Repository navigation
Conversation
…focus Step 1 called `herdr worktree create` with no source flag, so herdr branched off whichever workspace had focus. #209 proposed `--workspace "$HERDR_WORKSPACE_ID"`, but probing showed that picks the wrong repo too. herdr ties a workspace to the repo it opened in, so a pane that later cd'd elsewhere still branches off the old one. Every wrong case succeeded, and nothing told the run its worktree sat in another repo. Only `--cwd ROOT` picked the right repo. Step 1 now passes it and checks the response's `repo_root` against ROOT, stopping on a mismatch. Ledger row 192 records the probe. Closes #209
|
@codex review |
Codex Review Summary: ✅ CompletedThis 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: 59a477f806
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".
…lts on it Step 1's new `repo_root` check said "stop and name both paths" but printed no fixed message and had no row in Where a Run Stops. drive.py halts only on a quoted message from that table and relaunches anything else, so a driven run that hit the mismatch would start another agent instead of waiting for the user to remove the stray worktree (Codex review on #323). Step 1 now prints a fixed message, the table carries it as a new row, and the smoke suite's row count goes from 36 to 37.
|
@codex review |
|
Codex Review: Didn't find any major issues. 🎉 Reviewed commit: ℹ️ About Codex in GitHubYour team has set up Codex to review pull requests in this repo. Reviews are triggered when you
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". |
Step 1 now creates the herdr worktree with
--cwd ROOTand stops when the response'srepo_rootisn't ROOT. Before this change, a run started while another repo's workspace had focus could branch its worktree off that other repo, and herdr reported success.#209 proposed
--workspace "$HERDR_WORKSPACE_ID". I probed each form with focus on the long-term-memory-kit workspace before choosing:--workspaceon a workspace whose pane sits inskillsdeep-field, the repo that workspace opened in--workspaceon a pane in~--cwd ~/code/skillsskillsherdr binds a workspace to the repo it opened in and never updates the binding after a
cd, so--workspacefails the same way focus does. Every wrong case returnedworktree_created, which is why therepo_rootcheck is there. Ledger row 192 records the probe.Start with the Step 1 paragraph in
work-issue/SKILL.md. The other three files are a ledger row, a focus case added to EVALS scenario 11, and tworequire_textlines in the smoke suite.On a mismatch the run stops and leaves the stray worktree for the user to remove, because it sits in a repo the run doesn't own. That's the call I'm least sure of. An automatic
herdr workspace closeplusgit -C <repo_root> worktree removewould also work.Tested:
bash tests/work-issue-smoke.shpasses, and both new strings are absent frommain'sSKILL.md. I didn't run EVALS scenario 11 live.Closes #209