Skip to content

feat(pr-pipeline): hold for a person only on the owner's three reasons - #10954

Merged
kyle-sexton merged 32 commits into
mainfrom
feat/pr-unblock-three-reason-holds
Oct 30, 2026
Merged

kyle-sexton merged 32 commits into
mainfrom
feat/pr-unblock-three-reason-holds

Conversation

@kyle-sexton

@kyle-sexton kyle-sexton commented Oct 29, 2026 •

Copy link
Copy Markdown
Member

Refs: #10517
Refs: #10971

Authority

#10517 (owner-filed) asks that every needs-human path either route to a lane or be named as a kept trust control, because the owner's stop rule, recorded on main in docs/contracts/loop-lane/README.md ("When work stops for the owner"), allows a person only for three reasons. That section quotes the owner verbatim: "I (the human) should only ever be in the loop or involved via signals like needs-human / needs-decision labels if there is 1. A change in original intent/vision 2. Something comes up during implementation [...] that alters intent/vision or the primary design 3. A real HIGH/CRITICAL security or cost risk", and "Any other stop is a defect to fix, not a reason to ask me." The pr-unblock holds this PR reroutes are such paths. ADR 0074 records the dependency round cap as judgment ("The cap is judgment"), not an owner-kept control; no owner decision on main keeps it with a person.

A needs-session route is still a stop: route: fail posts a failing pr-unblock/<activity> status on the head, so the PR cannot merge until the cause is fixed. Only the actor changes, from a person to an agent session. No protected-path, supply-chain or workflow-pin check is removed or weakened; the lane still refuses to push those paths itself.

The owner's standing direction (2026-10-29), quoted verbatim, settles the protected-path, supply-chain and round-cap re-routes: "Holds that guard against our own actors (path holds, trust-path holds, person-approval holds, bot-author holds, pin/repin ceremony) are friction, not security boundaries, and come off now."; "Section 3's 'high/critical boundary' means only: a token or secret leaking, my accounts being compromised, or anything that bills me."; "A decision I made in any session binds every session." Every route this PR changes still stops the PR (a failing status) and only changes who answers it. None of them guards a token or secret leak.

Summary

The pr-unblock lanes handed a PR to a person (hold, human-gated label) for many stops that are not the owner's three reasons. This PR keeps hold only for an intent or design question, a HIGH or CRITICAL security or cost risk, or a person's own thread or hold, and sends the rest to the session-builder queue.

Route table (.github/scripts/pr-unblock/escalation-routes.json):

Reason Before After
protected-path-conflict hold fail + needs-session
conflict-patch-protected-path hold fail + needs-session
feedback-protected-path hold fail + needs-session
ci-fix-protected-path hold fail + needs-session
ci-fix-weaken-required hold fail + needs-session
dependency-round-cap hold fail + needs-session
dependency-fix-not-landed hold fail + needs-session
dependency-config-unreadable hold fail + needs-session
dependency-files-truncated hold fail + needs-session
dependency-control-uncited hold fail + needs-session
dependency-workflow-unpinned hold fail + needs-session
ci-fix-escalated fail fail + needs-session

Kept as holds, each with its reason:

  • feedback-round-cap: round cap reached while an open thread is a security finding rated IMPORTANT or above, an intent finding, or a person holds the PR.
  • feedback-uncertain: a CRITICAL finding, a security-review finding rated IMPORTANT or above or unrated, or one no known lane wrote; no adjudicator judges these.
  • feedback-disputed: a disputed finding only a person decides (same classes, or a thread a person directed).
  • feedback-question-intent: the dispute turns on what the change is meant to do.
  • feedback-question-design: the dispute turns on a design choice.
  • feedback-question-security: the dispute turns on a security judgment.
  • dependency-security-close: the review would close a security update.
  • dependency-escalate: the review found a risk the lane does not decide alone.
  • dependency-fix-refused: a non-Dependabot commit changes more than a manifest or lock version, so the lane cannot vouch for it.

A reason the table does not list now holds for a person with no activity to resume (fail-safe), where it used to make report-escalation.sh exit with an error.

