Skip to content

docs(harness-config): route auto-mode-setup's denial-based allow rules into draft-auto-mode-rules - #18046

Open
melodic-automation-lanes[bot] wants to merge 2 commits into
mainfrom
work-resolve-issue/issue-17911-44949329593-1
Open

melodic-automation-lanes[bot] wants to merge 2 commits into
mainfrom
work-resolve-issue/issue-17911-44949329593-1

Conversation

@melodic-automation-lanes

@melodic-automation-lanes melodic-automation-lanes Bot commented Nov 21, 2026 •

Copy link
Copy Markdown
Contributor

Closes #17911

Summary

Opened by the work-resolve-issue lane. The run's own summary, with the review round's change to the reference file:

draft-auto-mode-rules' Routing section now offers the allow rules /auto-mode-setup drafts from recent auto mode denials as input to the allow section. The person pastes the rules they accepted, or the allow list from claude auto-mode config, and the interview asks for each rule's exclusions and transcript-visible condition. Choice and basis: the Phase 1 lint script reports findings, not entries, so the rules reach the interview through the person (pasted rules are interview input, keeping the skill's 'interview plus effective merge, nothing else' input rule and its report-only posture). The skill still reads no history itself; the Inputs paragraph now says so and names this path.

reference/native-auto-mode-setup.md records what the command reads and writes as our decision plus a section-anchored pointer to the auto mode config page, with no stamp. The only stored value is the Claude Code 2.1.329 changelog line on the denial scan, with its as-of date and recheck trigger. The review-mode difference between the changelog and the config page is recorded as a source conflict with both links, neither position restated.

Not verified

  • check-affected skipped typos (not installed), comment-hygiene (fetch of the ci-workflows action failed: no network), size-check (audit-size.sh missing in the lane), and three validate-plugins checks plus changed-skills that run code from the checked tree; the PR's CI runs them
  • tidy pass 5 (ai-slop:audit fix): the detector script cannot run in this lane, whose Bash runs only the named wrappers; the listed lines were read against the rubric by hand and nothing was found. Passes 1-3 had no code lines to act on
  • verify: a docs-only change with no build or test suite to run here; the skill check's script tests and markdownlint were skipped by the check's own settings and are left to the PR's CI
  • the work-items CI-mode instruction to commit each changed file was not followed, because the runner's instruction to leave every change uncommitted in the working tree takes precedence
  • untracked .bash_profile, .bashrc, .gitconfig, .idea, .mcp.json, .profile, .ripgreprc, .vscode, .zprofile and .zshrc show in git status at the repo root only inside the Bash sandbox (file tools find no such files); they are sandbox mount stubs, not part of this change

Fix

  • plugins/harness-config/skills/draft-auto-mode-rules/SKILL.md: Routing offers /auto-mode-setup's denial-based allow rules as input to the allow section; the Inputs omission paragraph says denial history reaches the draft only through those saved rules
  • plugins/harness-config/skills/draft-auto-mode-rules/reference/native-auto-mode-setup.md: pointer records for what the command reads, what it writes and the writer row (no restated upstream behavior, no stamp), one stored-value record for the 2.1.329 changelog line, a source-conflict row for the review mode, the availability row's stamp dropped, and the complementary-verdict paragraph worded as our decision
  • .changes/harness-config/work-resolve-issue-issue-17911-44949329593-1-731a9f15.md: new patch changelog fragment

Verification

The runner ran each check from the base copy against cf9ebdd:

  • scripts/check-changelog-fragments.sh --check: passed
  • scripts/check-changed-skills.sh --merge-base: passed

The run reports:

  • read https: //code.claude.com/docs/en/changelog: the 2.1.329 (November 19, 2026) bug fixes include 'Fixed /auto-mode-setup ignoring recent auto mode denials when it scans a project for rules to suggest; it may now offer allow rules from them, and you review each before it's saved' (a first…
  • read https: //code.claude.com/docs/en/auto-mode-config: 'What /auto-mode-setup reads' lists project CLAUDE.md, README.md, config files, git remotes, autoMode and permissions.allow settings, command hosts/buckets/names from recent sessions, plus two opt-in scans; 'Review and save the draft' says the…
  • read https: //code.claude.com/docs/en/commands: the /auto-mode-setup row still describes drafting autoMode.environment entries only, with plan and version requirements
  • check-changelog-fragments: all 21 fragments valid
  • check-changed-skills: draft-auto-mode-rules PASS, 0 errors, 4 warnings (description length and body size above advisory medians, body grew 180 -> 186 lines, a pre-existing stale fresh-eyes-exempt directive at SKILL.md:111)
  • check-affected: 14 passed, 9 skipped, run again after tidy and after the review fixes
  • fresh-context review subagent over the diff: no wrong versions or misquotes; 6 findings (Phase 1 cannot show the saved allow rules, missing permissions.allow cleanup write, unstated review-mode discrepancy, stale writer row, version in SKILL.md body, a leftover reflow line), all fixed
  • The lane run: https://github.com/melodic-software/claude-code-plugins/actions/runs/44949329593

Left out: the items under Not verified.

Related

#17911

Lane-Activity: work-resolve-issue/resolve-issue
@melodic-automation-lanes

Copy link
Copy Markdown
Contributor Author

Unattended merge: native auto-merge leaves this pull request unarmed at head cf9ebdde24e29b48e93c0e109cdb7e1608aa8b30. No person needs to act: a lane clears this reason, and the next scheduled run judges it again.

pr-merge's gate refuses its verifier verdict (verify:missing)

@github-actions

github-actions Bot commented Nov 21, 2026 •

Copy link
Copy Markdown
Contributor

claude-security-review reviewed 866279c08db26e578f072387574f451e580e10ab: read 1 of 1 changed file(s); 0 finding(s), 0 as inline comments.

