[kyle-sexton]

No related issue: owner decision Q15 (owner-decisions.md:11)

Owner, verbatim (Q15, 2026-10-29): "We shouldn't really have any path restrictions at all... like that is just bullshit friction", then "We can go in and harden later".

Owner, verbatim (scope confirmation): "I want no paths to start so things just go through... we can harden once we get our issues to 0 and PRs to 0"

Owner, verbatim (Q16, no globs kept): "No fucking path restrictions"

Hardening trigger: once open issues = 0 and open PRs = 0, turn path holds back on by setting merge.diff-check.path-holds: true in docs/conventions/pr-pipeline.yaml. The default stays off.

Summary

This PR adds one switch, merge.diff-check.path-holds (boolean, default false). It sits beside denied-paths and high-path-floor, the other path switches of the pr-merge lane under merge.diff-check. While it is off, a changed path no longer holds a lanes PR just because of the path. Setting it to true brings back every hold exactly as it was. The hold code and the floor tables (FLOOR_CLASSES, HOLD_PATH_CLASSES, hold_patterns) are unchanged. The switch is read only at the points where the holds are decided. The babysit gate reads it at both the base commit and the default-branch head, and holds when either says true. ci-status reads it from the base commit. So a PR cannot turn off its own hold.

merge.person-held does not exist on main. It existed only on the superseded #10966 (lane-arm attestation) branch, so there was nothing to remove.

Fix

Sites the switch disables (file:line at the head):

• babysit_risk.py:2194-2221: _score moves every is_path_hold reason into RiskResult.skipped. That covers high-path:<floor or configured glob> and review-path:<glob>. RiskInput.path_holds is at :207.
• babysit_merge_policy.py:640, 885, 934, 1049, 1125, 1196: the PipelineConfig.path_holds field, how it is read, the stricter-of-base-and-main rule, and validation.
• babysit_merge.py:3567: passes the switch to the pr-merge lane only. The local babysit lane keeps every path hold. babysit_verify_evidence.py:2472 passes it to enable-auto-merge's reader.
• merge-holds.mjs:389-418 (pathHold, pathHoldsOf, readPathHolds, pathHoldKind, i.e. the HOLD_PATH_FLOOR match) and :1313-1318 (holds). The CLI at :1605-1618 reads PATH_HOLDS.
• enable-auto-merge.mjs:1219 (decide) and :1519 (no lifter read for a path that no longer holds). :648-680 lifts a standing refusal that names only hold paths, for example "it changes .github/scripts/pr-merge/clear-escalations.mjs, a hold path". It also lifts a risk refusal whose every code is a path hold, and needs no approval for either. The lift goes through the existing liftBlocker checks.
• pr-require-checks.yml:5585-5591: ci-status's merge-holds step reads the switch from pr-pipeline.yaml at BASE_SHA and passes it as PATH_HOLDS. This step is separate from hold_patterns.
• report-outcome.mjs heldPaths (Q16): after feat(pr-pipeline): hold lanes App human-merge pull requests with a label only a listed person lifts #11036 (human-merge-hold label), report-outcome's denied-path match becomes a label that ci-status enforces. So with the switch off it denies no path. An unreadable config still holds.
• pr-pipeline.yaml:1093-1101 sets path-holds: false. The schema adds the key, and the README documents it.

Measurability (§9, Q15's skipped-hold count): the gate report lists each skipped hold under risk.pathHoldsSkipped. enable-auto-merge and merge-holds.mjs each log one path holds are off (merge.diff-check.path-holds), so nothing holds it for this: <reason> line per PR.

Kept holds (not path restrictions):

• both review lanes and their bound verdicts (security findings still hold)
• ci-status checks and gitleaks
• do-not-merge, body holds and member vetoes
• cost holds (COST_PATHS / cost-review-required)
• the breaker, the kill switch and the post-merge hub audit
• the trusted-actor widening (trust-gate:), a content check
• the manifest review and the secret-exposure, agent-surface and permission-widening checks, all content checks
• kept, owner may drop: high-path:**, the unsafe or aliasing path check. It covers symlinks and segments such as .., : or non-ASCII characters that could alias a cost or other path.

Not changed:

• ci-status's trust-path hold (hold_patterns), .github/trust-path-hold.json and the ci-workflows trust-paths. Another lane owns them, and they still apply.
• clear-escalations. A refusal it would keep is lifted by enable-auto-merge's widening read when it names only path holds. Any other refusal clears as head-moved on the next push.

Verification

• New tests fail on the base code and pass here:
  • off: a PR changing .github/scripts/pr-merge/clear-escalations.mjs, and one changing a floor workflow, gets no path hold in the risk scorer, the gate, enable-auto-merge, the merge-holds CLI or the ci-status step test
  • on: the old holds return
  • with the switch off, cost paths, do-not-merge, body holds and high-path:** still hold
  • standing hold-path refusals and path-only risk refusals are lifted
• Existing tests that exercise path holds now pin path_holds=True / path-holds: true. The live-config tests assert that this repository's config has the switch off.
• pytest plugins/source-control/skills/babysit-prs/scripts/tests: 3561 passed.
• node --test .github/scripts/pr-merge: 653 passed. .github/actions/resolve-config: 750 passed (the schema validates the live config).
• scripts/ci-lanes-pr-hold.test.sh: 246 passed. The other ci-status hold tests (early-queue, human-gated, merge-group, cost-decision, review-lanes, pr-unblock-status) all pass.
• After merging main (fix(pr-merge): request a whole review for a lanes PR's review-path hold #10999, feat(pr-pipeline): hold lanes App human-merge pull requests with a label only a listed person lifts #11036 and others): pytest 3583 passed, pr-merge node 667 passed, report-outcome node 385 passed, ci-lanes-pr-hold.test.sh 246 passed, ci-hold-human-merge.test.sh 23 passed.
• The merge conflicted in enable-auto-merge.mjs at the two decide() calls. I kept main's wholeReview (fix(pr-merge): request a whole review for a lanes PR's review-path hold #10999) and this PR's pathHolds. fix(pr-merge): request a whole review for a lanes PR's review-path hold #10999 clears review-path: holds and adds no hold of its own. feat(pr-pipeline): hold lanes App human-merge pull requests with a label only a listed person lifts #11036's label hold is not path-based, except for the denied-path match covered above.
• scripts/sync-hold-floor.py --check: exit 0. test_merge_holds_sync.py is green.

