[melodic-automation-lanes]

Closes #7332

Summary

Opened by the work-resolve-issue lane. The run's own summary:

/source-control:worktree cleanup now takes \[<path>...\], running the same classification and guards on only those worktrees, and a bare cleanup asks session-worktree vs whole repository when the session worked one. Step 4c runs branch deletion and current-worktree removal inline on Linux/macOS after confirmation and a passed stranded-work guard, emitting them on Windows or a hook block.

Not verified

• plugins/source-control/evals/worktree-cleanup-emits-branch-delete/ (plugin eval case) not edited or run; its prompt has no confirmation so it should still hold, but eval runs are out of scope per the issue
• No evals added for the new scoped-path behavior; eval runs were not performed
• check-skill on babysit-prs could not pass here (read-only ~/.claude); untouched by this change
• new-changelog-fragment could not run on the detached HEAD, so the fragment was written by hand and validated with check-changelog-fragments

Fix

• plugins/source-control/skills/worktree/SKILL.md: action table, headline, scope paragraph, Step 4c invariant and non-goals line updated
• plugins/source-control/skills/worktree/context/cleanup.md: new Scope section; Step 4c rewritten (inline on POSIX, emit on Windows/hook block)
• plugins/source-control/skills/worktree/evals/evals.json: cleanup eval case reworded to the Windows/hook-block contract
• .changes/source-control/worktree-cleanup-scope-7332-3f9a1c7e.md: minor fragment

Verification

The run reports:

• check-changelog-fragments: all 143 fragments valid
• check-skill plugins/source-control/skills: worktree PASS, 0 errors (6 warnings, all pre-existing size/style); babysit-prs failed only on a read-only filesystem write to ~/.claude in the sandbox, unrelated to this change
• fresh-context subagent review of the diff: no guard weakened; found stale eval wording and a Step 4b/4c ambiguity, both fixed
• The lane run: https://github.com/melodic-software/claude-code-plugins/actions/runs/39105808952

Left out: the items under Not verified.

Related

#7332

[github-actions]

claude-security-review reviewed 68a527f046aa562fd6ec171dd68ecd4857c4b5e5: 0 finding(s), 0 as inline comments.

No security findings in the incremental change (186ca26..68a527f).

The change touches three files:

• plugins/source-control/skills/worktree/SKILL.md: the stranded-work invariant now reads "never run or emit its git branch -D" instead of "never emit". This widens the guardrail to cover a run as well as an emitted command. It deletes, narrows or softens no standing instruction, so the instruction-surface deletion lens does not apply.
• plugins/source-control/skills/worktree/evals/evals.json: the expected_output wording for the eval case matches the new rule.
• plugins/source-control/skills/worktree/reattach-slice.test.sh: the assert_gate phrase is updated to match the new SKILL.md text.

The diff adds no code paths, no injection surface and no handling of tokens or secrets.

[github-actions]

claude-review reviewed 68a527f046aa562fd6ec171dd68ecd4857c4b5e5: 0 finding(s), 0 as inline comments.

Incremental review of 186ca26..68a527f (3 files): no findings.

The change is one wording edit applied consistently in three places. The Step 4c invariant in plugins/source-control/skills/worktree/SKILL.md now reads "never run or emit its git branch -D". This matches the PR's move to running branch deletion inline on POSIX, so the guard now covers a branch deletion the skill runs as well as one it emits. The eval expected_output in evals.json (case 11) is reworded to match. The phrase assertion in reattach-slice.test.sh is updated to the new text.

• No stale copies: grep of plugins and docs found no remaining "never emit its" phrase, so no other test or document still asserts the old wording.
• Not run: I could not run reattach-slice.test.sh or validate the evals.json JSON here because the sandbox required approval for those commands. From reading the diff, the assertion string matches the new SKILL.md text exactly. Treat that as unverified until CI runs it.
• Closing issue source-control: worktree cleanup should take explicit paths and scope from the conversation #7332: this increment only tightens the safety invariant. It does not change the acceptance-criteria coverage reviewed at the earlier head.

[melodic-automation-lanes]

pr-unblock / address-feedback finished round 1 of 3 on 001be61.

[melodic-automation-lanes]

Unattended merge: risk MED at head 404090142ec8b594e9bb4efe1c98beca28d587f4. Only LOW merges without a human; this pull request waits for one.

MED findings:claude-review: 3 open

[melodic-automation-lanes]

The pr-unblock lane stopped on this pull request and needs a person: the merge stages a change under .github/workflows/, which the lane does not push.

• Reason: workflow-staged
• Resume key: d85604c14ec963898631910801574ec22c72cfd7 (the head the lane stopped on)

To resume, deal with the reason above, remove the needs-human label (and do-not-merge, if set), then dispatch the pr-unblock workflow with pr-number=8096 and activity=resolve-conflicts:

gh workflow run pr-unblock.yml -R melodic-software/claude-code-plugins -f pr-number=8096 -f activity=resolve-conflicts

[melodic-automation-lanes]

The pr-unblock lane stopped on this pull request and needs a person: the lane could not fix every failing check; its run log names each one and why.

• Reason: ci-fix-escalated
• Resume key: 404090142ec8b594e9bb4efe1c98beca28d587f4 (the head the lane stopped on)

To resume, deal with the reason above, remove the needs-human label (and do-not-merge, if set), then dispatch the pr-unblock workflow with pr-number=8096 and activity=fix-ci:

gh workflow run pr-unblock.yml -R melodic-software/claude-code-plugins -f pr-number=8096 -f activity=fix-ci