Repository navigation
pr-pipeline: let pr-unblock revert a file mode the pull request itself introduced #10080
Description
Problem
pr-unblock's address-feedback mode edits file contents only, so a review finding about a file mode it introduced cannot be fixed and escalates feedback-uncertain.
Example: #10052 changed plugins/docs-hygiene/skills/audit-noise/scripts/lib/noise-shapes.sh from 100644 to 100755. claude-review flagged it; the lane agreed ("nothing needs it to be executable") but replied "It cannot run git or chmod" and escalated. A person-run git update-index --chmod=-x fixed it.
Owner decision (2026-10-25)
A stop is worth a person only for security, intent, or a real charge. Reverting a mode bit the same PR introduced is none of those.
Fix
Let the lane's patch output carry a mode change for a path the PR itself changed the mode of, restoring the base's mode only (never adding the executable bit). Apply it in the existing patch path (.github/scripts/apply-patch/apply-patch.mjs), not a second writer. Adding an executable bit stays out of scope: high-path mode changes remain a security hold.
Done when
- A finding asking to revert a PR-introduced mode change is fixed by the lane on a test PR, with no
needs-human. - A patch that would set an executable bit not on the base is refused.
Activity
github-actions commentedon Oct 25, 2026
Proposed work class: work-class: scoped
Basis: automated intake triage; the lanes App applied this class.
Size: Medium
Complexity: Medium
Uncertainty: Known
Flags: none
Blocked by: none
#10080 (comment)
The decision settles all five open questions. The lane gets a narrow
chmod -xgrant. apply-patch gets an always-on refusal of any new executable bit, except for the issue row'sscript-exec-bit. A PR-introduced mode revert is admitted only for paths the kept-threads predicate admits. Thebabysit_risk.pyrating stays as it is.Automated triage, applied by the lanes App.
melodic-automation-lanes commentedon Oct 25, 2026
Not admitted to the issue lane: no trusted-human acceptance criteria: a listed human must put the confirmed
- [ ]criteria in the body or post them as their own comment, or, on an issue a listed human filed, intake's newest refinement comment must hold them unedited. Admission takes the issue once each of those is met, when intake triages it again (.github/actions/resolve-config/admission-rules.json).melodic-automation-lanes commentedon Oct 25, 2026
The latest triage verdict holds nothing: the issue is at
status: ready.melodic-automation-lanes commentedon Oct 25, 2026
Refinement. A read-only run explored this repository, searched its issues and pull requests, and read the allowed docs to draft what follows. On an issue a listed human filed, it counts as the issue's spec once posted, while unedited; it is still model text, so check it.
Intended outcome: apply-patch.mjs refuses any patch entry that sets 100755 where the PR base had 100644, and the review-threads lane has a way to emit a PR-introduced mode revert that apply-patch accepts.
Related work
Open questions
RECOMMENDED: Grant the review-threads lane a narrow chmod -x tool rule, so collect-patch.sh emits the old mode/new mode lines itself and apply-patch.mjs only validates them. Basis: .base/.github/scripts/read-patch/collect-patch.sh:82 (the patch is a git diff of the staged working tree, so the model cannot write a mode line directly); .base/.github/actions/collect-base-activity/action.yml:692 (an existing Bash(chmod +x *) grant for commit-capable lanes)
RECOMMENDED: Every row, as a new always-on or listed refusal, so no row can set an executable bit the base lacks. Exempt the issue row only for its existing script-exec-bit behavior. Basis: .base/.github/actions/resolve-config/patch-rules.json:63-72 (the issue row adds exec bits itself via script-exec-bit); .base/.github/scripts/apply-patch/apply-patch.mjs:870-882
RECOMMENDED: A path whose mode in the merge-base of the PR head and its base ref differs from the head's mode, and a patch entry that sets exactly the merge-base mode. The applier needs the base ref, which it already receives as BASE_REF. Basis: judgment; .base/.github/scripts/apply-patch/apply-patch.mjs:45-55 lists BASE_REF as optional and PATCH_BASE_SHA as issue-only, so the PR target has no pinned base commit yet
RECOMMENDED: No. Admit it only for a path the kept-threads predicate already admits, since the finding names that file. Basis: .base/.github/scripts/apply-patch/predicates.mjs:14-23
high-pathmode changes remain a security hold. Should that high-path rating in babysit_risk.py change?RECOMMENDED: No. Leave the risk rating alone, and have the lane's mode revert handled only inside apply-patch. Basis: .base/plugins/source-control/skills/babysit-prs/scripts/babysit_risk.py:413-436 (any mode-type change already rates HIGH); the issue's Fix section limits the change to apply-patch.mjs
Sources read
Automated refinement by the lanes App. Answer with a comment whose first line is
**Decision recorded.**.kyle-sexton commentedon Oct 26, 2026
Decision recorded. The owner answered in the Refinement session (2026-10-25), deciding the open items below on the facts:
Decided on the facts, as net hardening: lanes already hold
Bash(chmod +x *)on the commit path (.github/actions/collect-base-activity/action.yml:692);chmod -xcan only remove an executable bit; and apply-patch has no refusal for a new executable bit today (.github/scripts/apply-patch/apply-patch.mjs:117-129,EXECUTABLEdefined at:84but unused by any refusal). The change adds that refusal, so the lane ends with less power to set modes than it has now.Question asked: Grant the review-threads lane a narrow
chmod -xand add an always-on "no new executable bit" refusal in apply-patch (trust-boundary call)?The body's owner decision (2026-10-25) settles the intent:
Open items come from intake's refinement: #10080 (comment)
Decided: Grant the review-threads lane a narrow
Bash(chmod -x *), socollect-patch.shemits the mode lines itself andapply-patch.mjsonly validates them. Basis: origin/main.github/scripts/read-patch/collect-patch.sh:84(diff --cached --binary);.github/actions/collect-base-activity/action.yml:692(the existingBash(chmod +x *)grant). Intake's answer kept.patch-rules.jsonrows get the new mode check? Security call.Decided: Every row, as an always-on refusal, so no row can set an executable bit the base lacks; the issue row keeps its
script-exec-bitbehavior. Basis:.github/scripts/apply-patch/apply-patch.mjs:118-131(REFUSALS has no mode check today),:870-873(script-exec-bit). Intake's answer kept.Decided: A path whose mode at the merge base of head and base ref differs from the head's, and a patch entry that sets exactly the merge-base mode, read through
BASE_REF. Basis: judgment;apply-patch.mjs:46-58(BASE_REFoptional,PATCH_BASE_SHAissue-only). Intake's answer kept.Decided: No; only a path the kept-threads predicate admits. Basis:
.github/scripts/apply-patch/predicates.mjs:13-24. Intake's answer kept.high-pathrating for mode changes inbabysit_risk.pychange?Decided: No. Basis:
plugins/source-control/skills/babysit-prs/scripts/babysit_risk.py:88(a file mode change rateshigh-path); the body's Fix limits the change toapply-patch.mjs. Intake's answer kept.Acceptance criteria:
needs-human. (Converted from the body's Done when, wording unchanged.)patch-rules.jsonrow except the issue row'sscript-exec-bit, with a test case.apply-patch.mjs, with no second writer, andbabysit_risk.py's mode rating is unchanged.Builder note: the existing
Bash(chmod +x *)grant sits on collect-base-activity's commit path. Confirm the new refusal leaves that path working for a lane that commits a new#!script there; a test with a new.shon the commit path settles it.