Adds a "View Stack" to the bottom pane to allow for pushing/popping
bottom panels.
`esc` will go back instead of dismissing.
Benefit: We retain the "selection state" of a parent panel (e.g. the
review panel).
Backtracking multiple times could drop earlier turns. We now derive the
active user-turn positions from the transcript on demand (keying off the
latest session header) instead of caching state. This keeps the replayed
context intact during repeated edits and adds a regression test.
The only file to watch is the cargo.toml
All the others come from just fix + a few manual small fix
The set of rules have been taken from the list of clippy rules
arbitrarily while trying to optimise the learning and style of the code
while limiting the loss of productivity
Adds the following options:
1. Review current changes
2. Review a specific commit
3. Review against a base branch (PR style)
4. Custom instructions
<img width="487" height="330" alt="Screenshot 2025-09-20 at 2 11 36 PM"
src="https://github.com/user-attachments/assets/edb0aaa5-5747-47fa-881f-cc4c4f7fe8bc"
/>
---
\+ Adds the following UI helpers:
1. Makes list selection searchable
2. Adds navigation to the bottom pane, so you could add a stack of
popups
3. Basic custom prompt view
We currently get information about rate limits in the response headers.
We want to forward them to the clients to have better transparency.
UI/UX plans have been discussed and this information is needed.
Currently, we change the tool description according to the sandbox
policy and approval policy. This breaks the cache when the user hits
`/approvals`. This PR does the following:
- Always use the shell with escalation parameter:
- removes `create_shell_tool_for_sandbox` and always uses unified tool
via `create_shell_tool`
- Reject the func call when the model uses escalation parameter when it
cannot.
### Why Use `tokio::sync::Mutex`
`std::sync::Mutex` are not _async-aware_. As a result, they will block
the entire thread instead of just yielding the task. Furthermore they
can be poisoned which is not the case of `tokio` Mutex.
This allows the Tokio runtime to continue running other tasks while
waiting for the lock, preventing deadlocks and performance bottlenecks.
In general, this is preferred in async environment
This change instructs the model to install any missing command. Else
tokens are wasted when it tries to run
commands that aren't available multiple times before installing them.
Often, `gh` infers `--repo` when it is run from a Git clone, but our
`publish-npm` step is designed to avoid the overhead of cloning the
repo, so add the `--repo` option explicitly to fix things.
The build for `v0.37.0-alpha.3` failed on the `Create GitHub Release`
step:
https://github.com/openai/codex/actions/runs/17786866086/job/50556513221
with:
```
⚠️ GitHub release failed with status: 403
{"message":"Resource not accessible by integration","documentation_url":"https://docs.github.com/rest/releases/releases#create-a-release","status":"403"}
Skip retry — your GitHub token/PAT does not have the required permission to create a release
```
I believe I should have not introduced a top-level `permissions` for the
workflow in https://github.com/openai/codex/pull/3431 because that
affected the `permissions` for each job in the workflow.
This PR introduces `publish-npm` as its own job, which allows us to:
- consolidate all the Node.js-related steps required for publishing
- limit the reach of the `id-token: write` permission
- skip it altogether if is an alpha build
With this PR, each of `release`, `publish-npm`, and `update-branch` has
an explicit `permissions` block.
Proposal: We want to record a dev message like so:
```
{
"type": "message",
"role": "user",
"content": [
{
"type": "input_text",
"text": "<user_action>
<context>User initiated a review task. Here's the full review output from reviewer model. User may select one or more comments to resolve.</context>
<action>review</action>
<results>
{findings_str}
</results>
</user_action>"
}
]
},
```
Without showing in the chat transcript.
Rough idea, but it fixes issue where the user finishes a review thread,
and asks the parent "fix the rest of the review issues" thinking that
the parent knows about it.
### Question: Why not a tool call?
Because the agent didn't make the call, it was a human. + we haven't
implemented sub-agents yet, and we'll need to think about the way we
represent these human-led tool calls for the agent.
1. Adds the environment prompt (including cwd) to review thread
2. Prepends the review prompt as a user message (temporary fix so the
instructions are not replaced on backend)
3. Sets reasoning to low
4. Sets default review model to `gpt-5-codex`
## Summary
SendUserTurn has not been correctly handling updates to policies. While
the tui protocol handles this in `Op::OverrideTurnContext`, the
SendUserTurn should be appending `EnvironmentContext` messages when the
sandbox settings change. MCP client behavior should match the cli
behavior, so we update `SendUserTurn` message to match.
## Testing
- [x] Added prompt caching tests