Related

• Refs feat(pr-pipeline): lane-arm the merge gate's own code on a verified pr-merge attestation #10966 (lane-arm attestation for gate code). Superseded by Q15, labeled do-not-merge.

🤖 Generated with Claude Code

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-10-29T21:09:45.753374Z	e0f35b2	Draft marked ready

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" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.

[github-actions]

claude-security-review reviewed ecb0f1b984fbf2924d02b3316f4ef5fd5b2785ef: 0 finding(s), 0 as inline comments.

Incremental security review of ecb0f1b against a1f05af. I reviewed only the two changed changelog fragments, .changes/pr-pipeline/feat-merge-path-holds-switch-3c1d7a20.md and .changes/source-control/feat-merge-path-holds-switch-8e52b4f9.md. I used the PR description for context.

The change is prose only. Both fragments move bump: minor to bump: major, change the heading from ### Added to ### Changed, and add a "Breaking:" line. The line says a config that lists denied-paths or review-path: globs and has no path-holds key no longer holds on them, and that merge.diff-check.path-holds: true restores the old behavior.

• Trust controls: No code, workflow, config value or trusted-path file changed in this increment. The fragments sit under .changes/ and match no trusted-path glob. They widen no trust control and read by no gate as a hold, a token scope or a kill switch.
• Merge-hold weakening: The path-hold switch, which defaults to off, is in files unchanged since the last review. The new text documents it and adds nothing. The changelog now states the default-off behavior more accurately, and the major bump marks it as breaking. That is a small gain in transparency, not a loss.
• Instruction surface: The diff deletes and softens no standing instruction. The old "Added" wording is replaced by an accurate "Changed" description.
• Manifests: No plugin.json or marketplace.json changed, so no manifest <field> lines apply.
• PR context: The PR description contains no embedded directives aimed at this lane. Its quoted owner statements describe intent only.
• Cost: A major bump in the release fragments sets the next release version. It starts no billed resource and no paid plan.

critical-findings: 0

[github-actions]

claude-review reviewed ecb0f1b984fbf2924d02b3316f4ef5fd5b2785ef: 0 finding(s), 0 as inline comments.

Incremental review of the two change files edited since a1f05af. I found nothing to report.

The edit changes bump: minor to bump: major, ### Added to ### Changed, and adds a "Breaking:" line to both fragments. That matches the PR: path-holds defaults to false, so a config with denied-paths or review-path: globs and no path-holds key stops holding on them. The migration note ("Set merge.diff-check.path-holds: true") names the right key. Both fragments say the same thing, and their front matter matches the existing fragments.

I did not check how the release tooling handles a major bump. These are the only two fragments under .changes that use bump: major, so the maintainers should confirm that is what they want released.

This change only edits release-note text, so it has no cost or security impact. I did not re-review the rest of the PR, which is unchanged since the last review.

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: e0f35b2187

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".

[melodic-automation-lanes]

The pr-unblock lane stopped resolve-conflicts on this pull request: the merge stages the pull request's own .github/workflows/ changes together with the base's, which the lane token cannot push, so a session merges main by hand. The lane sets no label: the run ends in a failing pr-unblock/conflict status on the head the lane gated.

• Reason: workflow-staged
• Head: e0f35b2187bd827ce587a90f951bc86e89e3099a
• Run: https://github.com/melodic-software/claude-code-plugins/actions/runs/40958777850

When a retry on this head can change the outcome, the lane runs one on its own, up to its per-head cap, and edits this comment if it stops again; a run that does not stop clears the status. When no retry can (a review thread the lane does not take or settle, or a failing check after the retries are spent), a person resolves the threads or gets the checks green, then dispatches the workflow below: that run finds the cause gone and clears the status. On a pull request the lanes App opened, the dispatchers start that run once per stop; on any other, a person does. A new commit to the branch starts clean.

To run it again on this head now, dispatch the pr-unblock workflow:

gh workflow run pr-unblock.yml -R melodic-software/claude-code-plugins -f pr-number=11058 -f activity=resolve-conflicts

[github-actions]

claude-security-review: 1 inline finding(s) on a1f05af17eee9846cf4da00b6cbb39c228a661e1; the summary is in this lane's pull request comment.

claude-lane: claude-security-review

[github-actions]

claude-security-review holds this pull request for a person (critical-security) at a1f05af17eee9846cf4da00b6cbb39c228a661e1.

The change weakens the pr-merge lane's path-based merge holds: merge.diff-check.path-holds is set to false in docs/conventions/pr-pipeline.yaml:1101 (hold logic in merge-holds.mjs, enable-auto-merge.mjs, babysit_risk.py and pr-require-checks.yml). Lanes-App PRs that change the token broker, kill switch, trusted-trigger check, merge-hold scripts or review workflows can now merge without a person. The PR cites owner decision Q15, but the review could not verify it.

claude-lane: claude-security-review

[github-actions]

claude-review: 1 inline finding(s) on a1f05af17eee9846cf4da00b6cbb39c228a661e1; the summary is in this lane's pull request comment.

claude-lane: claude-review