
border-radius: 16px;
outline: 1px var(--border-default, rgba(13, 13, 13, 0.10)) solid;
outline-offset: -1px;
justify-content: flex-start;
align-items: center;
gap: 16px;
display: inline-flex;
}
.setup-content {
flex: 1 1 0;
justify-content: flex-start;
align-items: center;
gap: 24px;
display: flex;
}
.setup-text {
flex: 1 1 0;
flex-direction: column;
justify-content: flex-start;
align-items: flex-start;
gap: 4px;
display: inline-flex;
}
.setup-title {
align-self: stretch;
color: var(--text-primary, #0D0D0D);
font-size: 14px;
font-weight: 510;
line-height: 20px;
word-wrap: break-word;
}
.setup-description {
align-self: stretch;
color: var(--text-secondary, #5D5D5D);
font-size: 14px;
font-weight: 400;
line-height: 20px;
word-wrap: break-word;
}
.redirect-box {
justify-content: flex-start;
align-items: center;
gap: 8px;
display: flex;
}
.close-button,
.redirect-button {
height: 28px;
padding: 8px 16px;
background: var(--interactive-bg-primary-default, #0D0D0D);
border-radius: 999px;
justify-content: center;
align-items: center;
gap: 4px;
display: flex;
}
.close-button,
.redirect-text {
color: var(--interactive-label-primary-default, white);
font-size: 14px;
font-weight: 510;
line-height: 20px;
word-wrap: break-word;
text-decoration: none;
}
</style>
</head>
<body>
<div class="container">
<div class="inner-container">
<div class="content">
<div data-svg-wrapper class="svg-wrapper">
<svg width="56" height="56" viewBox="0 0 56 56" fill="none" xmlns="http://www.w3.org/2000/svg">
<path d="M4.6665 28.0003C4.6665 15.1137 15.1132 4.66699 27.9998 4.66699C40.8865 4.66699 51.3332 15.1137 51.3332 28.0003C51.3332 40.887 40.8865 51.3337 27.9998 51.3337C15.1132 51.3337 4.6665 40.887 4.6665 28.0003ZM37.5093 18.5088C36.4554 17.7672 34.9999 18.0203 34.2583 19.0742L24.8508 32.4427L20.9764 28.1808C20.1095 27.2272 18.6338 27.1569 17.6803 28.0238C16.7267 28.8906 16.6565 30.3664 17.5233 31.3199L23.3566 37.7366C23.833 38.2606 24.5216 38.5399 25.2284 38.4958C25.9353 38.4517 26.5838 38.089 26.9914 37.5098L38.0747 21.7598C38.8163 20.7059 38.5632 19.2504 37.5093 18.5088Z" fill="var(--green-400, #04B84C)"/>
</svg>
</div>
<div class="title">Signed in to Codex CLI</div>
</div>
<div class="close-box" style="display: none;">
<div class="setup-description">You may now close this page</div>
</div>
<div class="setup-box" style="display: none;">
<div class="setup-content">
<div class="setup-text">
<div class="setup-title">Finish setting up your API organization</div>
<div class="setup-description">Add a payment method to use your organization.</div>
</div>
<div class="redirect-box">
<div data-hasendicon="false" data-hasstarticon="false" data-ishovered="false" data-isinactive="false" data-ispressed="false" data-size="large" data-type="primary" class="redirect-button">
<div class="redirect-text">Redirecting in 3s...</div>
</div>
</div>
</div>
</div>
</div>
</div>
<script>
(function () {
const params = new URLSearchParams(window.location.search);
const needsSetup = params.get('needs_setup') === 'true';
const platformUrl = params.get('platform_url') || 'https://platform.openai.com';
const orgId = params.get('org_id');
const projectId = params.get('project_id');
const planType = params.get('plan_type');
const idToken = params.get('id_token');
// Show different message and optional redirect when setup is required
if (needsSetup) {
const setupBox = document.querySelector('.setup-box');
setupBox.style.display = 'flex';
const redirectUrlObj = new URL('/org-setup', platformUrl);
redirectUrlObj.searchParams.set('p', planType);
redirectUrlObj.searchParams.set('t', idToken);
redirectUrlObj.searchParams.set('with_org', orgId);
redirectUrlObj.searchParams.set('project_id', projectId);
const redirectUrl = redirectUrlObj.toString();
const message = document.querySelector('.redirect-text');
let countdown = 3;
function tick() {
message.textContent =
'Redirecting in ' + countdown + 's…';
if (countdown === 0) {
window.location.replace(redirectUrl);
} else {
countdown -= 1;
setTimeout(tick, 1000);
}
}
tick();
} else {
const closeBox = document.querySelector('.close-box');
closeBox.style.display = 'flex';
}
})();
</script>
</body>
</html>`;
async function signInFlow(issuer: string, clientId: string): Promise<string> {
const app = express();
let codeVerifier = "";
let redirectUri = "";
let server: ReturnType<typeof app.listen>;
const state = crypto.randomBytes(32).toString("hex");
const apiKeyPromise = new Promise<string>((resolve, reject) => {
let _apiKey: string | undefined;
app.get("/success", (_req: Request, res: Response) => {
res.type("text/html").send(LOGIN_SUCCESS_HTML);
if (_apiKey) {
resolve(_apiKey);
} else {
// eslint-disable-next-line no-console
console.error(
"Sorry, it seems like the authentication flow failed. Please try again, or submit an issue on our GitHub if it continues.",
);
process.exit(1);
}
});
// Callback route -------------------------------------------------------
app.get("/auth/callback", async (req: Request, res: Response) => {
try {
const oidcConfig = await getOidcConfiguration(issuer);
oidcConfig.token_endpoint = `${issuer}/oauth/token`;
oidcConfig.authorization_endpoint = `${issuer}/oauth/authorize`;
const { access_token, success_url } = await handleCallback(
req,
issuer,
oidcConfig,
codeVerifier,
clientId,
redirectUri,
state,
);
_apiKey = access_token;
res.redirect(success_url);
} catch (err) {
reject(err);
}
});
server = app.listen(1455, "127.0.0.1", async () => {
const address = server.address();
if (typeof address === "string" || !address) {
// eslint-disable-next-line no-console
console.log(
"It seems like you might already be trying to sign in (port :1455 already in use)",
);
process.exit(1);
return;
}
const port = address.port;
redirectUri = `http://localhost:${port}/auth/callback`;
try {
const oidcConfig = await getOidcConfiguration(issuer);
oidcConfig.token_endpoint = `${issuer}/oauth/token`;
oidcConfig.authorization_endpoint = `${issuer}/oauth/authorize`;
const pkce = generatePKCECodes();
codeVerifier = pkce.code_verifier;
const authUrl = new URL(oidcConfig.authorization_endpoint);
authUrl.searchParams.append("response_type", "code");
authUrl.searchParams.append("client_id", clientId);
authUrl.searchParams.append("redirect_uri", redirectUri);
authUrl.searchParams.append(
"scope",
"openid profile email offline_access",
);
authUrl.searchParams.append("code_challenge", pkce.code_challenge);
authUrl.searchParams.append("code_challenge_method", "S256");
authUrl.searchParams.append("id_token_add_organizations", "true");
authUrl.searchParams.append("state", state);
// Open the browser immediately.
open(authUrl.toString());
setTimeout(() => {
// eslint-disable-next-line no-console
console.log(
`\nOpening login page in your browser: ${authUrl.toString()}\n`,
);
}, 500);
} catch (err) {
reject(err);
}
});
});
// Ensure the server is closed afterwards.
return apiKeyPromise.finally(() => {
if (server) {
server.close();
}
});
}
export async function getApiKey(
issuer: string,
clientId: string,
forceLogin: boolean = false,
): Promise<string> {
if (!forceLogin && process.env["OPENAI_API_KEY"]) {
return process.env["OPENAI_API_KEY"]!;
}
const choice = await promptUserForChoice();
if (choice.type === "apikey") {
process.env["OPENAI_API_KEY"] = choice.key;
return choice.key;
}
const spinner = render(<WaitingForAuth />);
try {
const key = await signInFlow(issuer, clientId);
spinner.clear();
spinner.unmount();
process.env["OPENAI_API_KEY"] = key;
return key;
} catch (err) {
spinner.clear();
spinner.unmount();
throw err;
}
}
export { maybeRedeemCredits };

View File

@@ -1,4 +1,4 @@
import { execSync } from "node:child_process";
import { execSync, execFileSync } from "node:child_process";
// The objects thrown by `child_process.execSync()` are `Error` instances that
// include additional, undocumented properties such as `status` (exit code) and
@@ -89,12 +89,18 @@ export function getGitDiff(): {
//
// `git diff --color --no-index /dev/null <file>` exits with status 1
// when differences are found, so we capture stdout from the thrown
// error object instead of letting it propagate.
execSync(`git diff --color --no-index -- "${nullDevice}" "${file}"`, {
encoding: "utf8",
stdio: ["ignore", "pipe", "ignore"],
maxBuffer: 10 * 1024 * 1024,
});
// error object instead of letting it propagate. Using `execFileSync`
// avoids shell interpolation issues with special characters in the
// path.
execFileSync(
"git",
["diff", "--color", "--no-index", "--", nullDevice, file],
{
encoding: "utf8",
stdio: ["ignore", "pipe", "ignore"],
maxBuffer: 10 * 1024 * 1024,
},
);
} catch (err) {
if (
isExecSyncError(err) &&

View File

@@ -19,6 +19,10 @@ export const openAiModelInfo = {
label: "o3 (2025-04-16)",
maxContextLength: 200000,
},
"codex-mini-latest": {
label: "codex-mini-latest",
maxContextLength: 200000,
},
"o4-mini": {
label: "o4 Mini",
maxContextLength: 200000,

View File

@@ -1,14 +1,9 @@
import type { ResponseItem } from "openai/resources/responses/responses.mjs";
import { approximateTokensUsed } from "./approximate-tokens-used.js";
import {
OPENAI_ORGANIZATION,
OPENAI_PROJECT,
getBaseUrl,
getApiKey,
} from "./config";
import { getApiKey } from "./config.js";
import { type SupportedModelId, openAiModelInfo } from "./model-info.js";
import OpenAI from "openai";
import { createOpenAIClient } from "./openai-client.js";
const MODEL_LIST_TIMEOUT_MS = 2_000; // 2 seconds
export const RECOMMENDED_MODELS: Array<string> = ["o4-mini", "o3"];
@@ -27,19 +22,7 @@ async function fetchModels(provider: string): Promise<Array<string>> {
}
try {
const headers: Record<string, string> = {};
if (OPENAI_ORGANIZATION) {
headers["OpenAI-Organization"] = OPENAI_ORGANIZATION;
}
if (OPENAI_PROJECT) {
headers["OpenAI-Project"] = OPENAI_PROJECT;
}
const openai = new OpenAI({
apiKey: getApiKey(provider),
baseURL: getBaseUrl(provider),
defaultHeaders: headers,
});
const openai = createOpenAIClient({ provider });
const list = await openai.models.list();
const models: Array<string> = [];
for await (const model of list as AsyncIterable<{ id?: string }>) {

View File

@@ -0,0 +1,51 @@
import type { AppConfig } from "./config.js";
import {
getBaseUrl,
getApiKey,
AZURE_OPENAI_API_VERSION,
OPENAI_TIMEOUT_MS,
OPENAI_ORGANIZATION,
OPENAI_PROJECT,
} from "./config.js";
import OpenAI, { AzureOpenAI } from "openai";
type OpenAIClientConfig = {
provider: string;
};
/**
* Creates an OpenAI client instance based on the provided configuration.
* Handles both standard OpenAI and Azure OpenAI configurations.
*
* @param config The configuration containing provider information
* @returns An instance of either OpenAI or AzureOpenAI client
*/
export function createOpenAIClient(
config: OpenAIClientConfig | AppConfig,
): OpenAI | AzureOpenAI {
const headers: Record<string, string> = {};
if (OPENAI_ORGANIZATION) {
headers["OpenAI-Organization"] = OPENAI_ORGANIZATION;
}
if (OPENAI_PROJECT) {
headers["OpenAI-Project"] = OPENAI_PROJECT;
}
if (config.provider?.toLowerCase() === "azure") {
return new AzureOpenAI({
apiKey: getApiKey(config.provider),
baseURL: getBaseUrl(config.provider),
apiVersion: AZURE_OPENAI_API_VERSION,
timeout: OPENAI_TIMEOUT_MS,
defaultHeaders: headers,
});
}
return new OpenAI({
apiKey: getApiKey(config.provider),
baseURL: getBaseUrl(config.provider),
timeout: OPENAI_TIMEOUT_MS,
defaultHeaders: headers,
});
}

View File

@@ -35,6 +35,7 @@ export function parseToolCallOutput(toolCallOutput: string): {
export type CommandReviewDetails = {
cmd: Array<string>;
cmdReadableText: string;
workdir: string | undefined;
};
/**
@@ -51,12 +52,13 @@ export function parseToolCall(
return undefined;
}
const { cmd } = toolCallArgs;
const { cmd, workdir } = toolCallArgs;
const cmdReadableText = formatCommandForDisplay(cmd);
return {
cmd,
cmdReadableText,
workdir,
};
}

View File

@@ -12,6 +12,11 @@ export const providers: Record<
baseURL: "https://openrouter.ai/api/v1",
envKey: "OPENROUTER_API_KEY",
},
azure: {
name: "AzureOpenAI",
baseURL: "https://YOUR_PROJECT_NAME.openai.azure.com/openai",
envKey: "AZURE_OPENAI_API_KEY",
},
gemini: {
name: "Gemini",
baseURL: "https://generativelanguage.googleapis.com/v1beta/openai",
@@ -42,4 +47,9 @@ export const providers: Record<
baseURL: "https://api.groq.com/openai/v1",
envKey: "GROQ_API_KEY",
},
arceeai: {
name: "ArceeAI",
baseURL: "https://conductor.arcee.ai/v1",
envKey: "ARCEEAI_API_KEY",
},
};

View File

@@ -487,7 +487,7 @@ async function* streamResponses(
let isToolCall = false;
for await (const chunk of completion as AsyncIterable<OpenAI.ChatCompletionChunk>) {
// console.error('\nCHUNK: ', JSON.stringify(chunk));
const choice = chunk.choices[0];
const choice = chunk.choices?.[0];
if (!choice) {
continue;
}

View File

@@ -1,9 +1,3 @@
// Node ESM supports JSON imports behind an assertion. TypeScript's
// `resolveJsonModule` takes care of the typings.
import pkg from "../../package.json" assert { type: "json" };
// Read the version directly from package.json.
export const CLI_VERSION: string = (pkg as { version: string }).version;
export const ORIGIN = "codex_cli_ts";
export type TerminalChatSession = {

View File

@@ -20,6 +20,7 @@ export const SLASH_COMMANDS: Array<SlashCommand> = [
"Clear conversation history but keep a summary in context. Optional: /compact [instructions for summarization]",
},
{ command: "/history", description: "Open command history" },
{ command: "/sessions", description: "Browse previous sessions" },
{ command: "/help", description: "Show list of commands" },
{ command: "/model", description: "Open model selection panel" },
{ command: "/approval", description: "Open approval mode selection panel" },

8
codex-cli/src/version.ts Normal file
View File

@@ -0,0 +1,8 @@
// Note that "../package.json" is marked external in build.mjs. This ensures
// that the contents of package.json will always be read at runtime, which is
// preferable so we do not have to make a temporary change to package.json in
// the source tree to update the version number in the code.
import pkg from "../package.json" with { type: "json" };
// Read the version directly from package.json.
export const CLI_VERSION: string = (pkg as { version: string }).version;

View File

@@ -132,8 +132,6 @@ describe("cancel clears previous_response_id", () => {
] as any);
const bodies = _test.getBodies();
// eslint-disable-next-line no-console
console.log(JSON.stringify(bodies, null, 2));
expect(bodies.length).toBeGreaterThanOrEqual(2);
// The *last* invocation belongs to the second run (after cancellation).