Fix

  • escalation-routes.json gives 12 reasons route: fail with label: needs-session: the 11 holds above and ci-fix-escalated, already a fail route, which left a PR with no next actor once its same-head retry was spent (fix(pr-pipeline): give a ci-fix-escalated stop a next actor #10971). report-escalation.sh applies that label (only on a fail route, and only a well-formed label name) before the comment that explains it, and the comment names the queue.
  • Adjudicator contract reading (adjudicate-feedback.mjs): it reuses pr-verify's linkedIssues (now exported from resolve-pr.mjs) to stage the linked issues' spec text as adjudicate/contract.json per slot. A verdict that settles a dispute by quote must quote an exact substring of that contract text, at least 20 characters after trimming (QUOTE_MIN), and settles only question-intent and question-design. Security questions, CRITICAL findings and unrated security findings never settle by quote; they still reach a person.
  • Hold floor (babysit_risk.FLOOR_CLASSES, copies regenerated by scripts/sync-hold-floor.py): resolve-pr.mjs, write-verdict.mjs, intake-triage/admit.mjs, select-trusted-text.mjs, lib/publish-gate.mjs, .github/recurring-schedule.json and plugins/work-items/scripts/charge-words.json join as enforcement, because the adjudicator now loads them to read the linked spec a quote settles a question from, so they decide a stop route (as request-budget.mjs does).
  • pr-unblock.yml, pr-pipeline.yaml and the loop-lane and pr-pipeline convention docs describe the new routes and the needs-session queue.

Verification

  • node --test run inside .github/scripts/pr-unblock/ (the package's own test script) after merging origin/main: 978 tests, 978 pass, 0 fail.
  • report-escalation.test.sh covers the fail-plus-label route, the label-before-comment order and the unknown-reason hold. ci-fix-escalated is in its session reasons, and ci-fix-escalated, session: label is needs-session passes. The fail-comment edit-in-place cases now use ci-fix-round-cap, which is still unlabeled. The suite reports 0 not ok.
  • After the second origin/main merge, which resolved the resolve-pr.mjs linkedIssues conflict by keeping the export and main's webHosts parameter: node --test passes in .github/scripts/pr-unblock (978), pr-verify (178) and pr-merge (645), and in .github/actions/resolve-config for callers.test.mjs (200). scripts/check-changelog-fragments.sh --check and --check-required origin/main both pass.
  • Diff checked for secrets and machine-specific paths before the push: none.
  • Left out: the literal node --test .github/scripts/pr-unblock/ from the repository root, which Node v24.21.0 refuses with MODULE_NOT_FOUND (it treats the directory as a module); the same tests ran from inside the directory instead.

Related

🤖 Generated with Claude Code

https://claude.ai/code/session_01VQb9Q3ucKDVyHvpcU5cfU8

kyle-sexton and others added 4 commits October 29, 2026 09:40
…er's three reasons

escalation-routes.json now routes `hold` (needs-human) only for an intent or
design question, a person's own thread or hold, and a HIGH or CRITICAL
security risk. Protected-path stops, ci-fix-weaken-required after its fresh
fixer round, and the review-dependency stops other than security-close,
escalate and fix-refused route `fail` with "label": "needs-session", which
report-escalation.sh adds before its lane-stopped comment. A well-formed
reason the table does not list now holds for a person.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VQb9Q3ucKDVyHvpcU5cfU8
…item contract

The adjudicate-feedback job stages the linked issues' trusted spec text,
read by pr-verify's resolve-pr.mjs linkedIssues (now exported), as
contract.json. A lane or reviewer verdict on an intent or design question
must quote it: the script keeps the quote only as an exact substring of the
staged text, at least 20 characters, and the reply quotes it with the
issue's URL. Otherwise the question stays a person's hold. A security
question, a CRITICAL finding and a security-review IMPORTANT finding never
settle by quote.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VQb9Q3ucKDVyHvpcU5cfU8
@kyle-sexton
kyle-sexton marked this pull request as ready for review October 29, 2026 13:59
@chatgpt-codex-connector

chatgpt-codex-connector Bot commented Oct 29, 2026 •

Copy link
Copy Markdown

Codex Review Summary

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

Review Status Commit Review trigger
📝 Code Review ✅ Completed 2026-10-29T14:01:20.099144Z 2faac8b 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 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-security-review: 1 inline finding(s) on 2faac8b87714e8b021aeb6e09f9beca6fab5440b; the summary is in this lane's pull request comment.

claude-lane: claude-security-review

@github-actions

github-actions Bot commented Oct 29, 2026 •

Copy link
Copy Markdown
Contributor

claude-security-review reviewed bf75b33531960ca1482a4d6171b645a3c7ce631a: 0 finding(s), 0 as inline comments.

