Conversation
Review the commits merged to LibreChat dev since the last synced commit, let Claude document what needs it, and open or refresh one draft pull request. The sync marker in .github/docs-sync/state.json only advances when that pull request merges. The agent reads untrusted text, so it can only edit content/docs and its report, only run read-only git commands, and never sees a stored token; the job fails if git config or hooks change, and refuses locale files, files outside content/docs, and dashes used as punctuation.
|
The latest updates on your projects. Learn more about Vercel for GitHub.
|
Codex Review SummaryThis comment shows the latest Codex review activity on this pull request.
Security findingsAdvisory findings (1)ℹ️ About Codex in GitHubYour team has set up Codex to review pull requests in this repo. Reviews are triggered when you
Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings. |
📦 Next.js Bundle Analysis for librechat.aiThis analysis was generated by the Next.js Bundle Analysis action. 🤖 This PR introduced no changes to the JavaScript bundle! 🙌 |
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: d3e7db6dc5
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
| git -C .sync/librechat log --reverse --no-merges \ | ||
| --format='%H %as %s' "${since}..origin/dev" | head -n "$MAX_COMMITS" > .sync/commits.txt |
There was a problem hiding this comment.
When the range crosses an upstream merge with new commits on both parents, --no-merges leaves incomparable parent-side commits in this file, while last later selects only one of them as the persisted marker. Git's A..B revision range excludes only commits reachable from A, so the next run sees the other parent, saves that SHA, and subsequent runs can alternate between the two sides indefinitely. I reproduced this with a two-parent merge; the workflow repeatedly reviews the same commits and never converges. Derive the checkpoint from a merge or first-parent boundary that contains the entire reviewed set rather than the final line of the filtered log.
Useful? React with 👍 / 👎.
| claude_args: | | ||
| --model ${{ inputs.model || 'claude-opus-5-5' }} | ||
| --max-turns 400 | ||
| --allowedTools "Read,Glob,Grep,Task,TodoWrite,Edit(./content/docs/**),Write(./content/docs/**),Edit(./.sync/report.md),Write(./.sync/report.md),Bash(git -C .sync/librechat show:*),Bash(git -C .sync/librechat log:*),Bash(git -C .sync/librechat grep:*),Bash(git -C .sync/librechat ls-tree:*),Bash(git -C .sync/librechat diff:*),Bash(git diff:*),Bash(git status:*),Bash(ls:*),Bash(wc:*)" |
There was a problem hiding this comment.
Because this step deliberately processes untrusted commit messages and source, an injected instruction can use the permitted Bash rule to run git diff --no-index /dev/null - < /proc/self/environ; --no-index accepts arbitrary inputs, and the shell redirection exposes its inherited environment, including ANTHROPIC_API_KEY, in the tool result. The model can then place the recovered value in the permitted .sync/report.md, which the later gh pr create or gh pr edit publishes as the PR body. persist-credentials: false protects the GitHub credential but not this API key, so run the agent tools with secrets scrubbed from their subprocess environment or in a sandbox that prevents access to the parent environment.
Useful? React with 👍 / 👎.
There was a problem hiding this comment.
🛡️ Codex Security Review · Automatically triggered
Here are some automated security review suggestions for this pull request.
Reviewed commit: d3e7db6dc5
ℹ️ About Codex security reviews in GitHub
This is an experimental Codex feature. Security reviews are triggered when:
- You comment "@codex security review"
- A regular code review gets triggered (for example, "@codex review" or when a PR is opened), and you’re opted in so security review runs alongside code review
Once complete, Codex will leave suggestions, or a comment if no findings are found.
| claude_args: | | ||
| --model ${{ inputs.model || 'claude-opus-5-5' }} | ||
| --max-turns 400 | ||
| --allowedTools "Read,Glob,Grep,Task,TodoWrite,Edit(./content/docs/**),Write(./content/docs/**),Edit(./.sync/report.md),Write(./.sync/report.md),Bash(git -C .sync/librechat show:*),Bash(git -C .sync/librechat log:*),Bash(git -C .sync/librechat grep:*),Bash(git -C .sync/librechat ls-tree:*),Bash(git -C .sync/librechat diff:*),Bash(git diff:*),Bash(git status:*),Bash(ls:*),Bash(wc:*)" |
There was a problem hiding this comment.