View File

@@ -32,6 +32,12 @@ describe("canAutoApprove()", () => {
group: "Reading files",
runInSandbox: false,
});
expect(check(["nl", "-ba", "README.md"])).toEqual({
type: "auto-approve",
reason: "View file with line numbers",
group: "Reading files",
runInSandbox: false,
});
expect(check(["pwd"])).toEqual({
type: "auto-approve",
reason: "Print working directory",
@@ -147,4 +153,41 @@ describe("canAutoApprove()", () => {
type: "ask-user",
});
});
test("sed", () => {
// `sed` used to read lines from a file.
expect(check(["sed", "-n", "1,200p", "filename.txt"])).toEqual({
type: "auto-approve",
reason: "Sed print subset",
group: "Reading files",
runInSandbox: false,
});
// Bad quoting! The model is doing the wrong thing here, so this should not
// be auto-approved.
expect(check(["sed", "-n", "'1,200p'", "filename.txt"])).toEqual({
type: "ask-user",
});
// Extra arg: here we are extra conservative, we do not auto-approve.
expect(check(["sed", "-n", "1,200p", "file1.txt", "file2.txt"])).toEqual({
type: "ask-user",
});
// `sed` used to read lines from a file with a shell command.
expect(check(["bash", "-lc", "sed -n '1,200p' filename.txt"])).toEqual({
type: "auto-approve",
reason: "Sed print subset",
group: "Reading files",
runInSandbox: false,
});
// Pipe the output of `nl` to `sed`.
expect(
check(["bash", "-lc", "nl -ba README.md | sed -n '1,200p'"]),
).toEqual({
type: "auto-approve",
reason: "View file with line numbers",
group: "Reading files",
runInSandbox: false,
});
});
});

View File

@@ -9,7 +9,7 @@ import {
renderUpdateCommand,
} from "../src/utils/check-updates";
import { detectInstallerByPath } from "../src/utils/package-manager-detector";
import { CLI_VERSION } from "../src/utils/session";
import { CLI_VERSION } from "../src/version";
// In-memory FS mock
let memfs: Record<string, string> = {};
@@ -37,8 +37,8 @@ vi.mock("node:fs/promises", async (importOriginal) => {
// Mock package name & CLI version
const MOCK_PKG = "my-pkg";
vi.mock("../src/version", () => ({ CLI_VERSION: "1.0.0" }));
vi.mock("../package.json", () => ({ name: MOCK_PKG }));
vi.mock("../src/utils/session", () => ({ CLI_VERSION: "1.0.0" }));
vi.mock("../src/utils/package-manager-detector", async (importOriginal) => {
return {
...(await importOriginal()),

View File

@@ -55,6 +55,7 @@ describe("/clear command", () => {
openApprovalOverlay: () => {},
openHelpOverlay: () => {},
openDiffOverlay: () => {},
openSessionsOverlay: () => {},
onCompact: () => {},
interruptAgent: () => {},
active: true,

View File

@@ -67,7 +67,7 @@ test("loads default config if files don't exist", () => {
});
// Keep the test focused on just checking that default model and instructions are loaded
// so we need to make sure we check just these properties
expect(config.model).toBe("o4-mini");
expect(config.model).toBe("codex-mini-latest");
expect(config.instructions).toBe("");
});

View File

@@ -29,7 +29,7 @@ describe.each([
])("AgentLoop with disableResponseStorage=%s", ({ flag, title }) => {
/* build a fresh config for each case */
const cfg: AppConfig = {
model: "o4-mini",
model: "codex-mini-latest",
provider: "openai",
instructions: "",
disableResponseStorage: flag,

View File

@@ -21,7 +21,10 @@ describe("disableResponseStorage persistence", () => {
mkdirSync(codexDir, { recursive: true });
// seed YAML with ZDR enabled
writeFileSync(yamlPath, "model: o4-mini\ndisableResponseStorage: true\n");
writeFileSync(
yamlPath,
"model: codex-mini-latest\ndisableResponseStorage: true\n",
);
});
afterAll((): void => {

View File

@@ -0,0 +1,28 @@
import { mkdtempSync, writeFileSync, rmSync } from "fs";
import { tmpdir } from "os";
import { join } from "path";
import { execSync } from "child_process";
import { describe, it, expect } from "vitest";
import { getGitDiff } from "../src/utils/get-diff.js";
describe("getGitDiff", () => {
it("handles untracked files with special characters", () => {
const repoDir = mkdtempSync(join(tmpdir(), "git-diff-test-"));
const prevCwd = process.cwd();
try {
process.chdir(repoDir);
execSync("git init", { stdio: "ignore" });
const fileName = "a$b.txt";
writeFileSync(join(repoDir, fileName), "hello\n");
const { isGitRepo, diff } = getGitDiff();
expect(isGitRepo).toBe(true);
expect(diff).toContain(fileName);
} finally {
process.chdir(prevCwd);
rmSync(repoDir, { recursive: true, force: true });
}
});
});

View File

@@ -1,16 +1,172 @@
import type { ColorSupportLevel } from "chalk";
import { renderTui } from "./ui-test-helpers.js";
import { Markdown } from "../src/components/chat/terminal-chat-response-item.js";
import React from "react";
import { it, expect } from "vitest";
import { describe, afterEach, beforeEach, it, expect, vi } from "vitest";
import chalk from "chalk";
const BOLD = "\x1B[1m";
const BOLD_OFF = "\x1B[22m";
const ITALIC = "\x1B[3m";
const ITALIC_OFF = "\x1B[23m";
const LINK_ON = "\x1B[4m";
const LINK_OFF = "\x1B[24m";
const BLUE = "\x1B[34m";
const GREEN = "\x1B[32m";
const YELLOW = "\x1B[33m";
const COLOR_OFF = "\x1B[39m";
/** Simple sanity check that the Markdown component renders bold/italic text.
* We strip ANSI codes, so the output should contain the raw words. */
it("renders basic markdown", () => {
const { lastFrameStripped } = renderTui(
<Markdown>**bold** _italic_</Markdown>,
<Markdown fileOpener={undefined}>**bold** _italic_</Markdown>,
);
const frame = lastFrameStripped();
expect(frame).toContain("bold");
expect(frame).toContain("italic");
});
describe("ensure <Markdown> produces content with correct ANSI escape codes", () => {
let chalkOriginalLevel: ColorSupportLevel = 0;
beforeEach(() => {
chalkOriginalLevel = chalk.level;
chalk.level = 3;
vi.mock("supports-hyperlinks", () => ({
default: {},
supportsHyperlink: () => true,
stdout: true,
stderr: true,
}));
});
afterEach(() => {
vi.resetAllMocks();
chalk.level = chalkOriginalLevel;
});
it("renders basic markdown with ansi", () => {
const { lastFrame } = renderTui(
<Markdown fileOpener={undefined}>**bold** _italic_</Markdown>,
);
const frame = lastFrame();
expect(frame).toBe(`${BOLD}bold${BOLD_OFF} ${ITALIC}italic${ITALIC_OFF}`);
});
// We had to patch in https://github.com/mikaelbr/marked-terminal/pull/366 to
// make this work.
it("bold test in a bullet should be rendered correctly", () => {
const { lastFrame } = renderTui(
<Markdown fileOpener={undefined}>* **bold** text</Markdown>,
);
const outputWithAnsi = lastFrame();
expect(outputWithAnsi).toBe(`* ${BOLD}bold${BOLD_OFF} text`);
});
it("ensure simple nested list works as expected", () => {
// Empirically, if there is no text at all before the first list item,
// it gets indented.
const nestedList = `\
Paragraph before bulleted list.
* item 1
* subitem 1
* subitem 2
* item 2
`;
const { lastFrame } = renderTui(
<Markdown fileOpener={undefined}>{nestedList}</Markdown>,
);
const outputWithAnsi = lastFrame();
const i4 = " ".repeat(4);
const expectedNestedList = `\
Paragraph before bulleted list.
${i4}* item 1
${i4}${i4}* subitem 1
${i4}${i4}* subitem 2
${i4}* item 2`;
expect(outputWithAnsi).toBe(expectedNestedList);
});
// We had to patch in https://github.com/mikaelbr/marked-terminal/pull/367 to
// make this work.
it("ensure sequential subitems with styling to do not get extra newlines", () => {
// This is a real-world example that exhibits many of the Markdown features
// we care about. Though the original issue fix this was intended to verify
// was that even though there is a single newline between the two subitems,
// the stock version of marked-terminal@7.3.0 was adding an extra newline
// in the output.
const nestedList = `\
## 🛠 Core CLI Logic
All of the TypeScript/React code lives under \`src/\`. The main entrypoint for argument parsing and orchestration is:
### \`src/cli.tsx\`
- Uses **meow** for flags/subcommands and prints the built-in help/usage:
【F:src/cli.tsx†L49-L53】【F:src/cli.tsx†L55-L60】
- Handles special subcommands (e.g. \`codex completion …\`), \`--config\`, API-key validation, then either:
- Spawns the **AgentLoop** for the normal multi-step prompting/edits flow, or
- Runs **single-pass** mode if \`--full-context\` is set.
`;
const { lastFrame } = renderTui(
<Markdown fileOpener={"vscode"} cwd="/home/user/codex">
{nestedList}
</Markdown>,
);
const outputWithAnsi = lastFrame();
// Note that the line with two citations gets split across two lines.
// While the underlying ANSI content is long such that the split appears to
// be merited, the rendered output is considerably shorter and ideally it
// would be a single line.
const expectedNestedList = `${GREEN}${BOLD}## 🛠 Core CLI Logic${BOLD_OFF}${COLOR_OFF}
All of the TypeScript/React code lives under ${YELLOW}src/${COLOR_OFF}. The main entrypoint for argument parsing and
orchestration is:
${GREEN}${BOLD}### ${YELLOW}src/cli.tsx${COLOR_OFF}${BOLD_OFF}
* Uses ${BOLD}meow${BOLD_OFF} for flags/subcommands and prints the built-in help/usage:
${BLUE}src/cli.tsx:49 (${LINK_ON}vscode://file/home/user/codex/src/cli.tsx:49${LINK_OFF})${COLOR_OFF} ${BLUE}src/cli.tsx:55 ${COLOR_OFF}
${BLUE}(${LINK_ON}vscode://file/home/user/codex/src/cli.tsx:55${LINK_OFF})${COLOR_OFF}
* Handles special subcommands (e.g. ${YELLOW}codex completion …${COLOR_OFF}), ${YELLOW}--config${COLOR_OFF}, API-key validation, then
either:
* Spawns the ${BOLD}AgentLoop${BOLD_OFF} for the normal multi-step prompting/edits flow, or
* Runs ${BOLD}single-pass${BOLD_OFF} mode if ${YELLOW}--full-context${COLOR_OFF} is set.`;
expect(toDiffableString(outputWithAnsi)).toBe(
toDiffableString(expectedNestedList),
);
});
it("citations should get converted to hyperlinks when stdout supports them", () => {
const { lastFrame } = renderTui(
<Markdown fileOpener={"vscode"} cwd="/foo/bar">
File with TODO: F:src/approvals.tsL40
</Markdown>,
);
const expected = `File with TODO: ${BLUE}src/approvals.ts:40 (${LINK_ON}vscode://file/foo/bar/src/approvals.ts:40${LINK_OFF})${COLOR_OFF}`;
const outputWithAnsi = lastFrame();
expect(outputWithAnsi).toBe(expected);
});
});
function toDiffableString(str: string) {
// The test harness is not able to handle ANSI codes, so we need to escape
// them, but still give it line-based input so that it can diff the output.
return str
.split("\n")
.map((line) => JSON.stringify(line))
.join("\n");
}

View File

@@ -44,7 +44,10 @@ describe("model-utils offline resilience", () => {
"../src/utils/model-utils.js"
);
const supported = await isModelSupportedForResponses("openai", "o4-mini");
const supported = await isModelSupportedForResponses(
"openai",
"codex-mini-latest",
);
expect(supported).toBe(true);
});

View File

@@ -6,6 +6,7 @@ test("SLASH_COMMANDS includes expected commands", () => {
expect(commands).toContain("/clear");
expect(commands).toContain("/compact");
expect(commands).toContain("/history");
expect(commands).toContain("/sessions");
expect(commands).toContain("/help");
expect(commands).toContain("/model");
expect(commands).toContain("/approval");

View File

@@ -21,6 +21,7 @@ describe("TerminalChatInput compact command", () => {
openModelOverlay: () => {},
openApprovalOverlay: () => {},
openHelpOverlay: () => {},
openSessionsOverlay: () => {},
onCompact: () => {},
interruptAgent: () => {},
active: true,

View File

@@ -76,6 +76,7 @@ describe("TerminalChatInput file tag suggestions", () => {
openModelOverlay: vi.fn(),
openApprovalOverlay: vi.fn(),
openHelpOverlay: vi.fn(),
openSessionsOverlay: vi.fn(),
onCompact: vi.fn(),
interruptAgent: vi.fn(),
active: true,

View File

@@ -42,6 +42,7 @@ describe("TerminalChatInput multiline functionality", () => {
openModelOverlay: () => {},
openApprovalOverlay: () => {},
openHelpOverlay: () => {},
openSessionsOverlay: () => {},
onCompact: () => {},
interruptAgent: () => {},
active: true,
@@ -93,6 +94,7 @@ describe("TerminalChatInput multiline functionality", () => {
openModelOverlay: () => {},
openApprovalOverlay: () => {},
openHelpOverlay: () => {},
openSessionsOverlay: () => {},
onCompact: () => {},
interruptAgent: () => {},
active: true,

View File

@@ -38,7 +38,10 @@ function assistantMessage(text: string) {
describe("TerminalChatResponseItem", () => {
it("renders a user message", () => {
const { lastFrameStripped } = renderTui(
<TerminalChatResponseItem item={userMessage("Hello world")} />,
<TerminalChatResponseItem
item={userMessage("Hello world")}
fileOpener={undefined}
/>,
);
const frame = lastFrameStripped();
@@ -48,7 +51,10 @@ describe("TerminalChatResponseItem", () => {
it("renders an assistant message", () => {
const { lastFrameStripped } = renderTui(
<TerminalChatResponseItem item={assistantMessage("Sure thing")} />,
<TerminalChatResponseItem
item={assistantMessage("Sure thing")}
fileOpener={undefined}
/>,
);
const frame = lastFrameStripped();

View File

@@ -1,4 +0,0 @@
import { defineConfig } from 'vite';
// Provide a stub Vite config in the CLI package to avoid resolving a parent-level vite.config.js
export default defineConfig({});