Security review of PR #10954 (head bf75b33): no findings.

critical-findings: 0

What I read: the whole diff (32 files) and the PR context. Bash was unavailable, so I read changed files with Read and Grep only. I did not consult docs/conventions/instruction-exception-register/README.md. No instruction-surface file (CLAUDE.md, AGENTS.md, a skill body) was changed, and none of the removed prose is a rule.

Route changes (escalation-routes.json, report-escalation.sh).

  • Eleven hold reasons become fail plus the needs-session label, and ci-fix-escalated gains the label.
  • The reroutes cover protected-path, workflow-pin and pin-ceremony stops. The owner threat model names those as friction holds, so their removal is not a finding.
  • Dependency reasons that are not named friction holds (dependency-round-cap, -fix-not-landed, -config-unreadable, -files-truncated) still stop the PR. The route posts a failing pr-unblock/<activity> status, and the stop's own text says the lane "does not mark this dependency update for merge". Those reasons are not a path to a leak, a compromise or a charge.
  • The hold routes for security, cost and intent questions (dependency-security-close, dependency-escalate, dependency-fix-refused and the feedback-* holds) are unchanged.
  • An unknown reason now holds for a person instead of exiting with an error. That fails safe. The new jq and shell paths pass ESCALATION_REASON only after the ^[a-z]+(-[a-z]+)*$ check, and the label value is also validated, so there is no injection surface.
  • needs-session is not a merge hold anywhere in pr-merge/ or babysit-prs/scripts. A rerouted PR is blocked only by the failing status on its current head. pr-unblock.yml also now skips the lane on needs-session PRs.

Adjudicator quote path (adjudicate-feedback.mjs).

  • The contract text comes from linkedIssues, which reads only text people on the trusted list wrote. It is loaded from the base checkout, not a PR head.
  • A quote settles a question only if it is an exact substring of that text, at least 20 characters, and the question is intent or design. verdictOf and checkQuote fail toward a person.
  • A lane verdict on such a question is converted to human. A reviewer verdict with a quote only keeps the finding standing, so the quote cannot dismiss a finding.
  • parseVerdicts accepts a quote only with a URL that matches ISSUE_URL. Security questions never settle by quote.
  • The quoted text is rendered with each line prefixed by > and &lt;!-- escaped, so it cannot forge a lane marker.

Workflow token (pr-unblock.yml).

  • The adjudicate-feedback job gains issues: read and pull-requests: read.
  • github.token is passed only as GITHUB_TOKEN on the prepare step. The model steps (run-claude-cli) receive only the API key and run behind setup-runner-isolation. The widened read scope therefore does not reach untrusted text or the model.
  • LANE_IDENTITY comes from yq over the base pr-pipeline.yaml, not from the PR head.

Hold floor. resolve-pr.mjs, write-verdict.mjs, admit.mjs, select-trusted-text.mjs, lib/publish-gate.mjs, recurring-schedule.json and charge-words.json are added as enforcement in merge-holds.mjs, babysit_risk.py, trust-path-hold.json, the pr-require-checks.yml generated pattern list and the two pr-review-*-hosted.yml lists. This only widens what is held. resolve-pr.mjs gains an export and nothing else.

Removed holds, not findings. The path, protected-path and pin or repin holds listed above become failing statuses plus a label, so they fall under the owner's named friction holds. I hold nothing for them.

Spend. One extra adjudicator input file per disputed thread. It uses prepaid or subscription usage only and starts no billable resource.

Manifest fields: no plugin.json or marketplace.json change in this diff.

@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 2faac8b87714e8b021aeb6e09f9beca6fab5440b; the summary is in this lane's pull request comment.

claude-lane: claude-review

@github-actions

github-actions Bot commented Oct 29, 2026 •

Copy link
Copy Markdown
Contributor

claude-review reviewed bf75b33531960ca1482a4d6171b645a3c7ce631a: 0 finding(s), 0 as inline comments.

No blocking or advisory findings on head bf75b33. I did not run any tests: this lane has no shell. The 978/178/645/200 test counts in the PR description are the author's and unverified by me.