🛡️ Codex Security Review · Automatically triggered
When ANTHROPIC_API_KEY is configured, an attacker whose text reaches LibreChat dev can prompt-inject the scheduled agent into invoking git -C .sync/librechat grep --open-files-in-pager=<command> .... This matches the allowed git ... grep:* prefix, but Git executes the supplied pager as a shell command. It can persist BASH_ENV through $GITHUB_ENV; the later PR step starts Bash with GH_TOKEN (the optional PAT or contents/PR-write job token), enabling credential theft and repository writes. The workspace checks do not inspect runner environment files. Replace general Git Bash access with a fixed argument-validating wrapper.
Useful? React with 👍 / 👎.
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 8f7e2693fb
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
| git -C .sync/librechat log --reverse --no-merges \ | ||
| --format='%H %as %s' "${since}..origin/dev" | head -n "$MAX_COMMITS" > .sync/commits.txt |
There was a problem hiding this comment.
When a merge commit contains conflict-resolution or other manual edits not present in either parent, --no-merges removes the only commit whose diff exposes those changes. The agent therefore never triages them, and once a later non-merge commit advances the marker, the documentation change is permanently skipped; include nontrivial merge commits in the review input or separately inspect their merge-result diffs.
Useful? React with 👍 / 👎.
| exit 1 | ||
| fi | ||
| git -C .sync/librechat log --reverse --no-merges \ | ||
| --format='%H %as %s' "${since}..origin/dev" | head -n "$MAX_COMMITS" > .sync/commits.txt |
There was a problem hiding this comment.
When the pending range is large enough for head to close the pipe before git log finishes, the Actions default Bash shell runs with -o pipefail, so Git's resulting SIGPIPE status (141) makes this step fail instead of processing the requested batch. I reproduced the exact git log | head shape with more than 150 commits; because the marker is never advanced, every scheduled retry encounters the same backlog and fails again. The Bash pipeline semantics specify that pipefail returns the rightmost nonzero command status, so apply the limit within Git (for example with -n) rather than truncating its output through head.
Useful? React with 👍 / 👎.
📦 Next.js Bundle Analysis for librechat.aiThis analysis was generated by the Next.js Bundle Analysis action. 🤖 This PR introduced no changes to the JavaScript bundle! 🙌 |
This branch was successfully deployed
1 participant
Summary
Adds a weekly workflow that keeps the docs in step with LibreChat
dev. It lists the commits merged since the last synced commit, has Claude triage them, verify each change against the app source and write the missing English docs, then opens or refreshes a single draft PR with a report of what changed, what needed no docs, and what needs a human. Nothing merges on its own.The sync marker in
.github/docs-sync/state.jsonadvances only when that PR merges, so an unmerged run is regenerated with the wider range next time. It starts after the commit #795 was audited against. The instructions in.github/docs-sync/prompt.mdencode the conventions from #795 and assume the versioned docs from #770, so merge #770 first.Because the agent reads untrusted text, it may only edit
content/docsand its report, run read-only git commands, and never sees a stored token; the job fails if git config or hooks change, and rejects locale files, other paths, and dashes used as punctuation.Before it can run, a maintainer needs to add the
ANTHROPIC_API_KEYsecret and, so CI runs on the sync PR, an optionalDOCS_SYNC_TOKEN.Change Type
Testing
actionlint1.7.12: clean apart from a false positive on the action'sconclusionoutput, which exists inaction.ymlatv1workflow_dispatch)Checklist