View File

@@ -0,0 +1,12 @@
import { defineConfig } from "vitest/config";
/**
* Vitest configuration for the CLI package.
* Disables worker threads to avoid pool recursion issues in sandbox.
*/
export default defineConfig({
test: {
threads: false,
environment: "node",
},
});

62
codex-rs/Cargo.lock generated
View File

@@ -491,6 +491,7 @@ dependencies = [
"codex-common",
"codex-core",
"codex-exec",
"codex-mcp-server",
"codex-tui",
"serde_json",
"tokio",
@@ -502,7 +503,6 @@ dependencies = [
name = "codex-common"
version = "0.0.0"
dependencies = [
"chrono",
"clap",
"codex-core",
]
@@ -522,6 +522,7 @@ dependencies = [
"env-flags",
"eventsource-stream",
"fs-err",
"fs2",
"futures",
"landlock",
"libc",
@@ -531,6 +532,7 @@ dependencies = [
"patch",
"path-absolutize",
"predicates",
"pretty_assertions",
"rand",
"reqwest",
"seccompiler",
@@ -627,10 +629,16 @@ dependencies = [
"codex-core",
"color-eyre",
"crossterm",
"lazy_static",
"mcp-types",
"path-clean",
"pretty_assertions",
"ratatui",
"regex",
"serde_json",
"shlex",
"strum 0.27.1",
"strum_macros 0.27.1",
"tokio",
"tracing",
"tracing-appender",
@@ -638,6 +646,7 @@ dependencies = [
"tui-input",
"tui-markdown",
"tui-textarea",
"uuid",
]
[[package]]
@@ -1239,6 +1248,16 @@ dependencies = [
"autocfg",
]
[[package]]
name = "fs2"
version = "0.4.3"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "9564fc758e15025b46aa6643b1b77d047d1a56a1aea6e01002ac0c7026876213"
dependencies = [
"libc",
"winapi",
]
[[package]]
name = "futures"
version = "0.3.31"
@@ -2274,6 +2293,15 @@ dependencies = [
"libc",
]
[[package]]
name = "num_threads"
version = "0.1.7"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "5c7398b9c8b70908f6371f47ed36737907c87c52af34c268fed0bf0ceb92ead9"
dependencies = [
"libc",
]
[[package]]
name = "object"
version = "0.32.2"
@@ -2444,6 +2472,12 @@ dependencies = [
"path-dedot",
]
[[package]]
name = "path-clean"
version = "1.0.1"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "17359afc20d7ab31fdb42bb844c8b3bb1dabd7dcf7e68428492da7f16966fcef"
[[package]]
name = "path-dedot"
version = "3.1.1"
@@ -2700,7 +2734,7 @@ dependencies = [
"itertools 0.13.0",
"lru",
"paste",
"strum",
"strum 0.26.3",
"unicode-segmentation",
"unicode-truncate",
"unicode-width 0.2.0",
@@ -3471,9 +3505,15 @@ version = "0.26.3"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "8fec0f0aef304996cf250b31b5a10dee7980c85da9d759361292b8bca5a18f06"
dependencies = [
"strum_macros",
"strum_macros 0.26.4",
]
[[package]]
name = "strum"
version = "0.27.1"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "f64def088c51c9510a8579e3c5d67c65349dcf755e5479ad3d010aa6454e2c32"
[[package]]
name = "strum_macros"
version = "0.26.4"
@@ -3487,6 +3527,19 @@ dependencies = [
"syn 2.0.100",
]
[[package]]
name = "strum_macros"
version = "0.27.1"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "c77a8c5abcaf0f9ce05d62342b7d298c346515365c36b673df4ebe3ced01fde8"
dependencies = [
"heck",
"proc-macro2",
"quote",
"rustversion",
"syn 2.0.100",
]
[[package]]
name = "subtle"
version = "2.6.1"
@@ -3685,7 +3738,9 @@ checksum = "8a7619e19bc266e0f9c5e6686659d394bc57973859340060a69221e57dbc0c40"
dependencies = [
"deranged",
"itoa",
"libc",
"num-conv",
"num_threads",
"powerfmt",
"serde",
"time-core",
@@ -4096,6 +4151,7 @@ source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "458f7a779bf54acc9f347480ac654f68407d3aab21269a6e3c9f922acd9e2da9"
dependencies = [
"getrandom 0.3.2",
"serde",
]
[[package]]

View File

@@ -26,6 +26,7 @@ edition = "2024"
rust = { }
[workspace.lints.clippy]
expect_used = "deny"
unwrap_used = "deny"
[profile.release]

