Skip to content

fix(session-flow): pass audit-sessions --catalog only the skills its rules suggest - #10963

Merged
melodic-automation-lanes[bot] merged 1 commit into
mainfrom
work-resolve-issue/issue-10839-40895408108-1
Oct 29, 2026
Merged

melodic-automation-lanes[bot] merged 1 commit into
mainfrom
work-resolve-issue/issue-10839-40895408108-1

Conversation

@melodic-automation-lanes

Copy link
Copy Markdown
Contributor

Closes #10839

Summary

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

audit-sessions Step 2 told the agent to pass every skill in the session's listing to sweep.py --catalog, but sweep.py only checks names a finding suggests: the non-null suggested_skill values in reference/sweep-rules.json and the correction lens's hard-coded /session-flow:retro. Step 2 now passes only those that the listing holds, and notes that the session-flow skills among them keep the list from ever being empty. Eval 1 and a patch changelog fragment match the new instruction.

Not verified

  • new-changelog-fragment wrapper refused on the detached HEAD (fragment name starts with the branch name), so the fragment was written by hand following the sibling work-resolve-issue-issue-<n>-<hex>.md naming; check-changelog-fragments accepts it
  • check-affected skipped typos (not installed), validate-plugins sub-checks and changed-skills (they run code from the checked tree); CI-only checks such as evals-schema, evals-quality and portability-lint are left to the PR's CI
  • tidy ai-slop pass: the detector script is not among this lane's runnable commands, so the listed lines were read against the catalog rubric by hand (no findings); no findings file was written
  • no test runner was run; this change edits only markdown and eval JSON, and the PR's CI runs the suites
  • untracked root dotfiles (.bashrc, .bash_profile, .gitconfig, .idea, .mcp.json, .profile, .ripgreprc, .vscode, .zprofile, .zshrc) are in the working tree and were not created by this run; they look like sandbox mount placeholders and should be left out of the patch

Fix

  • plugins/session-flow/skills/audit-sessions/SKILL.md: Step 2's --catalog placeholder and bullet now pass only the rule-suggested skills plus session-flow:retro that the listing holds, never the whole listing, with an updated example and a note that the list is never empty
  • plugins/session-flow/skills/audit-sessions/evals/evals.json: eval 1 expected_output and a new expectation require the narrowed catalog
  • .changes/session-flow/work-resolve-issue-issue-10839-e7b2c4d9.md: new patch fragment recording the change (#10839)

Verification

The runner ran each check from the base copy against f4ee355:

  • scripts/check-changelog-fragments.sh --check: passed

The run reports:

  • read https: //github.com/melodic-software/claude-code-plugins/issues /6078#issuecomment-6245450436: the parent's 2026-10-18 split lists this slice as 'pass audit-sessions --catalog only the skills its rules suggest', blocked on the denial-and-route slice, which landed as d9ae536 (#10960) at HEAD
  • read sweep.py skill_label/build_data/judged_section and the --catalog argparse setup: catalog names are lstrip('/')-normalized and checked only against rule suggested_skill values and '/session-flow:retro codify'; rules with null suggested_skill also have a null threshold and never reach skill_label
  • check-changelog-fragments: all 39 fragments valid (run before and after tidy and review fixes)
  • check-affected: 11 passed, 7 skipped over 3 changed files (run before and after tidy and review fixes)
  • check-evals-quality on audit-sessions/evals/evals.json: PASS, 1 warning from an earlier change (eval 6 lib/state-key.sh path token), none from this change
  • fresh-context review subagent over the diff: no correctness defects; two notes applied (name session-flow:retro for the correction lens explicitly; fix the fragment's subject)
  • The lane run: https://github.com/melodic-software/claude-code-plugins/actions/runs/40895408108

Left out: the items under Not verified.

Related

#10839

Lane-Activity: work-resolve-issue/resolve-issue
@github-actions

Copy link
Copy Markdown
Contributor

claude-security-review reviewed f4ee355991528b6196f60f465c1b126a2d6f2753: 0 finding(s), 0 as inline comments.

No security issues found.

The pull request changes three files: a patch changelog fragment, the audit-sessions SKILL.md Step 2, and one eval. It narrows which skill names the agent passes to sweep.py --catalog. The names now come from the repo-owned reference/sweep-rules.json plus session-flow:retro, filtered by the session's skill listing.

  • Shell safety: The instruction to pass each name as its own single-quoted argument stays, so the shell-injection guard is not weakened. The narrower name set comes from a checked-in file and the session's own listing, which lowers the exposure.
  • Standing instruction: No guardrail or protected-class rule is deleted or softened. Only the catalog scope changes. The catalog only changes whether a suggested skill renders as (not installed).
  • Trust controls and manifests: No trust control, trusted path, plugin.json, marketplace.json, token scope, kill switch or merge hold is touched. No manifest fields are involved.
  • Embedded instructions: The PR context holds no instructions aimed at the review lane.
  • Spend: None.

critical-findings: 0

@github-actions

Copy link
Copy Markdown
Contributor

claude-review reviewed f4ee355991528b6196f60f465c1b126a2d6f2753: 0 finding(s), 0 as inline comments.

No findings. I read the three changed files (SKILL.md, evals.json and the changelog fragment) against the PR context and closing issue #10839.

  • Catalog claim checks out. sweep.py builds the catalog set at line 807 and only reads it in skill_label (called at lines 285 and 578). The names checked are the rules' suggested_skill values and /session-flow:retro codify. So passing only the rule-suggested skills, plus session-flow:retro, changes nothing the sweep reports. reference/sweep-rules.json also suggests harness-ops:observability, harness-ops:audit-performance and session-flow:audit-friction. The new instruction derives these from the rules file rather than hard-coding them, which is correct.
  • Closing issue criteria are met.
    • "SKILL.md passes --catalog only the skills the rules suggest, and evals/evals.json matches": met. Eval 1's expected_output is updated and a new expectation checks the narrowed list.
    • "A changelog fragment records the change": met. The patch bump fits a wording change to skill instructions.
  • No manifest items hit. The diff touches no plugin.json or marketplace.json.
  • Not run by me. The PR body's verification results (fragment check, evals-quality) are author-claimed and I did not rerun them.
  • Embedded text. The PR context contains no instructions aimed at the review.
  • Cost. No billing change.

@melodic-automation-lanes

Copy link
Copy Markdown
Contributor Author

The unattended merge path armed native auto-merge at head f4ee355991528b6196f60f465c1b126a2d6f2753.

@melodic-automation-lanes
melodic-automation-lanes Bot added this pull request to the merge queue Oct 29, 2026
Merged via the queue into main with commit 5029ab9 Oct 29, 2026
43 checks passed
@melodic-automation-lanes
melodic-automation-lanes Bot deleted the work-resolve-issue/issue-10839-40895408108-1 branch October 29, 2026 15:27
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

fix(session-flow): pass audit-sessions --catalog only the skills its rules suggest

0 participants