[kyle-sexton]

Refs: #8491 (the eval cases in its fourth checkbox are tracked by #9040)

Summary

Owner decision (2026-10-22): let the intake-triage lane apply needs-human as a plain label when an already-triaged issue comes back and still needs a human, so the run's own comment (quoted decision, new questions) is kept instead of the fixed escalation note.

Fix

• .github/workflows/intake-triage.yml: set allowed-labels to the reusable's defaults plus needs-human. The input replaces the default list, so the defaults are repeated.
• Re-entry step (d) now tells the run to apply needs-human with needs_human false.
• Reusable at the pinned v0.60.0 SHA: allowed-labels input at intake-triage.yml:160; needs_human: true escalates at :1496 and the fixed note is set at :1522-1527, while an allowlisted label passes allowed.has(label) at :1501 (admission() only rejects agent-ready and work-class:*). On a non-escalated run the floor label stays unless exactly one priority: label is applied and no status:/area: security label is.

Verification

• .github/standards/runner-policy/policy.json:232-257: the intake-triage.yml@82ca5ca9 contract lists allowed-labels in allowedInputs and compares names only, not values, so the contract is unaffected.
• needs-human exists as a label in this repository.
• No other file in .github/ or scripts/ pins the caller's allowed-labels value.
• Not run end to end: effect shows on the next re-entered issue.

Related

• feat(intake-triage): give the model each trusted comment's link ci-workflows#854
• chore(ci): pin ci-workflows v0.60.0 #9066

🤖 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 2026-10-22T15:32:09.153286Z	604452e	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 604452e5aae161885cbb04f41fedb3dc1334d918: 0 finding(s), 0 as inline comments.

No security findings in PR #9093 (head 604452e). The only changed file is .github/workflows/intake-triage.yml, which adds needs-human to the re-entry run's allowed-labels and changes the instruction from setting needs_human to applying the label.

• needs-human is a restrictive hold. The HOLDS set in label-unblocked.mjs treats it as stopping automation, and check-born-triaged.mjs lists it among role labels. Nothing I found makes it start or widen an action.
• work-unblock lifts a lone needs-human only when the lanes App's refs-hold marker and a merged Refs PR are present (label-unblocked.mjs:62-66, 731-734). A model-applied label carries no such marker.
• A prompt injection through issue text could at worst cause needs-human to be applied, which parks the issue for a person.
• The allowlist adds only that one label to the existing defaults. The issues: write permission and the pinned reusable are unchanged.
• No trigger, permission or pin change, so GitHub Actions hardening is zizmor's lane.
• Manifests: none touched, so there are no manifest <field> lines.
• Instruction-surface lens: the diff narrows a prompt instruction (needs_human true becomes applying the label). By consequence the same hold still applies, so no protection is lost. I did not consult docs/conventions/instruction-exception-register/README.md; this judgment is by consequence only.

[github-actions]

claude-review reviewed 604452e5aae161885cbb04f41fedb3dc1334d918: 1 finding(s), 1 as inline comments.

I reported one finding on this PR, a one-file change to .github/workflows/intake-triage.yml.

Finding (inline at .github/workflows/intake-triage.yml:226). The PR says Closes #8491, but #8491 asks for "an eval case for each of these, plus one where the answer settles everything". The PR adds none. I searched plugins/ and .github/ for Decision recorded and found no triage eval case. The existing reenter.test.mjs and write-issue.test.mjs cover the trigger and the comment selection, not the triage run's labels. Nothing would fail if step d regressed to needs_human: true. Either add the eval cases or change the link to Refs: #8491.

Unverified. The sandbox refused my gh api read of the reusable at the pinned v0.60.0 SHA. I could not check these claims, so they are author-claimed and unconfirmed:

• allowed-labels replaces the defaults rather than adding to them.
• The repeated default list matches the reusable's.
• needs_human: true escalates, and the fixed note is set at the cited lines.

If the list is wrong, a missing default label would be rejected on every triage run.

Scope note, not a finding. allowed-labels applies to every triage run, not only re-entry. The prompt limits needs-human to the re-entry section, so I left it unflagged.

Closing issue. I read #8491 from the PR context file. Its other criteria (quote the decision, re-park, keep needs-human) are covered by the instruction text.

[github-actions]

PR body contract — issue linkage

This PR body does not yet satisfy the issue-linkage contract:

• Missing a native closing keyword (Closes/Fixes/Resolves #N). If this PR references an issue it must not close, put "Refs: #N" (or "Relates to: #N") on its own line. If it relates to no GitHub issue at all, state "No linked issue" (or "No related issue:") in the body instead.

Edit the body and this comment updates itself on the next run.