What I read

  • The adjudicator change: adjudicate-feedback.mjs, its prompt and its schema.
  • escalation-routes.json and report-escalation.sh.
  • The pr-unblock.yml adjudicate job, including the new issues: read and pull-requests: read grants.
  • The clear-escalations and apply-patch test and comment edits, the pr-pipeline.yaml labels-absent change, and the hold-floor additions (merge-holds.mjs, babysit_risk.py, trust-path-hold.json, the generated pr-require-checks.yml patterns, and the two hosted review workflows).
  • linkedIssues in resolve-pr.mjs (now exported), to confirm the contract staging reads what the description says.

Checks that held

  • Quote check:
    • The quote check is fail-safe. A missing, short or non-matching quote, a lane verdict on an intent or design question, and any security question all end as human. A reviewer verdict can only make a finding stand.
    • collect reads contract.json from the runner-temp slot. The Read-only model works on a copy under $GITHUB_WORKSPACE/adjudicate, so it cannot edit the text it is checked against.
    • parseVerdicts accepts a quote only with an issue URL that passes ISSUE_URL.
    • linkedIssues keeps only same-repository issues, so the URL built from repository is correct.
  • report-escalation.sh: a label is applied only on a fail route, only if it is a well-formed name, and before the comment. An unlisted reason now holds with an empty activity, and the resume text omits the dispatch command in that case.
  • Hold floor: the seven new enforcement paths appear consistently in babysit_risk.py, merge-holds.mjs, both hosted review workflows and the generated hold patterns. trust-path-hold.json adds only the two non-.github files, since .github/** already covers the rest. The class counts in the Python tests (49, 116) match the additions.
  • Token scope: pr-unblock.yml is a workflow_dispatch workflow, so the job-level issues: read and pull-requests: read apply directly with no caller cap.

Observations, not findings

  • A quote shows only that the text exists in the issue, not that it bears on the finding. The reply still says "The linked issue settles this … question". The code comment in verdictOf and the README admit this. It is also why a quote can never dismiss a finding.
  • The substring match is exact. A multi-line quote from a CRLF-edited issue body would fail to match and fall back to a person. That is safe, but it reduces how often a quote settles anything.
  • A legacy needs-human label on ci-fix-protected-path or ci-fix-weaken-required now comes off when the head's checks read failed (clear-escalations.mjs). The PR then gets no needs-session label unless the lane stops again. The description says open PRs are swapped by hand.
  • needs-session blocks review-dependency and apply-dependency-patch until a session removes it. This is intended and documented, with a test in live-config.test.mjs.

Closing issues: the PR uses Refs: #10517 and Refs: #10971, not closing keywords, so there are no closing-issue sections and no acceptance criteria to check.

Manifests: no plugin.json or marketplace.json change, so no manifest items apply. The two changelog fragments carry bump: minor (pr-pipeline) and bump: patch (source-control). Both are reasonable for the changes described.

Spend: no billed resource or paid plan changes. The adjudicator model calls it already ran now also read contract.json. This draws on prepaid or subscription credit only, so I raise no hold.

kyle-sexton and others added 2 commits October 29, 2026 11:52
…-reason-holds

# Conflicts:
#	.github/scripts/pr-verify/resolve-pr.mjs
A ci-fix-escalated stop routed fail with no label, so once its same-head
retry was spent no queue held the pull request. It now names
"label": "needs-session", as ci-fix-weaken-required does, and
report-escalation.test.sh covers it among the session reasons; the
fail-comment cases move to ci-fix-round-cap, still unlabeled.

Refs: #10971

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VQb9Q3ucKDVyHvpcU5cfU8
kyle-sexton and others added 3 commits October 29, 2026 12:45
…-reason-holds

# Conflicts:
#	docs/conventions/loop-lane/README.md
…uote dismissals for a person

- review-dependency and apply-dependency-patch list needs-session in
  labels-absent, so a dependency stop routed to the session builder gets no
  same-head model re-run and no fresh round budget on a rebased head until
  a session removes the label.
- A quoted contract line only keeps a finding standing: a lane verdict on an
  intent or design question goes to a person, since the quote proves the
  text exists, not that it bears on the finding. The prompt and the lane
  reply drop the lane-side quote.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VQb9Q3ucKDVyHvpcU5cfU8

@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 dfd67b6dbd2be7aaf462ea5bc3f185733f5d469c; the summary is in this lane's pull request comment.