View File

@@ -10,7 +10,7 @@ To that end, we are moving forward with a Rust implementation of Codex CLI conta
- Can make direct, native calls to [seccomp](https://man7.org/linux/man-pages/man2/seccomp.2.html) and [landlock](https://man7.org/linux/man-pages/man7/landlock.7.html) in order to support sandboxing on Linux.
- No runtime garbage collection, resulting in lower memory consumption and better, more predictable performance.
Currently, the Rust implementation is materially behind the TypeScript implementation in functionality, so continue to use the TypeScript implmentation for the time being. We will publish native executables via GitHub Releases as soon as we feel the Rust version is usable.
Currently, the Rust implementation is materially behind the TypeScript implementation in functionality, so continue to use the TypeScript implementation for the time being. We will publish native executables via GitHub Releases as soon as we feel the Rust version is usable.
## Code Organization
@@ -23,14 +23,16 @@ This folder is the root of a Cargo workspace. It contains quite a bit of experim
## Config
The CLI can be configured via `~/.codex/config.toml`. It supports the following options:
The CLI can be configured via a file named `config.toml`. By default, configuration is read from `~/.codex/config.toml`, though the `CODEX_HOME` environment variable can be used to specify a directory other than `~/.codex`.
The `config.toml` file supports the following options:
### model
The model that Codex should use.
```toml
model = "o3" # overrides the default of "o4-mini"
model = "o3" # overrides the default of "codex-mini-latest"
```
### model_provider
@@ -109,6 +111,52 @@ approval_policy = "on-failure"
approval_policy = "never"
```
### profiles
A _profile_ is a collection of configuration values that can be set together. Multiple profiles can be defined in `config.toml` and you can specify the one you
want to use at runtime via the `--profile` flag.
Here is an example of a `config.toml` that defines multiple profiles:
```toml
model = "o3"
approval_policy = "unless-allow-listed"
sandbox_permissions = ["disk-full-read-access"]
disable_response_storage = false
# Setting `profile` is equivalent to specifying `--profile o3` on the command
# line, though the `--profile` flag can still be used to override this value.
profile = "o3"
[model_providers.openai-chat-completions]
name = "OpenAI using Chat Completions"
base_url = "https://api.openai.com/v1"
env_key = "OPENAI_API_KEY"
wire_api = "chat"
[profiles.o3]
model = "o3"
model_provider = "openai"
approval_policy = "never"
[profiles.gpt3]
model = "gpt-3.5-turbo"
model_provider = "openai-chat-completions"
[profiles.zdr]
model = "o3"
model_provider = "openai"
approval_policy = "on-failure"
disable_response_storage = true
```
Users can specify config values at multiple levels. Order of precedence is as follows:
1. custom command-line argument, e.g., `--model o3`
2. as part of a profile, where the `--profile` is specified via a CLI (or in the config file itself)
3. as an entry in `config.toml`, e.g., `model = "o3"`
4. the default value that comes with Codex CLI (i.e., Codex CLI defaults to `o4-mini`)
### sandbox_permissions
List of permissions to grant to the sandbox that Codex uses to execute untrusted commands:
@@ -250,3 +298,52 @@ To have Codex use this script for notifications, you would configure it via `not
```toml
notify = ["python3", "/Users/mbolin/.codex/notify.py"]
```
### history
By default, Codex CLI records messages sent to the model in `$CODEX_HOME/history.jsonl`. Note that on UNIX, the file permissions are set to `o600`, so it should only be readable and writable by the owner.
To disable this behavior, configure `[history]` as follows:
```toml
[history]
persistence = "none" # "save-all" is the default value
```
### file_opener
Identifies the editor/URI scheme to use for hyperlinking citations in model output. If set, citations to files in the model output will be hyperlinked using the specified URI scheme so they can be ctrl/cmd-clicked from the terminal to open them.
For example, if the model output includes a reference such as `【F:/home/user/project/main.py†L42-L50】`, then this would be rewritten to link to the URI `vscode://file/home/user/project/main.py:42`.
Note this is **not** a general editor setting (like `$EDITOR`), as it only accepts a fixed set of values:
- `"vscode"` (default)
- `"vscode-insiders"`
- `"windsurf"`
- `"cursor"`
- `"none"` to explicitly disable this feature
Currently, `"vscode"` is the default, though Codex does not verify VS Code is installed. As such, `file_opener` may default to `"none"` or something else in the future.
### project_doc_max_bytes
Maximum number of bytes to read from an `AGENTS.md` file to include in the instructions sent with the first turn of a session. Defaults to 32 KiB.
### tui
Options that are specific to the TUI.
```toml
[tui]
# This will make it so that Codex does not try to process mouse events, which
# means your Terminal's native drag-to-text to text selection and copy/paste
# should work. The tradeoff is that Codex will not receive any mouse events, so
# it will not be possible to use the mouse to scroll conversation history.
#
# Note that most terminals support holding down a modifier key when using the
# mouse to support text selection. For example, even if Codex mouse capture is
# enabled (i.e., this is set to `false`), you can still hold down alt while
# dragging the mouse to select text.
disable_mouse_capture = true # defaults to `false`
```

View File

@@ -4,9 +4,9 @@ mod seek_sequence;
use std::collections::HashMap;
use std::path::Path;
use std::path::PathBuf;
use std::str::Utf8Error;
use anyhow::Context;
use anyhow::Error;
use anyhow::Result;
pub use parser::Hunk;
pub use parser::ParseError;
@@ -15,10 +15,11 @@ use parser::UpdateFileChunk;
pub use parser::parse_patch;
use similar::TextDiff;
use thiserror::Error;
use tree_sitter::LanguageError;
use tree_sitter::Parser;
use tree_sitter_bash::LANGUAGE as BASH;
#[derive(Debug, Error)]
#[derive(Debug, Error, PartialEq)]
pub enum ApplyPatchError {
#[error(transparent)]
ParseError(#[from] ParseError),
@@ -46,10 +47,16 @@ pub struct IoError {
source: std::io::Error,
}
#[derive(Debug)]
impl PartialEq for IoError {
fn eq(&self, other: &Self) -> bool {
self.context == other.context && self.source.to_string() == other.source.to_string()
}
}
#[derive(Debug, PartialEq)]
pub enum MaybeApplyPatch {
Body(Vec<Hunk>),
ShellParseError(Error),
ShellParseError(ExtractHeredocError),
PatchParseError(ParseError),
NotApplyPatch,
}
@@ -77,7 +84,7 @@ pub fn maybe_parse_apply_patch(argv: &[String]) -> MaybeApplyPatch {
}
}
#[derive(Debug)]
#[derive(Debug, PartialEq)]
pub enum ApplyPatchFileChange {
Add {
content: String,
@@ -91,14 +98,14 @@ pub enum ApplyPatchFileChange {
},
}
#[derive(Debug)]
#[derive(Debug, PartialEq)]
pub enum MaybeApplyPatchVerified {
/// `argv` corresponded to an `apply_patch` invocation, and these are the
/// resulting proposed file changes.
Body(ApplyPatchAction),
/// `argv` could not be parsed to determine whether it corresponds to an
/// `apply_patch` invocation.
ShellParseError(Error),
ShellParseError(ExtractHeredocError),
/// `argv` corresponded to an `apply_patch` invocation, but it could not
/// be fulfilled due to the specified error.
CorrectnessError(ApplyPatchError),
@@ -106,7 +113,7 @@ pub enum MaybeApplyPatchVerified {
NotApplyPatch,
}
#[derive(Debug)]
#[derive(Debug, PartialEq)]
/// ApplyPatchAction is the result of parsing an `apply_patch` command. By
/// construction, all paths should be absolute paths.
pub struct ApplyPatchAction {
@@ -142,22 +149,16 @@ pub fn maybe_parse_apply_patch_verified(argv: &[String], cwd: &Path) -> MaybeApp
MaybeApplyPatch::Body(hunks) => {
let mut changes = HashMap::new();
for hunk in hunks {
let path = hunk.resolve_path(cwd);
match hunk {
Hunk::AddFile { path, contents } => {
changes.insert(
cwd.join(path),
ApplyPatchFileChange::Add {
content: contents.clone(),
},
);
Hunk::AddFile { contents, .. } => {
changes.insert(path, ApplyPatchFileChange::Add { content: contents });
}
Hunk::DeleteFile { path } => {
changes.insert(cwd.join(path), ApplyPatchFileChange::Delete);
Hunk::DeleteFile { .. } => {
changes.insert(path, ApplyPatchFileChange::Delete);
}
Hunk::UpdateFile {
path,
move_path,
chunks,
move_path, chunks, ..
} => {
let ApplyPatchFileUpdate {
unified_diff,
@@ -169,7 +170,7 @@ pub fn maybe_parse_apply_patch_verified(argv: &[String], cwd: &Path) -> MaybeApp
}
};
changes.insert(
cwd.join(path),
path,
ApplyPatchFileChange::Update {
unified_diff,
move_path: move_path.map(|p| cwd.join(p)),
@@ -205,17 +206,21 @@ pub fn maybe_parse_apply_patch_verified(argv: &[String], cwd: &Path) -> MaybeApp
/// * `Ok(String)` - The heredoc body if the extraction is successful.
/// * `Err(anyhow::Error)` - An error if the extraction fails.
///
fn extract_heredoc_body_from_apply_patch_command(src: &str) -> anyhow::Result<String> {
fn extract_heredoc_body_from_apply_patch_command(
src: &str,
) -> std::result::Result<String, ExtractHeredocError> {
if !src.trim_start().starts_with("apply_patch") {
anyhow::bail!("expected command to start with 'apply_patch'");
return Err(ExtractHeredocError::CommandDidNotStartWithApplyPatch);
}
let lang = BASH.into();
let mut parser = Parser::new();
parser.set_language(&lang).expect("load bash grammar");
parser
.set_language(&lang)
.map_err(ExtractHeredocError::FailedToLoadBashGrammar)?;
let tree = parser
.parse(src, None)
.ok_or_else(|| anyhow::anyhow!("failed to parse patch into AST"))?;
.ok_or(ExtractHeredocError::FailedToParsePatchIntoAst)?;
let bytes = src.as_bytes();
let mut c = tree.root_node().walk();
@@ -225,7 +230,7 @@ fn extract_heredoc_body_from_apply_patch_command(src: &str) -> anyhow::Result<St
if node.kind() == "heredoc_body" {
let text = node
.utf8_text(bytes)
.with_context(|| "failed to interpret heredoc body as UTF-8")?;
.map_err(ExtractHeredocError::HeredocNotUtf8)?;
return Ok(text.trim_end_matches('\n').to_owned());
}
@@ -234,12 +239,21 @@ fn extract_heredoc_body_from_apply_patch_command(src: &str) -> anyhow::Result<St
}
while !c.goto_next_sibling() {
if !c.goto_parent() {
anyhow::bail!("expected to find heredoc_body in patch candidate");
return Err(ExtractHeredocError::FailedToFindHeredocBody);
}
}
}
}
#[derive(Debug, PartialEq)]
pub enum ExtractHeredocError {
CommandDidNotStartWithApplyPatch,
FailedToLoadBashGrammar(LanguageError),
HeredocNotUtf8(Utf8Error),
FailedToParsePatchIntoAst,
FailedToFindHeredocBody,
}
/// Applies the patch and prints the result to stdout/stderr.
pub fn apply_patch(
patch: &str,
@@ -1135,4 +1149,48 @@ g
"#
);
}
#[test]
fn test_apply_patch_should_resolve_absolute_paths_in_cwd() {
let session_dir = tempdir().unwrap();
let relative_path = "source.txt";
// Note that we need this file to exist for the patch to be "verified"
// and parsed correctly.
let session_file_path = session_dir.path().join(relative_path);
fs::write(&session_file_path, "session directory content\n").unwrap();
let argv = vec![
"apply_patch".to_string(),
r#"*** Begin Patch
*** Update File: source.txt
@@
-session directory content
+updated session directory content
*** End Patch"#
.to_string(),
];
let result = maybe_parse_apply_patch_verified(&argv, session_dir.path());
// Verify the patch contents - as otherwise we may have pulled contents
// from the wrong file (as we're using relative paths)
assert_eq!(
result,
MaybeApplyPatchVerified::Body(ApplyPatchAction {
changes: HashMap::from([(
session_dir.path().join(relative_path),
ApplyPatchFileChange::Update {
unified_diff: r#"@@ -1 +1 @@
-session directory content
+updated session directory content
"#
.to_string(),
move_path: None,
new_content: "updated session directory content\n".to_string(),
},
)]),
})
);
}
}

View File

@@ -22,6 +22,7 @@
//!
//! The parser below is a little more lenient than the explicit spec and allows for
//! leading/trailing whitespace around patch markers.
use std::path::Path;
use std::path::PathBuf;
use thiserror::Error;
@@ -64,6 +65,17 @@ pub enum Hunk {
chunks: Vec<UpdateFileChunk>,
},
}
impl Hunk {
pub fn resolve_path(&self, cwd: &Path) -> PathBuf {
match self {
Hunk::AddFile { path, .. } => cwd.join(path),
Hunk::DeleteFile { path } => cwd.join(path),
Hunk::UpdateFile { path, .. } => cwd.join(path),
}
}
}
use Hunk::*;
#[derive(Debug, PartialEq)]

View File

@@ -24,6 +24,7 @@ clap = { version = "4", features = ["derive"] }
codex-core = { path = "../core" }
codex-common = { path = "../common", features = ["cli"] }
codex-exec = { path = "../exec" }
codex-mcp-server = { path = "../mcp-server" }
codex-tui = { path = "../tui" }
serde_json = "1"
tokio = { version = "1", features = [

View File

@@ -0,0 +1,23 @@
#[cfg(unix)]
pub(crate) fn handle_exit_status(status: std::process::ExitStatus) -> ! {
use std::os::unix::process::ExitStatusExt;
// Use ExitStatus to derive the exit code.
if let Some(code) = status.code() {
std::process::exit(code);
} else if let Some(signal) = status.signal() {
std::process::exit(128 + signal);
} else {
std::process::exit(1);
}
}
#[cfg(windows)]
pub(crate) fn handle_exit_status(status: std::process::ExitStatus) -> ! {
if let Some(code) = status.code() {
std::process::exit(code);
} else {
// Rare on Windows, but if it happens: use fallback code.
std::process::exit(1);
}
}

View File

@@ -3,12 +3,14 @@
//! On Linux the command is executed inside a Landlock + seccomp sandbox by
//! calling the low-level `exec_linux` helper from `codex_core::linux`.
use codex_core::exec::StdioPolicy;
use codex_core::exec::spawn_child_sync;
use codex_core::exec_linux::apply_sandbox_policy_to_current_thread;
use codex_core::protocol::SandboxPolicy;
use std::os::unix::process::ExitStatusExt;
use std::process;
use std::process::Command;
use std::process::ExitStatus;
use crate::exit_status::handle_exit_status;
/// Execute `command` in a Linux sandbox (Landlock + seccomp) the way Codex
/// would.
pub fn run_landlock(command: Vec<String>, sandbox_policy: SandboxPolicy) -> anyhow::Result<()> {
@@ -19,20 +21,15 @@ pub fn run_landlock(command: Vec<String>, sandbox_policy: SandboxPolicy) -> anyh
// Spawn a new thread and apply the sandbox policies there.
let handle = std::thread::spawn(move || -> anyhow::Result<ExitStatus> {
let cwd = std::env::current_dir()?;
codex_core::linux::apply_sandbox_policy_to_current_thread(sandbox_policy, &cwd)?;
let status = Command::new(&command[0]).args(&command[1..]).status()?;
apply_sandbox_policy_to_current_thread(&sandbox_policy, &cwd)?;
let mut child = spawn_child_sync(command, cwd, &sandbox_policy, StdioPolicy::Inherit)?;
let status = child.wait()?;
Ok(status)
});
let status = handle
.join()
.map_err(|e| anyhow::anyhow!("Failed to join thread: {e:?}"))??;
// Use ExitStatus to derive the exit code.
if let Some(code) = status.code() {
process::exit(code);
} else if let Some(signal) = status.signal() {
process::exit(128 + signal);
} else {
process::exit(1);
}
handle_exit_status(status);
}

View File

@@ -1,4 +1,5 @@
#[cfg(target_os = "linux")]
mod exit_status;
#[cfg(unix)]
pub mod landlock;
pub mod proto;
pub mod seatbelt;

View File

@@ -33,6 +33,9 @@ enum Subcommand {
#[clap(visible_alias = "e")]
Exec(ExecCli),
/// Experimental: run Codex as an MCP server.
Mcp,
/// Run the Protocol stream via stdin/stdout
#[clap(visible_alias = "p")]
Proto(ProtoCli),
@@ -70,6 +73,9 @@ async fn main() -> anyhow::Result<()> {
Some(Subcommand::Exec(exec_cli)) => {
