Repository navigation
Report the session's worktree and borrow on the status wire - #745
Conversation
repo_path is resolved on the daemon from the worktree's source checkout, so a borrowed reviewer and the primary whose worktree it borrowed report the same repository; the app's own resolution serves only an older daemon. Consent-prompt and activity labels show a launch request's own checkout leaf, since that wire carries no repository behind it. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
|
ⓘ Qodo reviews are paused because your trial has ended. Ask your workspace admin to add credits to resume reviews. Manage billing |
Codex Review SummaryThis comment shows the latest Codex review activity on this pull request.
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. |
![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: 3410dc72a3
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".
| .ThenBy(a => a.Id, StringComparer.Ordinal) | ||
| .Select(a => new AgentStatusDto( | ||
| a.Id, KindText(a.Kind), a.Vendor, a.RepoPath, a.Status, | ||
| a.Id, KindText(a.Kind), a.Vendor, a.Checkout.Repository, a.Status, |
There was a problem hiding this comment.
When an older desktop client connects to this upgraded daemon, it ignores the new trailing checkout fields but still interprets repo_path as the session's checkout. Replacing it with Checkout.Repository therefore moves borrowed reviewers from their borrowed-worktree rail node into the main-checkout node; version skew is allowed when the user declines the takeover prompt. Preserve the existing repo_path meaning and add a separate repository field, or negotiate the semantic change before sending it.
Useful? React with 👍 / 👎.
There was a problem hiding this comment.
Keeping repo_path as the repository is the point of this change (AI-2320): the field carried two meanings, a primary's repository and a reviewer's borrowed checkout, and the new trailing members are what say which checkout is which. A second repository field would leave repo_path ambiguous for good.
Under version skew the effect on an older desktop app is not a regression. Before, it filed a primary under "main checkout" and its borrowed reviewer one group away under agent-…; against this daemon it files both under "main checkout", mislabeled but together. Tray, card and Home labels resolve to the repository name either way. The one degrade is the older app's chat relativizing tool paths against repo_path: for a reviewer borrowing a daemon worktree the older ToolDetail still strips the .capacitor/worktrees/<id>/ segment, so only a reviewer borrowing a .claude/worktrees checkout shows that prefix, and only until the app updates. Nothing in the CLI reads repo_path from this payload; the agent list and server registration use the instance's own path, which is unchanged.
CheckoutLabel picks the checkout a session is presented under, the one a reviewer borrowed first, so the rail and the workspace subtitle cannot drift. The chat relativizes tool paths against worktree_path, the checkout the agent runs in, which for a snapshot reviewer is its private copy. Against an older daemon both fall back to RepoPath: the repository for a primary, the borrowed checkout for a reviewer. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
4541a6b to
96a3103
Compare
1 participant
AI-2320 — no GitHub issue exists for this one; the Linear issue is the only tracker entry.
What & why
repo_pathcarried two conventions — a primary's repository, a borrowed reviewer's worktree — so the rail filed a reviewer one group away from the session it reviews, with nothing on the wire saying they share a checkout. Three trailing nullable members now say which is which:worktree_path,work_location,borrowed_from, andrepo_pathis the repository for every agent, resolved on the daemon. The rail keys worktree nodes onborrowed_fromfirst, so a snapshot reviewer files under the checkout it reviews, and the header subtitle readseventuous / agent-6da28a8cb8e94a · borrowed.Where to look
AgentCheckout.Resolveis the one source of the three values.RepoLabel.Leafdrops the.claude/.capacitortail guess; the consent prompt and activity log label a launch request's path, which carries no repository, so they show that checkout's own leaf.Verification
dotnet publish -c Release: no IL2026/IL3050.🤖 Generated with Claude Code