claude-lane: claude-review

Comment thread docs/conventions/pr-pipeline/README.md Resolved
…the needs-session queue

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VQb9Q3ucKDVyHvpcU5cfU8
kyle-sexton and others added 4 commits October 29, 2026 15:16
… as a file in it

admit.mjs reads `new URL("../../workflows/")`, and import_closure crashed
with IsADirectoryError on it. Follow loaded_closure's convention: a module
that reads a whole directory stands for a file in it.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VQb9Q3ucKDVyHvpcU5cfU8

@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-security-review: 1 inline finding(s) on b3d6a6c21ca444ccb660d24cc2cbf7bce25110be; the summary is in this lane's pull request comment.

claude-lane: claude-security-review

kyle-sexton and others added 2 commits October 29, 2026 16:27
…losure

admit.mjs reads `../../workflows/` only for the workflow names; it loads no
code from there, and every gated workflow is floored by name. Standing for
`.github/workflows/any.yml` (loaded_closure's convention) fits only an
assertion that also accepts CARRY_REFUSES; this one checks the floor alone.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…ked spec through

adjudicate-feedback.mjs now loads resolve-pr.mjs and admit.mjs to stage
the linked issues' spec text a quote settles a question from, so their
closure decides a stop route. Floor the seven files as enforcement, as
request-budget.mjs is for the same reason, and regenerate the copies.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@melodic-automation-lanes

Copy link
Copy Markdown
Contributor

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.

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=10954 -f activity=resolve-conflicts

Lane-Activity: pr-unblock/resolve-conflicts

@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 e1d75eb03a1b1b1bb2609eb83c21ef91da12e377; the summary is in this lane's pull request comment.

claude-lane: claude-review

@kyle-sexton
kyle-sexton enabled auto-merge October 30, 2026 03:39
…tions path

The contracts move left the relative link one level short, which failed
lint-repo's offline link check.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@melodic-automation-lanes melodic-automation-lanes Bot added the needs-human Human-in-the-loop required; autonomous sessions must not resolve items carrying this. label Oct 30, 2026
@melodic-automation-lanes

Copy link
Copy Markdown
Contributor

The pr-unblock lane stopped on this pull request and needs a person: a conflicted path is a workflow, submodule config or a settings file, which the lane does not resolve.

  • Reason: protected-path-conflict
  • Resume key: e1d75eb03a1b1b1bb2609eb83c21ef91da12e377 (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=10954 and activity=resolve-conflicts:

gh workflow run pr-unblock.yml -R melodic-software/claude-code-plugins -f pr-number=10954 -f activity=resolve-conflicts

kyle-sexton and others added 3 commits October 30, 2026 00:04
A tests-only verdict carrying a design or intent question and a quote was
kept and then acted on from green CI. Reject it as malformed, as before.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@kyle-sexton kyle-sexton removed the needs-human Human-in-the-loop required; autonomous sessions must not resolve items carrying this. label Oct 30, 2026

@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 a031e35140aa72efc06815ac3b30cce22133d1f1; the summary is in this lane's pull request comment.

claude-lane: claude-review

@kyle-sexton

Copy link
Copy Markdown
Member Author

Settling the pr-verify findings from head 40477ab. That run read the old body, which cited "No related issue: owner standing direction". The body now cites #10517 as the trusted source.