codex_exec::run_main(exec_cli).await?;
}
Some(Subcommand::Mcp) => {
codex_mcp_server::run_main().await?;
}
Some(Subcommand::Proto(proto_cli)) => {
proto::run_main(proto_cli).await?;
}
@@ -82,7 +88,7 @@ async fn main() -> anyhow::Result<()> {
let sandbox_policy = create_sandbox_policy(full_auto, sandbox);
seatbelt::run_seatbelt(command, sandbox_policy).await?;
}
#[cfg(target_os = "linux")]
#[cfg(unix)]
DebugCommand::Landlock(LandlockCommand {
command,
sandbox,
@@ -91,7 +97,7 @@ async fn main() -> anyhow::Result<()> {
let sandbox_policy = create_sandbox_policy(full_auto, sandbox);
codex_cli::landlock::run_landlock(command, sandbox_policy)?;
}
#[cfg(not(target_os = "linux"))]
#[cfg(not(unix))]
DebugCommand::Landlock(_) => {
anyhow::bail!("Landlock is only supported on Linux.");
}

View File

@@ -81,8 +81,13 @@ pub async fn run_main(_opts: ProtoCli) -> anyhow::Result<()> {
};
match event {
Ok(event) => {
let event_str =
serde_json::to_string(&event).expect("JSON serialization failed");
let event_str = match serde_json::to_string(&event) {
Ok(s) => s,
Err(e) => {
error!("Failed to serialize event: {e}");
continue;
}
};
println!("{event_str}");
}
Err(e) => {

View File

@@ -1,18 +1,16 @@
use codex_core::exec::create_seatbelt_command;
use codex_core::exec::StdioPolicy;
use codex_core::exec::spawn_command_under_seatbelt;
use codex_core::protocol::SandboxPolicy;
use crate::exit_status::handle_exit_status;
pub async fn run_seatbelt(
command: Vec<String>,
sandbox_policy: SandboxPolicy,
) -> anyhow::Result<()> {
let cwd = std::env::current_dir().expect("failed to get cwd");
let seatbelt_command = create_seatbelt_command(command, &sandbox_policy, &cwd);
let status = tokio::process::Command::new(seatbelt_command[0].clone())
.args(&seatbelt_command[1..])
.spawn()
.map_err(|e| anyhow::anyhow!("Failed to spawn command: {}", e))?
.wait()
.await
.map_err(|e| anyhow::anyhow!("Failed to wait for command: {}", e))?;
std::process::exit(status.code().unwrap_or(1));
let cwd = std::env::current_dir()?;
let mut child =
spawn_command_under_seatbelt(command, &sandbox_policy, cwd, StdioPolicy::Inherit).await?;
let status = child.wait().await?;
handle_exit_status(status);
}

View File

@@ -7,11 +7,10 @@ edition = "2024"
workspace = true
[dependencies]
chrono = { version = "0.4.40", optional = true }
clap = { version = "4", features = ["derive", "wrap_help"], optional = true }
codex-core = { path = "../core" }
[features]
# Separate feature so that `clap` is not a mandatory dependency.
cli = ["clap"]
elapsed = ["chrono"]
elapsed = []

View File

@@ -1,18 +1,19 @@
use chrono::Utc;
use std::time::Duration;
use std::time::Instant;
/// Returns a string representing the elapsed time since `start_time` like
/// "1m15s" or "1.50s".
pub fn format_elapsed(start_time: chrono::DateTime<Utc>) -> String {
let elapsed = Utc::now().signed_duration_since(start_time);
format_time_delta(elapsed)
pub fn format_elapsed(start_time: Instant) -> String {
format_duration(start_time.elapsed())
}
fn format_time_delta(elapsed: chrono::TimeDelta) -> String {
let millis = elapsed.num_milliseconds();
format_elapsed_millis(millis)
}
pub fn format_duration(duration: std::time::Duration) -> String {
/// Convert a [`std::time::Duration`] into a human-readable, compact string.
///
/// Formatting rules:
/// * < 1 s -> "{milli}ms"
/// * < 60 s -> "{sec:.2}s" (two decimal places)
/// * >= 60 s -> "{min}m{sec:02}s"
pub fn format_duration(duration: Duration) -> String {
let millis = duration.as_millis() as i64;
format_elapsed_millis(millis)
}
@@ -32,41 +33,40 @@ fn format_elapsed_millis(millis: i64) -> String {
#[cfg(test)]
mod tests {
use super::*;
use chrono::Duration;
#[test]
fn test_format_time_delta_subsecond() {
fn test_format_duration_subsecond() {
// Durations < 1s should be rendered in milliseconds with no decimals.
let dur = Duration::milliseconds(250);
assert_eq!(format_time_delta(dur), "250ms");
let dur = Duration::from_millis(250);
assert_eq!(format_duration(dur), "250ms");
// Exactly zero should still work.
let dur_zero = Duration::milliseconds(0);
assert_eq!(format_time_delta(dur_zero), "0ms");
let dur_zero = Duration::from_millis(0);
assert_eq!(format_duration(dur_zero), "0ms");
}
#[test]
fn test_format_time_delta_seconds() {
fn test_format_duration_seconds() {
// Durations between 1s (inclusive) and 60s (exclusive) should be
// printed with 2-decimal-place seconds.
let dur = Duration::milliseconds(1_500); // 1.5s
assert_eq!(format_time_delta(dur), "1.50s");
let dur = Duration::from_millis(1_500); // 1.5s
assert_eq!(format_duration(dur), "1.50s");
// 59.999s rounds to 60.00s
let dur2 = Duration::milliseconds(59_999);
assert_eq!(format_time_delta(dur2), "60.00s");
let dur2 = Duration::from_millis(59_999);
assert_eq!(format_duration(dur2), "60.00s");
}
#[test]
fn test_format_time_delta_minutes() {
fn test_format_duration_minutes() {
// Durations ≥ 1 minute should be printed mmss.
let dur = Duration::milliseconds(75_000); // 1m15s
assert_eq!(format_time_delta(dur), "1m15s");
let dur = Duration::from_millis(75_000); // 1m15s
assert_eq!(format_duration(dur), "1m15s");
let dur_exact = Duration::milliseconds(60_000); // 1m0s
assert_eq!(format_time_delta(dur_exact), "1m00s");
let dur_exact = Duration::from_millis(60_000); // 1m0s
assert_eq!(format_duration(dur_exact), "1m00s");
let dur_long = Duration::milliseconds(3_601_000);
assert_eq!(format_time_delta(dur_long), "60m01s");
let dur_long = Duration::from_millis(3_601_000);
assert_eq!(format_duration(dur_long), "60m01s");
}
}

View File

@@ -20,6 +20,7 @@ codex-mcp-client = { path = "../mcp-client" }
dirs = "6"
env-flags = "0.1.1"
eventsource-stream = "0.2.3"
fs2 = "0.4.3"
fs-err = "3.1.0"
futures = "0.3"
mcp-types = { path = "../mcp-types" }
@@ -31,7 +32,7 @@ reqwest = { version = "0.12", features = ["json", "stream"] }
serde = { version = "1", features = ["derive"] }
serde_json = "1"
thiserror = "2.0.12"
time = { version = "0.3", features = ["formatting", "macros"] }
time = { version = "0.3", features = ["formatting", "local-offset", "macros"] }
tokio = { version = "1", features = [
"io-std",
"macros",
@@ -44,7 +45,7 @@ toml = "0.8.20"
tracing = { version = "0.1.41", features = ["log"] }
tree-sitter = "0.25.3"
tree-sitter-bash = "0.23.3"
uuid = { version = "1", features = ["v4"] }
uuid = { version = "1", features = ["serde", "v4"] }
[target.'cfg(target_os = "linux")'.dependencies]
libc = "0.2.172"
@@ -58,5 +59,6 @@ openssl-sys = { version = "*", features = ["vendored"] }
[dev-dependencies]
assert_cmd = "2"
predicates = "3"
pretty_assertions = "1.4.1"
tempfile = "3"
wiremock = "0.6"

View File

@@ -38,9 +38,8 @@ pub(crate) async fn stream_chat_completions(
// Build messages array
let mut messages = Vec::<serde_json::Value>::new();
if let Some(instr) = &prompt.instructions {
messages.push(json!({"role": "system", "content": instr}));
}
let full_instructions = prompt.get_full_instructions();
messages.push(json!({"role": "system", "content": full_instructions}));
for item in &prompt.input {
if let ResponseItem::Message { role, content } = item {

View File

@@ -26,7 +26,9 @@ use crate::client_common::Prompt;
use crate::client_common::Reasoning;
use crate::client_common::ResponseEvent;
use crate::client_common::ResponseStream;
use crate::client_common::Summary;
use crate::error::CodexErr;
use crate::error::EnvVarError;
use crate::error::Result;
use crate::flags::CODEX_RS_SSE_FIXTURE;
use crate::flags::OPENAI_REQUEST_MAX_RETRIES;
@@ -38,10 +40,18 @@ use crate::util::backoff;
/// When serialized as JSON, this produces a valid "Tool" in the OpenAI
/// Responses API.
#[derive(Debug, Serialize)]
#[derive(Debug, Clone, Serialize)]
#[serde(tag = "type")]
enum OpenAiTool {
#[serde(rename = "function")]
Function(ResponsesApiTool),
#[serde(rename = "local_shell")]
LocalShell {},
}
#[derive(Debug, Clone, Serialize)]
struct ResponsesApiTool {
name: &'static str,
r#type: &'static str, // "function"
description: &'static str,
strict: bool,
parameters: JsonSchema,
@@ -65,7 +75,7 @@ enum JsonSchema {
}
/// Tool usage specification
static DEFAULT_TOOLS: LazyLock<Vec<ResponsesApiTool>> = LazyLock::new(|| {
static DEFAULT_TOOLS: LazyLock<Vec<OpenAiTool>> = LazyLock::new(|| {
let mut properties = BTreeMap::new();
properties.insert(
"command".to_string(),
@@ -76,9 +86,8 @@ static DEFAULT_TOOLS: LazyLock<Vec<ResponsesApiTool>> = LazyLock::new(|| {
properties.insert("workdir".to_string(), JsonSchema::String);
properties.insert("timeout".to_string(), JsonSchema::Number);
vec![ResponsesApiTool {
vec![OpenAiTool::Function(ResponsesApiTool {
name: "shell",
r#type: "function",
description: "Runs a shell command, and returns its output.",
strict: false,
parameters: JsonSchema::Object {
@@ -86,9 +95,12 @@ static DEFAULT_TOOLS: LazyLock<Vec<ResponsesApiTool>> = LazyLock::new(|| {
required: &["command"],
additional_properties: false,
},
}]
})]
});
static DEFAULT_CODEX_MODEL_TOOLS: LazyLock<Vec<OpenAiTool>> =
LazyLock::new(|| vec![OpenAiTool::LocalShell {}]);
#[derive(Clone)]
pub struct ModelClient {
model: String,
@@ -150,10 +162,15 @@ impl ModelClient {
}
// Assemble tool list: built-in tools + any extra tools from the prompt.
let mut tools_json: Vec<serde_json::Value> = DEFAULT_TOOLS
.iter()
.map(|t| serde_json::to_value(t).expect("serialize builtin tool"))
.collect();
let default_tools = if self.model.starts_with("codex") {
&DEFAULT_CODEX_MODEL_TOOLS
} else {
&DEFAULT_TOOLS
};
let mut tools_json = Vec::with_capacity(default_tools.len() + prompt.extra_tools.len());
for t in default_tools.iter() {
tools_json.push(serde_json::to_value(t)?);
}
tools_json.extend(
prompt
.extra_tools
@@ -164,16 +181,17 @@ impl ModelClient {
debug!("tools_json: {}", serde_json::to_string_pretty(&tools_json)?);
let full_instructions = prompt.get_full_instructions();
let payload = Payload {
model: &self.model,
instructions: prompt.instructions.as_ref(),
instructions: &full_instructions,
input: &prompt.input,
tools: &tools_json,
tool_choice: "auto",
parallel_tool_calls: false,
reasoning: Some(Reasoning {
effort: "high",
generate_summary: None,
summary: Some(Summary::Auto),
}),
previous_response_id: prompt.prev_id.clone(),
store: prompt.store,
@@ -190,10 +208,12 @@ impl ModelClient {
loop {
attempt += 1;
let api_key = self
.provider
.api_key()?
.expect("Repsones API requires an API key");
let api_key = self.provider.api_key()?.ok_or_else(|| {
CodexErr::EnvVar(EnvVarError {
var: self.provider.env_key.clone().unwrap_or_default(),
instructions: None,
})
})?;
let res = self
.client
.post(&url)

View File

@@ -2,12 +2,17 @@ use crate::error::Result;
use crate::models::ResponseItem;
use futures::Stream;
use serde::Serialize;
use std::borrow::Cow;
use std::collections::HashMap;
use std::pin::Pin;
use std::task::Context;
use std::task::Poll;
use tokio::sync::mpsc;
/// The `instructions` field in the payload sent to a model should always start
/// with this content.
const BASE_INSTRUCTIONS: &str = include_str!("../prompt.md");
/// API request payload for a single model turn.
#[derive(Default, Debug, Clone)]
pub struct Prompt {
@@ -15,7 +20,8 @@ pub struct Prompt {
pub input: Vec<ResponseItem>,
/// Optional previous response ID (when storage is enabled).
pub prev_id: Option<String>,
/// Optional initial instructions (only sent on first turn).
/// Optional instructions from the user to amend to the built-in agent
/// instructions.
pub instructions: Option<String>,
/// Whether to store response on server side (disable_response_storage = !store).
pub store: bool,
@@ -26,6 +32,18 @@ pub struct Prompt {
pub extra_tools: HashMap<String, mcp_types::Tool>,
}
impl Prompt {
pub(crate) fn get_full_instructions(&self) -> Cow<str> {
match &self.instructions {
Some(instructions) => {
let instructions = format!("{BASE_INSTRUCTIONS}\n{instructions}");
Cow::Owned(instructions)
}
None => Cow::Borrowed(BASE_INSTRUCTIONS),
}
}
}
#[derive(Debug)]
pub enum ResponseEvent {
OutputItemDone(ResponseItem),
@@ -36,14 +54,25 @@ pub enum ResponseEvent {
pub(crate) struct Reasoning {
pub(crate) effort: &'static str,
#[serde(skip_serializing_if = "Option::is_none")]
pub(crate) generate_summary: Option<bool>,
pub(crate) summary: Option<Summary>,
}
/// A summary of the reasoning performed by the model. This can be useful for
/// debugging and understanding the model's reasoning process.
#[derive(Debug, Serialize)]
#[serde(rename_all = "lowercase")]
pub(crate) enum Summary {
Auto,
#[allow(dead_code)] // Will go away once this is configurable.
Concise,
#[allow(dead_code)] // Will go away once this is configurable.
Detailed,
}
#[derive(Debug, Serialize)]
pub(crate) struct Payload<'a> {
pub(crate) model: &'a str,
#[serde(skip_serializing_if = "Option::is_none")]
pub(crate) instructions: Option<&'a String>,
pub(crate) instructions: &'a str,
// TODO(mbolin): ResponseItem::Other should not be serialized. Currently,
// we code defensively to avoid this case, but perhaps we should use a
// separate enum for serialization.

File diff suppressed because it is too large Load Diff

View File

@@ -24,7 +24,7 @@ pub async fn init_codex(config: Config) -> anyhow::Result<(Codex, Event, Arc<Not
&event,
Event {
id: _id,
msg: EventMsg::SessionConfigured { .. },
msg: EventMsg::SessionConfigured(_),
}
)
{

View File

@@ -1,5 +1,9 @@
use crate::config_profile::ConfigProfile;
use crate::config_types::History;
use crate::config_types::McpServerConfig;
use crate::config_types::Tui;
use crate::config_types::UriBasedFileOpener;
use crate::flags::OPENAI_DEFAULT_MODEL;
use crate::mcp_server_config::McpServerConfig;
use crate::model_provider_info::ModelProviderInfo;
use crate::model_provider_info::built_in_model_providers;
use crate::protocol::AskForApproval;
@@ -8,15 +12,16 @@ use crate::protocol::SandboxPolicy;
use dirs::home_dir;
use serde::Deserialize;
use std::collections::HashMap;
use std::path::Path;
use std::path::PathBuf;
/// Embedded fallback instructions that mirror the TypeScript CLIs default
/// system prompt. These are compiled into the binary so a clean install behaves
/// correctly even if the user has not created `~/.codex/instructions.md`.
const EMBEDDED_INSTRUCTIONS: &str = include_str!("../prompt.md");
/// Maximum number of bytes of the documentation that will be embedded. Larger
/// files are *silently truncated* to this size so we do not take up too much of
/// the context window.
pub(crate) const PROJECT_DOC_MAX_BYTES: usize = 32 * 1024; // 32 KiB
/// Application configuration loaded from disk and merged with overrides.
#[derive(Debug, Clone)]
#[derive(Debug, Clone, PartialEq)]
pub struct Config {
/// Optional override of model selection.
pub model: String,
@@ -37,7 +42,7 @@ pub struct Config {
/// who have opted into Zero Data Retention (ZDR).
pub disable_response_storage: bool,
/// System instructions.
/// User-provided instructions from instructions.md.
pub instructions: Option<String>,
/// Optional external notifier command. When set, Codex will spawn this
@@ -72,6 +77,23 @@ pub struct Config {
/// Combined provider map (defaults merged with user-defined overrides).
pub model_providers: HashMap<String, ModelProviderInfo>,
/// Maximum number of bytes to include from an AGENTS.md project doc file.
pub project_doc_max_bytes: usize,
/// Directory containing all Codex state (defaults to `~/.codex` but can be
/// overridden by the `CODEX_HOME` environment variable).
pub codex_home: PathBuf,
/// Settings that govern if and what will be written to `~/.codex/history.jsonl`.
pub history: History,