Security review: PR #18046 (incremental since cf9ebdd)

I found no security issues. This was a blocking-only review because this lane already reviewed an earlier head, so it reports only CRITICAL and IMPORTANT findings.

Scope. I read the whole incremental change. It touches one file, plugins/harness-config/skills/draft-auto-mode-rules/reference/native-auto-mode-setup.md, which matches no trusted path.

What changed.

  • The reference file's rows about what the built-in /auto-mode-setup reads and writes no longer restate the upstream behavior with dates. Each row now records our decision and a pointer to fetch the matching docs section live.
  • The file adds one row noting a disagreement between the changelog and the config page. It keeps one stored changelog line with its date and recheck trigger.
  • The "Why the verdict is complementary" paragraph is reworded as our decision.

What I checked.

  • Weakened instructions: nothing removed is a guardrail. The change adds a stance that the command's inputs are "never a source this skill reads itself", which narrows what the skill reads. The skill still never writes, and it gains no file access, command execution or write path. No instruction-surface deletion applies, so I did not consult the instruction exception register.
  • Trust controls, plugin manifests and workflows: none are touched, so there are no manifest lines.
  • Embedded instructions: the PR context holds none aimed at this lane.

Cost: none.

critical-findings: 0

@github-actions github-actions Bot left a comment

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

claude-review: 1 inline finding(s) on cf9ebdde24e29b48e93c0e109cdb7e1608aa8b30; the summary is in this lane's pull request comment.

claude-lane: claude-review

@github-actions

github-actions Bot commented Nov 21, 2026 •

Copy link
Copy Markdown
Contributor

claude-review reviewed 866279c08db26e578f072387574f451e580e10ab: read 1 of 1 changed file(s); 0 finding(s), 0 as inline comments.

Code review: PR #18046 (incremental, cf9ebdde → 866279c0)

No blocking findings in this push. This lane already reviewed an earlier head, so this review reports blocking findings only.

What changed. One file changed: plugins/harness-config/skills/draft-auto-mode-rules/reference/native-auto-mode-setup.md. The push answers this lane's earlier finding that the reference rows restated upstream behavior:

  • Writer, reads, writes and availability rows. Each is now our decision plus a live-fetch pointer to a section anchor (commands#all-commands, auto-mode-config#what-auto-mode-setup-reads, auto-mode-config#generate-environment-entries), with empty As of and Recheck cells.
  • The 2.1.329 changelog line. It is the only stored value, and it keeps its date and recheck trigger.
  • The review-mode difference. It is now a separate row that only says the changelog and the config page disagree, with both links.
  • The intro and the "Why the verdict is complementary" paragraph. The intro now says which rows store a value. The paragraph is worded as our decision.

Against the rules. I checked the new rows against docs/contracts/upstream-drift/README.md "Required parts": a pointer record takes no stamp, and a source conflict is recorded only as "X and Y disagree on T". The rows now conform.

scripts/check-as-of-recheck.sh flags only an as-of date that has no trigger, so the empty As of and Recheck cells should pass it. I read the script; I did not run it.

Earlier finding. The thread for it is resolved, its fix commit is this head, and this push addresses it. I am not raising it again.

Closing issue #17911, "Done when". The reference file still records what the command reads and writes, now as pointers. It also records the version, 2.1.329, on the stored changelog row. The criterion is still met. These are prose criteria, so no test could guard them.

Not verified by me. The author says check-changelog-fragments.sh and check-changed-skills.sh passed. I had no shell, so those results are the author's claims.

Spend: none. The change adds no prepaid or billed usage.

Loosened checks: the change loosens no check. Loosened checks would be marked: fail (the default, because the wrapper wrote no loosened_check_verdict: or loosened_check_patterns: line); patterns: 0. With no decision-record line, I didn't check the citation requirement.

Paved paths: not checked. I did not resolve the standards index for this docs-only push, so I report nothing about paved paths.

I found no embedded directives in the PR context, the settled threads or the prior findings.

Verifier dropped 0 of 0 candidates.

@github-actions

Copy link
Copy Markdown
Contributor

Lane run report

Field Value
Activity pr-unblock/address-feedback

Run 44950889751, attempt 1

Field Value
Model rung 0
Leg handoff
Run job success
Activity step success
Commits pushed none (read lane)
Run 44950889751

Failed GitHub API calls

No failed call recorded by the run job's scripts.

Lane-Activity: pr-unblock/address-feedback
Lane-Thread: PRRT_kwDOTCGFQM66cfml
@github-actions

Copy link
Copy Markdown
Contributor

Lane run report

Field Value
Activity pr-unblock/apply-feedback-patch

Run 44950889751, attempt 1

Field Value
Model rung 0
Leg base
Run job success
Activity step success
Commits pushed 866279c
Run 44950889751

Failed GitHub API calls

No failed call recorded by the run job's scripts.

@melodic-automation-lanes

Copy link
Copy Markdown
Contributor Author

pr-unblock / address-feedback finished round 1 of max 4 on 866279c.

@github-actions

Copy link
Copy Markdown
Contributor

Lane run report

Field Value
Activity pr-unblock/publish-feedback

Run 44950889751, attempt 1

Field Value
Model rung 0
Leg base
Run job success
Activity step success
Commits pushed none recorded
Run 44950889751

Failed GitHub API calls

No failed call recorded by the run job's scripts.

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.

harness-config: draft-auto-mode-rules takes allow-rule input from /auto-mode-setup's denial scan (2.1.329)

0 participants