  • Owner direction (2026-10-29, verbatim). "Holds that guard against our own actors (path holds, trust-path holds, person-approval holds, bot-author holds, pin/repin ceremony) are friction, not security boundaries, and come off now." "Section 3's 'high/critical boundary' means only: a token or secret leaking, my accounts being compromised, or anything that bills me." "A decision I made in any session binds every session." Each route here is still a stop, with a failing status, and only the actor who answers it changes. None guards a token or secret leak, so none of them is an intent change. The PR body's Authority section now carries these quotes.
  • Authority. fix(pr-pipeline): align every needs-human path with the owner stop rule #10517 is owner-filed. It asks that every needs-human path "either routes to a lane (no needs-human) or is named in the 'When work stops for the owner' section as a kept trust control". That section is on main in docs/contracts/loop-lane/README.md. fix(pr-pipeline): give a ci-fix-escalated stop a next actor #10971 is only the first slice.
  • dependency-round-cap. ADR 0074 calls the round cap judgment ("The cap is judgment"). No owner decision on main keeps it with a person. This PR still makes a fail route post a failing pr-unblock/<activity> status, so the PR cannot merge until the cause is fixed. Only the actor who acts on it changes.
  • Protected-path and supply-chain holds (protected-path-conflict, conflict-patch-protected-path, feedback-protected-path, dependency-*). The owner decided this on 2026-10-29: an agent's change to a protected path stops for a person only when a deterministic check finds a privilege change. All other cases go through both review lanes. The lane still refuses to push those paths itself, and no check is removed. fix(pr-pipeline): align every needs-human path with the owner stop rule #10517's security bar ("No HIGH/CRITICAL security path is loosened without a security-review pass at the head") is met by the security-review lane on each head.
  • Adjudicator quote settling, its read-only permissions and the seven hold-floor paths. This covers fix(pr-pipeline): align every needs-human path with the owner stop rule #10517's "Review rounds hand off on 'Unsure: escalate'" item. An intent or design question is settled only by an exact quote of owner-filed issue text, never a security question, and the floor only grows. The open review thread on it is fixed in a031e35: only a reviewer verdict settles by quote.
  • The ci-fix-escalated to needs-session slice (fix(pr-pipeline): give a ci-fix-escalated stop a next actor #10971) moves to its own PR. Once that PR merges, this PR drops its copy when main is merged in.

@melodic-automation-lanes melodic-automation-lanes Bot added the needs-human Human-in-the-loop required; autonomous sessions must not resolve items carrying this. label Oct 30, 2026
@melodic-automation-lanes

Copy link
Copy Markdown
Contributor

The pr-unblock lane stopped on this pull request and needs a person: a conflicted path is a workflow, submodule config or a settings file, which the lane does not resolve.

  • Reason: protected-path-conflict
  • Resume key: a031e35140aa72efc06815ac3b30cce22133d1f1 (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=10954 and activity=resolve-conflicts:

gh workflow run pr-unblock.yml -R melodic-software/claude-code-plugins -f pr-number=10954 -f activity=resolve-conflicts

@github-actions github-actions Bot removed the pr-verify-failed pr-verify failed this head; a pass at a new commit, or after a trusted body edit, removes it. label Oct 30, 2026
…-reason-holds

# Conflicts:
#	.github/workflows/pr-require-checks.yml
#	plugins/source-control/skills/babysit-prs/scripts/tests/test_babysit_risk.py
#	plugins/source-control/skills/babysit-prs/scripts/tests/test_merge_holds_sync.py
@melodic-automation-lanes

Copy link
Copy Markdown
Contributor

The pr-unblock lane stopped on this pull request and needs a person: the pull request changes .github/workflows/, which the lane token cannot push, so the lane does not address its review threads.

  • Reason: feedback-protected-path
  • Resume key: a031e35140aa72efc06815ac3b30cce22133d1f1 (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=10954 and activity=address-feedback:

gh workflow run pr-unblock.yml -R melodic-software/claude-code-plugins -f pr-number=10954 -f activity=address-feedback

…-reason-holds

Compose both sides in the eight conflicted files:
- escalation-routes.json: keep the branch's ci-fix-escalated why; both
  sides add the needs-session label (#11175).
- escalation-routes.mjs, report-escalation.sh header: main's wording plus
  the branch's three-reason hold and fail-safe for an unlisted reason.
- report-escalation.test.sh: the branch's HOLD/SESSION reason sets with
  main's stricter no-other-label check.
- pr-pipeline.yaml: review-dependency and apply-dependency-patch wait on
  needs-session and "security: outside-actor".
- live-config.test.mjs: keep both new tests.
- pr-require-checks.yml: the branch's floored paths with main's
  path-holds guard (sync-hold-floor.py --check clean).
- pr-pipeline README: the branch's routing text plus main's note that a
  session, not a lane, removes needs-session.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@kyle-sexton kyle-sexton removed the needs-human Human-in-the-loop required; autonomous sessions must not resolve items carrying this. label Oct 30, 2026
@kyle-sexton
kyle-sexton added this pull request to the merge queue Oct 30, 2026
Merged via the queue into main with commit 11a4ee9 Oct 30, 2026
104 of 106 checks passed
@kyle-sexton
kyle-sexton deleted the feat/pr-unblock-three-reason-holds branch October 30, 2026 14:49
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.

1 participant