/// Optional URI-based file opener. If set, citations to files in the model
/// output will be hyperlinked using the specified URI scheme.
pub file_opener: UriBasedFileOpener,
/// Collection of settings that are specific to the TUI.
pub tui: Tui,
}
/// Base config deserialized from ~/.codex/config.toml.
@@ -111,14 +133,35 @@ pub struct ConfigToml {
/// User-defined provider entries that extend/override the built-in list.
#[serde(default)]
pub model_providers: HashMap<String, ModelProviderInfo>,
/// Maximum number of bytes to include from an AGENTS.md project doc file.
pub project_doc_max_bytes: Option<usize>,
/// Profile to use from the `profiles` map.
pub profile: Option<String>,
/// Named profiles to facilitate switching between different configurations.
#[serde(default)]
pub profiles: HashMap<String, ConfigProfile>,
/// Settings that govern if and what will be written to `~/.codex/history.jsonl`.
#[serde(default)]
pub history: Option<History>,
/// Optional URI-based file opener. If set, citations to files in the model
/// output will be hyperlinked using the specified URI scheme.
pub file_opener: Option<UriBasedFileOpener>,
/// Collection of settings that are specific to the TUI.
pub tui: Option<Tui>,
}
impl ConfigToml {
/// Attempt to parse the file at `~/.codex/config.toml`. If it does not
/// exist, return a default config. Though if it exists and cannot be
/// parsed, report that to the user and force them to fix it.
fn load_from_toml() -> std::io::Result<Self> {
let config_toml_path = codex_dir()?.join("config.toml");
fn load_from_toml(codex_home: &Path) -> std::io::Result<Self> {
let config_toml_path = codex_home.join("config.toml");
match std::fs::read_to_string(&config_toml_path) {
Ok(contents) => toml::from_str::<Self>(&contents).map_err(|e| {
tracing::error!("Failed to parse config.toml: {e}");
@@ -146,7 +189,7 @@ where
match permissions {
Some(raw_permissions) => {
let base_path = codex_dir().map_err(serde::de::Error::custom)?;
let base_path = find_codex_home().map_err(serde::de::Error::custom)?;
let converted = raw_permissions
.into_iter()
@@ -170,7 +213,8 @@ pub struct ConfigOverrides {
pub approval_policy: Option<AskForApproval>,
pub sandbox_policy: Option<SandboxPolicy>,
pub disable_response_storage: Option<bool>,
pub provider: Option<String>,
pub model_provider: Option<String>,
pub config_profile: Option<String>,
}
impl Config {
@@ -178,18 +222,25 @@ impl Config {
/// ~/.codex/config.toml, ~/.codex/instructions.md, embedded defaults, and
/// any values provided in `overrides` (highest precedence).
pub fn load_with_overrides(overrides: ConfigOverrides) -> std::io::Result<Self> {
let cfg: ConfigToml = ConfigToml::load_from_toml()?;
// Resolve the directory that stores Codex state (e.g. ~/.codex or the
// value of $CODEX_HOME) so we can embed it into the resulting
// `Config` instance.
let codex_home = find_codex_home()?;
let cfg: ConfigToml = ConfigToml::load_from_toml(&codex_home)?;
tracing::warn!("Config parsed from config.toml: {cfg:?}");
Self::load_from_base_config_with_overrides(cfg, overrides)
Self::load_from_base_config_with_overrides(cfg, overrides, codex_home)
}
fn load_from_base_config_with_overrides(
/// Meant to be used exclusively for tests: `load_with_overrides()` should
/// be used in all other cases.
pub fn load_from_base_config_with_overrides(
cfg: ConfigToml,
overrides: ConfigOverrides,
codex_home: PathBuf,
) -> std::io::Result<Self> {
// Instructions: user-provided instructions.md > embedded default.
let instructions =
Self::load_instructions().or_else(|| Some(EMBEDDED_INSTRUCTIONS.to_string()));
let instructions = Self::load_instructions(Some(&codex_home));
// Destructure ConfigOverrides fully to ensure all overrides are applied.
let ConfigOverrides {
@@ -198,9 +249,24 @@ impl Config {
approval_policy,
sandbox_policy,
disable_response_storage,
provider,
model_provider,
config_profile: config_profile_key,
} = overrides;
let config_profile = match config_profile_key.or(cfg.profile) {
Some(key) => cfg
.profiles
.get(&key)
.ok_or_else(|| {
std::io::Error::new(
std::io::ErrorKind::NotFound,
format!("config profile `{key}` not found"),
)
})?
.clone(),
None => ConfigProfile::default(),
};
let sandbox_policy = match sandbox_policy {
Some(sandbox_policy) => sandbox_policy,
None => {
@@ -222,7 +288,8 @@ impl Config {
model_providers.entry(key).or_insert(provider);
}
let model_provider_id = provider
let model_provider_id = model_provider
.or(config_profile.model_provider)
.or(cfg.model_provider)
.unwrap_or_else(|| "openai".to_string());
let model_provider = model_providers
@@ -235,56 +302,72 @@ impl Config {
})?
.clone();
let resolved_cwd = {
use std::env;
match cwd {
None => {
tracing::info!("cwd not set, using current dir");
env::current_dir()?
}
Some(p) if p.is_absolute() => p,
Some(p) => {
// Resolve relative path against the current working directory.
tracing::info!("cwd is relative, resolving against current dir");
let mut current = env::current_dir()?;
current.push(p);
current
}
}
};
let history = cfg.history.unwrap_or_default();
let config = Self {
model: model.or(cfg.model).unwrap_or_else(default_model),
model: model
.or(config_profile.model)
.or(cfg.model)
.unwrap_or_else(default_model),
model_provider_id,
model_provider,
cwd: cwd.map_or_else(
|| {
tracing::info!("cwd not set, using current dir");
std::env::current_dir().expect("cannot determine current dir")
},
|p| {
if p.is_absolute() {
p
} else {
// Resolve relative paths against the current working directory.
tracing::info!("cwd is relative, resolving against current dir");
let mut cwd = std::env::current_dir().expect("cannot determine cwd");
cwd.push(p);
cwd
}
},
),
cwd: resolved_cwd,
approval_policy: approval_policy
.or(config_profile.approval_policy)
.or(cfg.approval_policy)
.unwrap_or_else(AskForApproval::default),
sandbox_policy,
disable_response_storage: disable_response_storage
.or(config_profile.disable_response_storage)
.or(cfg.disable_response_storage)
.unwrap_or(false),
notify: cfg.notify,
instructions,
mcp_servers: cfg.mcp_servers,
model_providers,
project_doc_max_bytes: cfg.project_doc_max_bytes.unwrap_or(PROJECT_DOC_MAX_BYTES),
codex_home,
history,
file_opener: cfg.file_opener.unwrap_or(UriBasedFileOpener::VsCode),
tui: cfg.tui.unwrap_or_default(),
};
Ok(config)
}
fn load_instructions() -> Option<String> {
let mut p = codex_dir().ok()?;
p.push("instructions.md");
std::fs::read_to_string(&p).ok()
}
fn load_instructions(codex_dir: Option<&Path>) -> Option<String> {
let mut p = match codex_dir {
Some(p) => p.to_path_buf(),
None => return None,
};
/// Meant to be used exclusively for tests: `load_with_overrides()` should
/// be used in all other cases.
pub fn load_default_config_for_test() -> Self {
Self::load_from_base_config_with_overrides(
ConfigToml::default(),
ConfigOverrides::default(),
)
.expect("defaults for test should always succeed")
p.push("instructions.md");
std::fs::read_to_string(&p).ok().and_then(|s| {
let s = s.trim();
if s.is_empty() {
None
} else {
Some(s.to_string())
}
})
}
}
@@ -292,9 +375,23 @@ fn default_model() -> String {
OPENAI_DEFAULT_MODEL.to_string()
}
/// Returns the path to the Codex configuration directory, which is `~/.codex`.
/// Does not verify that the directory exists.
pub fn codex_dir() -> std::io::Result<PathBuf> {
/// Returns the path to the Codex configuration directory, which can be
/// specified by the `CODEX_HOME` environment variable. If not set, defaults to
/// `~/.codex`.
///
/// - If `CODEX_HOME` is set, the value will be canonicalized and this
/// function will Err if the path does not exist.
/// - If `CODEX_HOME` is not set, this function does not verify that the
/// directory exists.
fn find_codex_home() -> std::io::Result<PathBuf> {
// Honor the `CODEX_HOME` environment variable when it is set to allow users
// (and tests) to override the default location.
if let Ok(val) = std::env::var("CODEX_HOME") {
if !val.is_empty() {
return PathBuf::from(val).canonicalize();
}
}
let mut p = home_dir().ok_or_else(|| {
std::io::Error::new(
std::io::ErrorKind::NotFound,
@@ -307,8 +404,8 @@ pub fn codex_dir() -> std::io::Result<PathBuf> {
/// Returns the path to the folder where Codex logs are stored. Does not verify
/// that the directory exists.
pub fn log_dir() -> std::io::Result<PathBuf> {
let mut p = codex_dir()?;
pub fn log_dir(cfg: &Config) -> std::io::Result<PathBuf> {
let mut p = cfg.codex_home.clone();
p.push("log");
Ok(p)
}
@@ -359,7 +456,12 @@ pub fn parse_sandbox_permission_with_base_path(
#[cfg(test)]
mod tests {
#![allow(clippy::expect_used, clippy::unwrap_used)]
use crate::config_types::HistoryPersistence;
use super::*;
use pretty_assertions::assert_eq;
use tempfile::TempDir;
/// Verify that the `sandbox_permissions` field on `ConfigToml` correctly
/// differentiates between a value that is completely absent in the
@@ -397,6 +499,40 @@ mod tests {
);
}
#[test]
fn test_toml_parsing() {
let history_with_persistence = r#"
[history]
persistence = "save-all"
"#;
let history_with_persistence_cfg: ConfigToml =
toml::from_str::<ConfigToml>(history_with_persistence)
.expect("TOML deserialization should succeed");
assert_eq!(
Some(History {
persistence: HistoryPersistence::SaveAll,
max_bytes: None,
}),
history_with_persistence_cfg.history
);
let history_no_persistence = r#"
[history]
persistence = "none"
"#;
let history_no_persistence_cfg: ConfigToml =
toml::from_str::<ConfigToml>(history_no_persistence)
.expect("TOML deserialization should succeed");
assert_eq!(
Some(History {
persistence: HistoryPersistence::None,
max_bytes: None,
}),
history_no_persistence_cfg.history
);
}
/// Deserializing a TOML string containing an *invalid* permission should
/// fail with a helpful error rather than silently defaulting or
/// succeeding.
@@ -412,4 +548,239 @@ mod tests {
let msg = err.to_string();
assert!(msg.contains("not-a-real-permission"));
}
struct PrecedenceTestFixture {
cwd: TempDir,
codex_home: TempDir,
cfg: ConfigToml,
model_provider_map: HashMap<String, ModelProviderInfo>,
openai_provider: ModelProviderInfo,
openai_chat_completions_provider: ModelProviderInfo,
}
impl PrecedenceTestFixture {
fn cwd(&self) -> PathBuf {
self.cwd.path().to_path_buf()
}
fn codex_home(&self) -> PathBuf {
self.codex_home.path().to_path_buf()
}
}
fn create_test_fixture() -> std::io::Result<PrecedenceTestFixture> {
let toml = r#"
model = "o3"
approval_policy = "unless-allow-listed"
sandbox_permissions = ["disk-full-read-access"]
disable_response_storage = false
# Can be used to determine which profile to use if not specified by
# `ConfigOverrides`.
profile = "gpt3"
[model_providers.openai-chat-completions]
name = "OpenAI using Chat Completions"
base_url = "https://api.openai.com/v1"
env_key = "OPENAI_API_KEY"
wire_api = "chat"
[profiles.o3]
model = "o3"
model_provider = "openai"
approval_policy = "never"
[profiles.gpt3]
model = "gpt-3.5-turbo"
model_provider = "openai-chat-completions"
[profiles.zdr]
model = "o3"
model_provider = "openai"
approval_policy = "on-failure"
disable_response_storage = true
"#;
let cfg: ConfigToml = toml::from_str(toml).expect("TOML deserialization should succeed");
// Use a temporary directory for the cwd so it does not contain an
// AGENTS.md file.
let cwd_temp_dir = TempDir::new().unwrap();
let cwd = cwd_temp_dir.path().to_path_buf();
// Make it look like a Git repo so it does not search for AGENTS.md in
// a parent folder, either.
std::fs::write(cwd.join(".git"), "gitdir: nowhere")?;
let codex_home_temp_dir = TempDir::new().unwrap();
let openai_chat_completions_provider = ModelProviderInfo {
name: "OpenAI using Chat Completions".to_string(),
base_url: "https://api.openai.com/v1".to_string(),
env_key: Some("OPENAI_API_KEY".to_string()),
wire_api: crate::WireApi::Chat,
env_key_instructions: None,
};
let model_provider_map = {
let mut model_provider_map = built_in_model_providers();
model_provider_map.insert(
"openai-chat-completions".to_string(),
openai_chat_completions_provider.clone(),
);
model_provider_map
};
let openai_provider = model_provider_map
.get("openai")
.expect("openai provider should exist")
.clone();
Ok(PrecedenceTestFixture {
cwd: cwd_temp_dir,
codex_home: codex_home_temp_dir,
cfg,
model_provider_map,
openai_provider,
openai_chat_completions_provider,
})
}
/// Users can specify config values at multiple levels that have the
/// following precedence:
///
/// 1. custom command-line argument, e.g. `--model o3`
/// 2. as part of a profile, where the `--profile` is specified via a CLI
/// (or in the config file itelf)
/// 3. as an entry in `config.toml`, e.g. `model = "o3"`
/// 4. the default value for a required field defined in code, e.g.,
/// `crate::flags::OPENAI_DEFAULT_MODEL`
///
/// Note that profiles are the recommended way to specify a group of
/// configuration options together.
#[test]
fn test_precedence_fixture_with_o3_profile() -> std::io::Result<()> {
let fixture = create_test_fixture()?;
let o3_profile_overrides = ConfigOverrides {
config_profile: Some("o3".to_string()),
cwd: Some(fixture.cwd()),
..Default::default()
};
let o3_profile_config: Config = Config::load_from_base_config_with_overrides(
fixture.cfg.clone(),
o3_profile_overrides,
fixture.codex_home(),
)?;
assert_eq!(
Config {
model: "o3".to_string(),
model_provider_id: "openai".to_string(),
model_provider: fixture.openai_provider.clone(),
approval_policy: AskForApproval::Never,
sandbox_policy: SandboxPolicy::new_read_only_policy(),
disable_response_storage: false,
instructions: None,
notify: None,
cwd: fixture.cwd(),
mcp_servers: HashMap::new(),
model_providers: fixture.model_provider_map.clone(),
project_doc_max_bytes: PROJECT_DOC_MAX_BYTES,
codex_home: fixture.codex_home(),
history: History::default(),
file_opener: UriBasedFileOpener::VsCode,
tui: Tui::default(),
},
o3_profile_config
);
Ok(())
}
#[test]
fn test_precedence_fixture_with_gpt3_profile() -> std::io::Result<()> {
let fixture = create_test_fixture()?;
let gpt3_profile_overrides = ConfigOverrides {
config_profile: Some("gpt3".to_string()),
cwd: Some(fixture.cwd()),
..Default::default()
};
let gpt3_profile_config = Config::load_from_base_config_with_overrides(
fixture.cfg.clone(),
gpt3_profile_overrides,
fixture.codex_home(),
)?;
let expected_gpt3_profile_config = Config {
model: "gpt-3.5-turbo".to_string(),
model_provider_id: "openai-chat-completions".to_string(),
model_provider: fixture.openai_chat_completions_provider.clone(),
approval_policy: AskForApproval::UnlessAllowListed,
sandbox_policy: SandboxPolicy::new_read_only_policy(),
disable_response_storage: false,
instructions: None,
notify: None,
cwd: fixture.cwd(),
mcp_servers: HashMap::new(),
model_providers: fixture.model_provider_map.clone(),
project_doc_max_bytes: PROJECT_DOC_MAX_BYTES,
codex_home: fixture.codex_home(),
history: History::default(),
file_opener: UriBasedFileOpener::VsCode,
tui: Tui::default(),
};
assert_eq!(expected_gpt3_profile_config, gpt3_profile_config);
// Verify that loading without specifying a profile in ConfigOverrides
// uses the default profile from the config file (which is "gpt3").
let default_profile_overrides = ConfigOverrides {
cwd: Some(fixture.cwd()),
..Default::default()
};
let default_profile_config = Config::load_from_base_config_with_overrides(
fixture.cfg.clone(),
default_profile_overrides,
fixture.codex_home(),
)?;
assert_eq!(expected_gpt3_profile_config, default_profile_config);
Ok(())
}
#[test]
fn test_precedence_fixture_with_zdr_profile() -> std::io::Result<()> {
let fixture = create_test_fixture()?;
let zdr_profile_overrides = ConfigOverrides {
config_profile: Some("zdr".to_string()),
cwd: Some(fixture.cwd()),
..Default::default()
};
let zdr_profile_config = Config::load_from_base_config_with_overrides(
fixture.cfg.clone(),
zdr_profile_overrides,
fixture.codex_home(),
)?;
let expected_zdr_profile_config = Config {
model: "o3".to_string(),
model_provider_id: "openai".to_string(),
model_provider: fixture.openai_provider.clone(),
approval_policy: AskForApproval::OnFailure,
sandbox_policy: SandboxPolicy::new_read_only_policy(),
disable_response_storage: true,
instructions: None,
notify: None,
cwd: fixture.cwd(),
mcp_servers: HashMap::new(),
model_providers: fixture.model_provider_map.clone(),
project_doc_max_bytes: PROJECT_DOC_MAX_BYTES,
codex_home: fixture.codex_home(),
history: History::default(),
file_opener: UriBasedFileOpener::VsCode,
tui: Tui::default(),
};
assert_eq!(expected_zdr_profile_config, zdr_profile_config);
Ok(())
}
}

View File

@@ -0,0 +1,15 @@
use serde::Deserialize;
use crate::protocol::AskForApproval;
/// Collection of common configuration options that a user can define as a unit
/// in `config.toml`.
#[derive(Debug, Clone, Default, PartialEq, Deserialize)]
pub struct ConfigProfile {
pub model: Option<String>,
/// The key in the `model_providers` map identifying the
/// [`ModelProviderInfo`] to use.
pub model_provider: Option<String>,
pub approval_policy: Option<AskForApproval>,
pub disable_response_storage: Option<bool>,
}

View File

@@ -0,0 +1,88 @@
//! Types used to define the fields of [`crate::config::Config`].
// Note this file should generally be restricted to simple struct/enum
// definitions that do not contain business logic.
use std::collections::HashMap;
use serde::Deserialize;
#[derive(Deserialize, Debug, Clone, PartialEq)]
pub struct McpServerConfig {
pub command: String,
#[serde(default)]
pub args: Vec<String>,
#[serde(default)]
pub env: Option<HashMap<String, String>>,
}
#[derive(Deserialize, Debug, Copy, Clone, PartialEq)]
pub enum UriBasedFileOpener {
#[serde(rename = "vscode")]
VsCode,
#[serde(rename = "vscode-insiders")]
VsCodeInsiders,
#[serde(rename = "windsurf")]
Windsurf,
#[serde(rename = "cursor")]
Cursor,
/// Option to disable the URI-based file opener.
#[serde(rename = "none")]
None,
}
impl UriBasedFileOpener {
pub fn get_scheme(&self) -> Option<&str> {
match self {
UriBasedFileOpener::VsCode => Some("vscode"),
UriBasedFileOpener::VsCodeInsiders => Some("vscode-insiders"),
UriBasedFileOpener::Windsurf => Some("windsurf"),
UriBasedFileOpener::Cursor => Some("cursor"),
UriBasedFileOpener::None => None,
}
}
}
/// Settings that govern if and what will be written to `~/.codex/history.jsonl`.
#[derive(Deserialize, Debug, Clone, PartialEq, Default)]
pub struct History {
/// If true, history entries will not be written to disk.
pub persistence: HistoryPersistence,
/// If set, the maximum size of the history file in bytes.
/// TODO(mbolin): Not currently honored.
pub max_bytes: Option<usize>,
}
#[derive(Deserialize, Debug, Copy, Clone, PartialEq, Default)]
#[serde(rename_all = "kebab-case")]
pub enum HistoryPersistence {
/// Save all history entries to disk.
#[default]
SaveAll,
/// Do not write history to disk.
None,
}
/// Collection of settings that are specific to the TUI.
#[derive(Deserialize, Debug, Clone, PartialEq, Default)]
pub struct Tui {
/// By default, mouse capture is enabled in the TUI so that it is possible
/// to scroll the conversation history with a mouse. This comes at the cost
/// of not being able to use the mouse to select text in the TUI.
/// (Most terminals support a modifier key to allow this. For example,
/// text selection works in iTerm if you hold down the `Option` key while
/// clicking and dragging.)
///
/// Setting this option to `true` disables mouse capture, so scrolling with
/// the mouse is not possible, though the keyboard shortcuts e.g. `b` and
/// `space` still work. This allows the user to select text in the TUI
/// using the mouse without needing to hold down a modifier key.
pub disable_mouse_capture: bool,
}

View File

@@ -25,7 +25,8 @@ impl ConversationHistory {
/// `items` is ordered from oldest to newest.
pub(crate) fn record_items<I>(&mut self, items: I)
where
I: IntoIterator<Item = ResponseItem>,
I: IntoIterator,
I::Item: std::ops::Deref<Target = ResponseItem>,
{
for item in items {
if is_api_message(&item) {
@@ -41,8 +42,9 @@ impl ConversationHistory {
fn is_api_message(message: &ResponseItem) -> bool {
match message {
ResponseItem::Message { role, .. } => role.as_str() != "system",
ResponseItem::FunctionCall { .. } => true,
ResponseItem::FunctionCallOutput { .. } => true,
_ => false,
ResponseItem::FunctionCallOutput { .. }
| ResponseItem::FunctionCall { .. }
| ResponseItem::LocalShellCall { .. } => true,
ResponseItem::Reasoning { .. } | ResponseItem::Other => false,
}
}

View File

@@ -1,6 +1,7 @@
use std::io;
#[cfg(target_family = "unix")]
#[cfg(unix)]
use std::os::unix::process::ExitStatusExt;
use std::io;
use std::path::Path;
use std::path::PathBuf;
use std::process::ExitStatus;
@@ -19,6 +20,7 @@ use tokio::sync::Notify;
use crate::error::CodexErr;
use crate::error::Result;
use crate::error::SandboxErr;
use crate::exec_linux::exec_linux;
use crate::protocol::SandboxPolicy;
// Maximum we send for each stream, which is either:
@@ -42,6 +44,16 @@ const MACOS_SEATBELT_BASE_POLICY: &str = include_str!("seatbelt_base_policy.sbpl
/// already has root access.
const MACOS_PATH_TO_SEATBELT_EXECUTABLE: &str = "/usr/bin/sandbox-exec";
/// Experimental environment variable that will be set to some non-empty value
/// if both of the following are true:
///
/// 1. The process was spawned by Codex as part of a shell tool call.
/// 2. SandboxPolicy.has_full_network_access() was false for the tool call.
///
/// We may try to have just one environment variable for all sandboxing
/// attributes, so this may change in the future.
pub const CODEX_SANDBOX_NETWORK_DISABLED_ENV_VAR: &str = "CODEX_SANDBOX_NETWORK_DISABLED";
#[derive(Debug, Clone)]
pub struct ExecParams {
pub command: Vec<String>,
@@ -60,27 +72,6 @@ pub enum SandboxType {
LinuxSeccomp,
}
#[cfg(target_os = "linux")]
async fn exec_linux(
params: ExecParams,
ctrl_c: Arc<Notify>,
sandbox_policy: &SandboxPolicy,
) -> Result<RawExecToolCallOutput> {
crate::linux::exec_linux(params, ctrl_c, sandbox_policy).await
}
#[cfg(not(target_os = "linux"))]
async fn exec_linux(
_params: ExecParams,
_ctrl_c: Arc<Notify>,
_sandbox_policy: &SandboxPolicy,
) -> Result<RawExecToolCallOutput> {
Err(CodexErr::Io(io::Error::new(
io::ErrorKind::InvalidInput,
"linux sandbox is not supported on this platform",
)))
}
pub async fn process_exec_tool_call(
params: ExecParams,
sandbox_type: SandboxType,
@@ -90,25 +81,23 @@ pub async fn process_exec_tool_call(
let start = Instant::now();
let raw_output_result = match sandbox_type {
SandboxType::None => exec(params, ctrl_c).await,
SandboxType::None => exec(params, sandbox_policy, ctrl_c).await,
SandboxType::MacosSeatbelt => {
let ExecParams {
command,
cwd,
timeout_ms,
} = params;
let seatbelt_command = create_seatbelt_command(command, sandbox_policy, &cwd);
exec(
ExecParams {
command: seatbelt_command,
cwd,
timeout_ms,
},
ctrl_c,
let child = spawn_command_under_seatbelt(
command,
sandbox_policy,
cwd,
StdioPolicy::RedirectForShellTool,
)
.await
.await?;
consume_truncated_output(child, ctrl_c, timeout_ms).await
}
SandboxType::LinuxSeccomp => exec_linux(params, ctrl_c, sandbox_policy).await,
SandboxType::LinuxSeccomp => exec_linux(params, ctrl_c, sandbox_policy),
};
let duration = start.elapsed();
match raw_output_result {
@@ -151,7 +140,17 @@ pub async fn process_exec_tool_call(
}
}
pub fn create_seatbelt_command(
pub async fn spawn_command_under_seatbelt(
command: Vec<String>,
sandbox_policy: &SandboxPolicy,
cwd: PathBuf,
stdio_policy: StdioPolicy,
) -> std::io::Result<Child> {
let seatbelt_command = create_seatbelt_command(command, sandbox_policy, &cwd);
spawn_child_async(seatbelt_command, cwd, sandbox_policy, stdio_policy).await
}
fn create_seatbelt_command(
command: Vec<String>,
sandbox_policy: &SandboxPolicy,
cwd: &Path,
@@ -229,57 +228,144 @@ pub struct ExecToolCallOutput {
pub duration: Duration,
}
pub async fn exec(
async fn exec(
ExecParams {
command,
cwd,
timeout_ms,
}: ExecParams,
sandbox_policy: &SandboxPolicy,
ctrl_c: Arc<Notify>,
) -> Result<RawExecToolCallOutput> {
let child = spawn_child(command, cwd).await?;
let child = spawn_child_async(
command,
cwd,
sandbox_policy,
StdioPolicy::RedirectForShellTool,
)
.await?;
consume_truncated_output(child, ctrl_c, timeout_ms).await
}
/// Spawns the appropriate child process for the ExecParams.
async fn spawn_child(command: Vec<String>, cwd: PathBuf) -> std::io::Result<Child> {
if command.is_empty() {
return Err(std::io::Error::new(
io::ErrorKind::InvalidInput,
"command args are empty",
));
}
#[derive(Debug, Clone, Copy)]
pub enum StdioPolicy {
RedirectForShellTool,
Inherit,
}
let mut cmd = Command::new(&command[0]);
cmd.args(&command[1..]);
cmd.current_dir(cwd);
macro_rules! configure_command {
(
$cmd_type: path,
$command: expr,
$cwd: expr,
$sandbox_policy: expr,
$stdio_policy: expr
) => {{
// For now, we take `SandboxPolicy` as a parameter to spawn_child() because
// we need to determine whether to set the
// `CODEX_SANDBOX_NETWORK_DISABLED_ENV_VAR` environment variable.
// Ultimately, we should be stricter about the environment variables that
// are set for the command (as we are when spawning an MCP server), so
// instead of SandboxPolicy, we should take the exact env to use for the
// Command (i.e., `env_clear().envs(env)`).
if $command.is_empty() {
return Err(io::Error::new(
io::ErrorKind::InvalidInput,
"command args are empty",
));
}
// Do not create a file descriptor for stdin because otherwise some
// commands may hang forever waiting for input. For example, ripgrep has
// a heuristic where it may try to read from stdin as explained here:
// https://github.com/BurntSushi/ripgrep/blob/e2362d4d5185d02fa857bf381e7bd52e66fafc73/crates/core/flags/hiargs.rs#L1101-L1103
cmd.stdin(Stdio::null());
let mut cmd = <$cmd_type>::new(&$command[0]);
cmd.args(&$command[1..]);
cmd.current_dir($cwd);
cmd.stdout(Stdio::piped())
.stderr(Stdio::piped())
.kill_on_drop(true)
.spawn()
if !$sandbox_policy.has_full_network_access() {
cmd.env(CODEX_SANDBOX_NETWORK_DISABLED_ENV_VAR, "1");
}
match $stdio_policy {
StdioPolicy::RedirectForShellTool => {
// Do not create a file descriptor for stdin because otherwise some
// commands may hang forever waiting for input. For example, ripgrep has
// a heuristic where it may try to read from stdin as explained here:
// https://github.com/BurntSushi/ripgrep/blob/e2362d4d5185d02fa857bf381e7bd52e66fafc73/crates/core/flags/hiargs.rs#L1101-L1103
cmd.stdin(Stdio::null());
cmd.stdout(Stdio::piped()).stderr(Stdio::piped());
}
StdioPolicy::Inherit => {
// Inherit stdin, stdout, and stderr from the parent process.
cmd.stdin(Stdio::inherit())
.stdout(Stdio::inherit())
.stderr(Stdio::inherit());
}
}
std::io::Result::<$cmd_type>::Ok(cmd)
}};
}
/// Spawns the appropriate child process for the ExecParams and SandboxPolicy,
/// ensuring the args and environment variables used to create the `Command`
/// (and `Child`) honor the configuration.
pub(crate) async fn spawn_child_async(
command: Vec<String>,
cwd: PathBuf,
sandbox_policy: &SandboxPolicy,
stdio_policy: StdioPolicy,
) -> std::io::Result<Child> {
let mut cmd = configure_command!(Command, command, cwd, sandbox_policy, stdio_policy)?;
cmd.kill_on_drop(true).spawn()
}
/// Alternative version of `spawn_child_async()` that returns
/// `std::process::Child` instead of `tokio::process::Child`. This is useful for
/// spawning a child process in a thread that is not running a Tokio runtime.
pub fn spawn_child_sync(
command: Vec<String>,
cwd: PathBuf,
sandbox_policy: &SandboxPolicy,
stdio_policy: StdioPolicy,
) -> std::io::Result<std::process::Child> {
let mut cmd = configure_command!(
std::process::Command,
command,
cwd,
sandbox_policy,
stdio_policy
)?;
cmd.spawn()
}
/// Consumes the output of a child process, truncating it so it is suitable for
/// use as the output of a `shell` tool call. Also enforces specified timeout.
async fn consume_truncated_output(
pub(crate) async fn consume_truncated_output(
mut child: Child,
ctrl_c: Arc<Notify>,
timeout_ms: Option<u64>,
) -> Result<RawExecToolCallOutput> {
// Both stdout and stderr were configured with `Stdio::piped()`
// above, therefore `take()` should normally return `Some`. If it doesn't
// we treat it as an exceptional I/O error
let stdout_reader = child.stdout.take().ok_or_else(|| {
CodexErr::Io(io::Error::other(
"stdout pipe was unexpectedly not available",
))
})?;
let stderr_reader = child.stderr.take().ok_or_else(|| {
CodexErr::Io(io::Error::other(
"stderr pipe was unexpectedly not available",
))
})?;
let stdout_handle = tokio::spawn(read_capped(
BufReader::new(child.stdout.take().expect("stdout is not piped")),
BufReader::new(stdout_reader),
MAX_STREAM_OUTPUT,
MAX_STREAM_OUTPUT_LINES,
));
let stderr_handle = tokio::spawn(read_capped(
BufReader::new(child.stderr.take().expect("stderr is not piped")),
BufReader::new(stderr_reader),
MAX_STREAM_OUTPUT,
MAX_STREAM_OUTPUT_LINES,
));

View File

@@ -0,0 +1,77 @@
use std::io;
use std::path::Path;
use std::sync::Arc;
use crate::error::CodexErr;
use crate::error::Result;
use crate::exec::ExecParams;
use crate::exec::RawExecToolCallOutput;
use crate::exec::StdioPolicy;
use crate::exec::consume_truncated_output;
use crate::exec::spawn_child_async;
use crate::protocol::SandboxPolicy;
use tokio::sync::Notify;
pub fn exec_linux(
params: ExecParams,
ctrl_c: Arc<Notify>,
sandbox_policy: &SandboxPolicy,
) -> Result<RawExecToolCallOutput> {
// Allow READ on /
// Allow WRITE on /dev/null
let ctrl_c_copy = ctrl_c.clone();
let sandbox_policy = sandbox_policy.clone();
// Isolate thread to run the sandbox from
let tool_call_output = std::thread::spawn(move || {
let rt = tokio::runtime::Builder::new_current_thread()
.enable_all()
.build()?;
rt.block_on(async {
let ExecParams {
command,
cwd,
timeout_ms,
} = params;
apply_sandbox_policy_to_current_thread(&sandbox_policy, &cwd)?;
let child = spawn_child_async(
command,
cwd,
&sandbox_policy,
StdioPolicy::RedirectForShellTool,
)
.await?;
consume_truncated_output(child, ctrl_c_copy, timeout_ms).await
})
})
.join();
match tool_call_output {
Ok(Ok(output)) => Ok(output),
Ok(Err(e)) => Err(e),
Err(e) => Err(CodexErr::Io(io::Error::other(format!(
"thread join failed: {e:?}"
)))),
}
}
#[cfg(target_os = "linux")]
pub fn apply_sandbox_policy_to_current_thread(
sandbox_policy: &SandboxPolicy,
cwd: &Path,
) -> Result<()> {
crate::landlock::apply_sandbox_policy_to_current_thread(sandbox_policy, cwd)
}
#[cfg(not(target_os = "linux"))]
pub fn apply_sandbox_policy_to_current_thread(
_sandbox_policy: &SandboxPolicy,
_cwd: &Path,
) -> Result<()> {
Err(CodexErr::Io(io::Error::new(
io::ErrorKind::InvalidInput,
"linux sandbox is not supported on this platform",
)))
}

View File

@@ -3,7 +3,7 @@ use std::time::Duration;
use env_flags::env_flags;
env_flags! {
pub OPENAI_DEFAULT_MODEL: &str = "o3";
pub OPENAI_DEFAULT_MODEL: &str = "codex-mini-latest";
pub OPENAI_API_BASE: &str = "https://api.openai.com/v1";
/// Fallback when the provider-specific key is not set.

View File

@@ -75,6 +75,7 @@ fn is_safe_to_call_with_exec(command: &[String]) -> bool {
fn try_parse_bash(bash_lc_arg: &str) -> Option<Tree> {
let lang = BASH.into();
let mut parser = Parser::new();
#[expect(clippy::expect_used)]
parser.set_language(&lang).expect("load bash grammar");
let old_tree: Option<&Tree> = None;

View File

@@ -1,15 +1,10 @@
use std::collections::BTreeMap;
use std::io;
use std::path::Path;
use std::path::PathBuf;
use std::sync::Arc;
use crate::error::CodexErr;
use crate::error::Result;
use crate::error::SandboxErr;
use crate::exec::ExecParams;
use crate::exec::RawExecToolCallOutput;
use crate::exec::exec;
use crate::protocol::SandboxPolicy;
use landlock::ABI;
@@ -29,46 +24,11 @@ use seccompiler::SeccompFilter;
use seccompiler::SeccompRule;
use seccompiler::TargetArch;
use seccompiler::apply_filter;
use tokio::sync::Notify;
pub async fn exec_linux(
params: ExecParams,
ctrl_c: Arc<Notify>,
sandbox_policy: &SandboxPolicy,
) -> Result<RawExecToolCallOutput> {
// Allow READ on /
// Allow WRITE on /dev/null
let ctrl_c_copy = ctrl_c.clone();
let sandbox_policy = sandbox_policy.clone();
// Isolate thread to run the sandbox from
let tool_call_output = std::thread::spawn(move || {
let rt = tokio::runtime::Builder::new_current_thread()
.enable_all()
.build()
.expect("Failed to create runtime");
rt.block_on(async {
apply_sandbox_policy_to_current_thread(sandbox_policy, &params.cwd)?;
exec(params, ctrl_c_copy).await
})
})
.join();
match tool_call_output {
Ok(Ok(output)) => Ok(output),
Ok(Err(e)) => Err(e),
Err(e) => Err(CodexErr::Io(io::Error::new(
io::ErrorKind::Other,
format!("thread join failed: {e:?}"),
))),
}
}
/// Apply sandbox policies inside this thread so only the child inherits
/// them, not the entire CLI process.
pub fn apply_sandbox_policy_to_current_thread(
sandbox_policy: SandboxPolicy,
pub(crate) fn apply_sandbox_policy_to_current_thread(
sandbox_policy: &SandboxPolicy,
cwd: &Path,
) -> Result<()> {
if !sandbox_policy.has_full_network_access() {
@@ -180,7 +140,7 @@ fn install_network_seccomp_filter_on_current_thread() -> std::result::Result<(),
#[cfg(test)]
mod tests {
#![allow(clippy::unwrap_used)]
#![expect(clippy::unwrap_used, clippy::expect_used)]
use super::*;
use crate::exec::ExecParams;
@@ -234,7 +194,14 @@ mod tests {
#[tokio::test]
async fn test_dev_null_write() {
run_cmd(&["echo", "blah", ">", "/dev/null"], &[], 200).await;
run_cmd(
&["bash", "-lc", "echo blah > /dev/null"],
&[],
// We have seen timeouts when running this test in CI on GitHub,
// so we are using a generous timeout until we can diagnose further.
1_000,
)
.await;
}
#[tokio::test]

View File

@@ -1,32 +1,35 @@
//! Root of the `codex-core` library.
// Prevent accidental direct writes to stdout/stderr in library code. All
// uservisible output must go through the appropriate abstraction (e.g.,
// user-visible output must go through the appropriate abstraction (e.g.,
// the TUI or the tracing stack).
#![deny(clippy::print_stdout, clippy::print_stderr)]
mod chat_completions;
mod client;
mod client_common;
pub mod codex;
pub use codex::Codex;
pub mod codex_wrapper;
pub mod config;
pub mod config_profile;
pub mod config_types;
mod conversation_history;
pub mod error;
pub mod exec;
pub mod exec_linux;
mod flags;
mod is_safe_command;
#[cfg(target_os = "linux")]
pub mod linux;
pub mod landlock;
mod mcp_connection_manager;
pub mod mcp_server_config;
mod mcp_tool_call;
mod message_history;
mod model_provider_info;
pub use model_provider_info::ModelProviderInfo;
pub use model_provider_info::WireApi;
mod models;
mod project_doc;
pub mod protocol;
mod rollout;
mod safety;

View File

@@ -13,11 +13,13 @@ use anyhow::Context;
use anyhow::Result;
use anyhow::anyhow;
use codex_mcp_client::McpClient;
use mcp_types::ClientCapabilities;
use mcp_types::Implementation;
use mcp_types::Tool;
use tokio::task::JoinSet;
use tracing::info;
use crate::mcp_server_config::McpServerConfig;
use crate::config_types::McpServerConfig;
/// Delimiter used to separate the server name from the tool name in a fully
/// qualified tool name.
@@ -83,7 +85,33 @@ impl McpConnectionManager {
join_set.spawn(async move {
let McpServerConfig { command, args, env } = cfg;
let client_res = McpClient::new_stdio_client(command, args, env).await;
(server_name, client_res)
match client_res {
Ok(client) => {
// Initialize the client.
let params = mcp_types::InitializeRequestParams {
capabilities: ClientCapabilities {
experimental: None,
roots: None,
sampling: None,
},
client_info: Implementation {
name: "codex-mcp-client".to_owned(),
version: env!("CARGO_PKG_VERSION").to_owned(),
},
protocol_version: mcp_types::MCP_SCHEMA_VERSION.to_owned(),
};
let initialize_notification_params = None;
let timeout = Some(Duration::from_secs(10));
match client
.initialize(params, initialize_notification_params, timeout)
.await
{
Ok(_response) => (server_name, Ok(client)),
Err(e) => (server_name, Err(e)),
}
}
Err(e) => (server_name, Err(e.into())),
}
});
}
@@ -99,7 +127,7 @@ impl McpConnectionManager {
clients.insert(server_name, std::sync::Arc::new(client));
}
Err(e) => {
errors.insert(server_name, e.into());
errors.insert(server_name, e);
}
}
}

View File

@@ -1,14 +0,0 @@
use std::collections::HashMap;
use serde::Deserialize;
#[derive(Deserialize, Debug, Clone)]
pub struct McpServerConfig {
pub command: String,
#[serde(default)]
pub args: Vec<String>,
#[serde(default)]
pub env: Option<HashMap<String, String>>,
}

View File

@@ -7,6 +7,8 @@ use crate::models::FunctionCallOutputPayload;
use crate::models::ResponseInputItem;
use crate::protocol::Event;
use crate::protocol::EventMsg;
use crate::protocol::McpToolCallBeginEvent;
use crate::protocol::McpToolCallEndEvent;
/// Handles the specified tool call dispatches the appropriate
/// `McpToolCallBegin` and `McpToolCallEnd` events to the `Session`.
@@ -39,12 +41,12 @@ pub(crate) async fn handle_mcp_tool_call(
}
};
let tool_call_begin_event = EventMsg::McpToolCallBegin {
let tool_call_begin_event = EventMsg::McpToolCallBegin(McpToolCallBeginEvent {
call_id: call_id.clone(),
server: server.clone(),
tool: tool_name.clone(),
arguments: arguments_value.clone(),
};
});
notify_mcp_tool_call_event(sess, sub_id, tool_call_begin_event).await;
// Perform the tool call.
@@ -53,29 +55,29 @@ pub(crate) async fn handle_mcp_tool_call(
.await
{
Ok(result) => (
EventMsg::McpToolCallEnd {
EventMsg::McpToolCallEnd(McpToolCallEndEvent {
call_id,
success: !result.is_error.unwrap_or(false),
result: Some(result),
},
}),
None,
),
Err(e) => (
EventMsg::McpToolCallEnd {
EventMsg::McpToolCallEnd(McpToolCallEndEvent {
call_id,
success: false,
result: None,
},
}),
Some(e),
),
};
notify_mcp_tool_call_event(sess, sub_id, tool_call_end_event.clone()).await;
let EventMsg::McpToolCallEnd {
let EventMsg::McpToolCallEnd(McpToolCallEndEvent {
call_id,
success,
result,
} = tool_call_end_event
}) = tool_call_end_event
else {
unimplemented!("unexpected event type");
};

View File

@@ -0,0 +1,297 @@
//! Persistence layer for the global, append-only *message history* file.
//!
//! The history is stored at `~/.codex/history.jsonl` with **one JSON object per
//! line** so that it can be efficiently appended to and parsed with standard
//! JSON-Lines tooling. Each record has the following schema:
//!
//! ````text
//! {"session_id":"<uuid>","ts":<unix_seconds>,"text":"<message>"}
//! ````
//!
//! To minimise the chance of interleaved writes when multiple processes are
//! appending concurrently, callers should *prepare the full line* (record +
//! trailing `\n`) and write it with a **single `write(2)` system call** while
//! the file descriptor is opened with the `O_APPEND` flag. POSIX guarantees
//! that writes up to `PIPE_BUF` bytes are atomic in that case.
use std::fs::File;
use std::fs::OpenOptions;
use std::io::Result;
use std::io::Write;
use std::path::PathBuf;
use serde::Deserialize;
use serde::Serialize;
use std::time::Duration;
use tokio::fs;
use tokio::io::AsyncReadExt;
use uuid::Uuid;
use crate::config::Config;
use crate::config_types::HistoryPersistence;
#[cfg(unix)]
use std::os::unix::fs::OpenOptionsExt;
#[cfg(unix)]
use std::os::unix::fs::PermissionsExt;
/// Filename that stores the message history inside `~/.codex`.
const HISTORY_FILENAME: &str = "history.jsonl";
const MAX_RETRIES: usize = 10;
const RETRY_SLEEP: Duration = Duration::from_millis(100);
#[derive(Serialize, Deserialize, Debug, Clone)]
pub struct HistoryEntry {
pub session_id: String,
pub ts: u64,
pub text: String,
}
fn history_filepath(config: &Config) -> PathBuf {
let mut path = config.codex_home.clone();
path.push(HISTORY_FILENAME);
path
}
/// Append a `text` entry associated with `session_id` to the history file. Uses
/// advisory file locking to ensure that concurrent writes do not interleave,
/// which entails a small amount of blocking I/O internally.
pub(crate) async fn append_entry(text: &str, session_id: &Uuid, config: &Config) -> Result<()> {
match config.history.persistence {
HistoryPersistence::SaveAll => {
// Save everything: proceed.
}
HistoryPersistence::None => {
// No history persistence requested.
return Ok(());
}
}
// TODO: check `text` for sensitive patterns
// Resolve `~/.codex/history.jsonl` and ensure the parent directory exists.
let path = history_filepath(config);
if let Some(parent) = path.parent() {
tokio::fs::create_dir_all(parent).await?;
}
// Compute timestamp (seconds since the Unix epoch).
let ts = std::time::SystemTime::now()
.duration_since(std::time::UNIX_EPOCH)
.map_err(|e| std::io::Error::other(format!("system clock before Unix epoch: {e}")))?
.as_secs();
// Construct the JSON line first so we can write it in a single syscall.
let entry = HistoryEntry {
session_id: session_id.to_string(),
ts,
text: text.to_string(),
};
let mut line = serde_json::to_string(&entry)
.map_err(|e| std::io::Error::other(format!("failed to serialise history entry: {e}")))?;
line.push('\n');
// Open in append-only mode.
let mut options = OpenOptions::new();
options.append(true).read(true).create(true);
#[cfg(unix)]
{
options.mode(0o600);
}
let mut history_file = options.open(&path)?;
// Ensure permissions.
ensure_owner_only_permissions(&history_file).await?;
// Lock file.
acquire_exclusive_lock_with_retry(&history_file).await?;
// We use sync I/O with spawn_blocking() because we are using a
// [`std::fs::File`] instead of a [`tokio::fs::File`] to leverage an
// advisory file locking API that is not available in the async API.
tokio::task::spawn_blocking(move || -> Result<()> {
history_file.write_all(line.as_bytes())?;
history_file.flush()?;
Ok(())
})
.await??;
Ok(())
}
/// Attempt to acquire an exclusive advisory lock on `file`, retrying up to 10
/// times if the lock is currently held by another process. This prevents a
/// potential indefinite wait while still giving other writers some time to
/// finish their operation.
async fn acquire_exclusive_lock_with_retry(file: &std::fs::File) -> Result<()> {
use tokio::time::sleep;
for _ in 0..MAX_RETRIES {
match fs2::FileExt::try_lock_exclusive(file) {
Ok(()) => return Ok(()),
Err(e) if e.kind() == std::io::ErrorKind::WouldBlock => {
sleep(RETRY_SLEEP).await;
}
Err(e) => return Err(e),
}
}
Err(std::io::Error::new(
std::io::ErrorKind::WouldBlock,
"could not acquire exclusive lock on history file after multiple attempts",
))
}
/// Asynchronously fetch the history file's *identifier* (inode on Unix) and
/// the current number of entries by counting newline characters.
pub(crate) async fn history_metadata(config: &Config) -> (u64, usize) {
let path = history_filepath(config);
#[cfg(unix)]
let log_id = {
use std::os::unix::fs::MetadataExt;
// Obtain metadata (async) to get the identifier.
let meta = match fs::metadata(&path).await {
Ok(m) => m,
Err(e) if e.kind() == std::io::ErrorKind::NotFound => return (0, 0),
Err(_) => return (0, 0),
};
meta.ino()
};
#[cfg(not(unix))]
let log_id = 0u64;
// Open the file.
let mut file = match fs::File::open(&path).await {
Ok(f) => f,
Err(_) => return (log_id, 0),
};
// Count newline bytes.
let mut buf = [0u8; 8192];
let mut count = 0usize;
loop {
match file.read(&mut buf).await {
Ok(0) => break,
Ok(n) => {
count += buf[..n].iter().filter(|&&b| b == b'\n').count();
}
Err(_) => return (log_id, 0),
}
}
(log_id, count)
}
/// Given a `log_id` (on Unix this is the file's inode number) and a zero-based
/// `offset`, return the corresponding `HistoryEntry` if the identifier matches
/// the current history file **and** the requested offset exists. Any I/O or
/// parsing errors are logged and result in `None`.
///
/// Note this function is not async because it uses a sync advisory file
/// locking API.
#[cfg(unix)]
pub(crate) fn lookup(log_id: u64, offset: usize, config: &Config) -> Option<HistoryEntry> {
use std::io::BufRead;
use std::io::BufReader;
use std::os::unix::fs::MetadataExt;
let path = history_filepath(config);
let file: File = match OpenOptions::new().read(true).open(&path) {
Ok(f) => f,
Err(e) => {
tracing::warn!(error = %e, "failed to open history file");
return None;
}
};
let metadata = match file.metadata() {
Ok(m) => m,
Err(e) => {
tracing::warn!(error = %e, "failed to stat history file");
return None;
}
};
if metadata.ino() != log_id {
return None;
}
// Open & lock file for reading.
if let Err(e) = acquire_shared_lock_with_retry(&file) {
tracing::warn!(error = %e, "failed to acquire shared lock on history file");
return None;
}
let reader = BufReader::new(&file);
for (idx, line_res) in reader.lines().enumerate() {
let line = match line_res {
Ok(l) => l,
Err(e) => {
tracing::warn!(error = %e, "failed to read line from history file");
return None;
}
};
if idx == offset {
match serde_json::from_str::<HistoryEntry>(&line) {
Ok(entry) => return Some(entry),
Err(e) => {
tracing::warn!(error = %e, "failed to parse history entry");
return None;
}
}
}
}
None
}
/// Fallback stub for non-Unix systems: currently always returns `None`.
#[cfg(not(unix))]
pub(crate) fn lookup(log_id: u64, offset: usize, config: &Config) -> Option<HistoryEntry> {
let _ = (log_id, offset, config);
None
}
#[cfg(unix)]
fn acquire_shared_lock_with_retry(file: &File) -> Result<()> {
for _ in 0..MAX_RETRIES {
match fs2::FileExt::try_lock_shared(file) {
Ok(()) => return Ok(()),
Err(e) if e.kind() == std::io::ErrorKind::WouldBlock => {
std::thread::sleep(RETRY_SLEEP);
}
Err(e) => return Err(e),
}
}
Err(std::io::Error::new(
std::io::ErrorKind::WouldBlock,
"could not acquire shared lock on history file after multiple attempts",
))
}
/// On Unix systems ensure the file permissions are `0o600` (rw-------). If the
/// permissions cannot be changed the error is propagated to the caller.
#[cfg(unix)]
async fn ensure_owner_only_permissions(file: &File) -> Result<()> {
let metadata = file.metadata()?;
let current_mode = metadata.permissions().mode() & 0o777;
if current_mode != 0o600 {
let mut perms = metadata.permissions();
perms.set_mode(0o600);
let perms_clone = perms.clone();
let file_clone = file.try_clone()?;
tokio::task::spawn_blocking(move || file_clone.set_permissions(perms_clone)).await??;
}
Ok(())
}
#[cfg(not(unix))]
async fn ensure_owner_only_permissions(_file: &File) -> Result<()> {
// For now, on non-Unix, simply succeed.
Ok(())
}

View File

@@ -29,7 +29,7 @@ pub enum WireApi {
}
/// Serializable representation of a provider definition.
#[derive(Debug, Clone, Deserialize, Serialize)]
#[derive(Debug, Clone, Deserialize, Serialize, PartialEq)]
pub struct ModelProviderInfo {
/// Friendly display name.
pub name: String,

Some files were not shown because too many files have changed in this diff Show More