[JacobStephens2]

Docs only: docs/agents/triage-labels.md, docs/guide.md and the regenerated site/docs/index.html.

A scheduled Triage Pass passes over an issue with sub-issues, but a Spec has none until its first Ticket is attached. Published as needs-triage, as the Day shift procedure said, it could be triaged as a single issue in the middle of publication.

• The Day shift publication steps now hold an unfinished Spec as ready-for-human, then swap it to ready-for-agent once every Ticket and blocking link is in place.
• "Choosing between needs-triage and ready-for-human" now says: use ready-for-human for anything a scheduled Pass should not take yet.
• The user manual's advice for to-spec/to-tickets says the same, and gives the reason.

Not changed: the Factory skills thirdshift-to-spec/thirdshift-to-tickets still publish Architect plans as needs-triage, and thirdshift swaps the label in code. That fix belongs to #1965. A Spec that thirdshift hands back to needs-triage after a waiting Spec run already has sub-issues, so a Triage Pass passes over it.

npm run docs:check --prefix site passes.

CI notes

• process::process_session_evidence_check_all_private_fix_identifiers_and_abbreviated_commits (run 43054869408) is intermittent and unrelated: this PR changes only docs. The test asserts that request.json does not contain 421. That file includes the random experiment-id path, and in that run the id 5e2e28d9…d6087fdd054211fa05b0c30c happened to contain 421.

Built with claude 2.1.322 · claude-opus-5-5 · medium

[JacobStephens2]

Review run. Pushed fixes to publication-ready-for-human at fb3f3c2. Spec axis used the pull request body alone; no closing issue was linked.

Review of PR #1991: Publish an unfinished Spec as ready-for-human

The reviewers raised 3 Standards findings (all style judgement calls) and no Spec findings. I fixed one in commit fb3f3c29, not pushed, and declined the other two. Final validation passed. A broken run of the full Rust suite, triggered by my own log files, showed 2 failures in code this branch does not touch.

Fixed point: the supplied ref refs/remotes/origin/74b072d07fe2d8b7f6116564c724074c5870e572 does not exist locally. I used the commit it names, 74b072d07fe2d8b7f6116564c724074c5870e572. One commit was reviewed: e7ec3794. The PR is still open and not a draft.

Spec source: the PR has no linked closing issue, so the PR body alone was the Spec.

Standards (3 findings, all style judgement calls; nothing breaks a documented rule)

• Documented standards: nothing is broken.
  • CODING_STANDARDS.md only covers deadlines in tests, so it doesn't apply to docs.
  • The AGENTS.md rule for the generated user manual was followed: the source was edited and the manual regenerated.
  • The AGENTS.md rule for hand-written site pages needed no edits: those pages mention needs-triage, to-spec and to-tickets only in general terms.
  • GLOSSARY.md and ADR 0029 agree with the change.
• The code matches what the docs claim:
  • A Triage Pass skips ready-for-human issues and issues with sub-issues (src/triage/pass.rs:138-158).
  • Pickup skips ready-for-human (src/labels.rs:73).
  • The thirdshift-to-spec Factory skill still publishes as needs-triage (skills/thirdshift-to-spec/SKILL.md:23).
• Finding 1, repeated wording: the same reason ("needs-triage won't do, because a Spec has no sub-issues until its first Ticket is attached") appears three times in docs/agents/triage-labels.md and again at docs/guide.md:3875. Declined (see Unaddressed findings).
• Finding 2, unclear wording: in "In short: use ready-for-human for anything a scheduled thirdshift Pass should not take yet", the next sentence doesn't follow from "In short", and "thirdshift Pass" is broader than the Triage Pass the section is about. Declined.
• Finding 3, unclear wording: "They still publish a plan as needs-triage (A Triage Pass doesn't wait for an issue to settle, so it can triage a Spec before its Tickets are published #1965)" didn't say what risk remains. Fixed: the sentence now says a scheduled Triage Pass can take a Spec the Factory skills publish before its first Ticket is attached. I checked that claim against skills/thirdshift-to-spec/SKILL.md:23, src/architect.rs:492-498 and the body of issue A Triage Pass doesn't wait for an issue to settle, so it can triage a Spec before its Tickets are published #1965.

Spec (0 findings)

• No requirement from the body is missing or partial, nothing was added beyond it, and nothing it asks for is implemented wrongly.
• Every requirement in the body is in the diff.
• Every claim about the code holds:
  • A Triage Pass takes needs-triage issues and skips parents with sub-issues (src/triage/pass.rs:69,137-160).
  • Plans are swapped from needs-triage to ready-for-agent in code (src/architect.rs:492-498).
  • A Spec run starts only on an issue with sub-issues (src/ready.rs:555, src/spec_run.rs:70), so a Spec handed back to needs-triage (src/claim.rs:528-575) already has them.
• Leaving the Factory skills for A Triage Pass doesn't wait for an issue to settle, so it can triage a Spec before its Tickets are published #1965 matches the options that issue lists.
• No other doc, skill, ADR or hand-written site page still holds a Spec at needs-triage during publication.

Unaddressed findings

Standards

• Finding 1 (repeated wording): declined. The body asks the user manual to state the rule "and give the reason", so the reason in docs/guide.md:3875 is required. The manual is for users, while docs/agents/triage-labels.md is the procedure agents must follow, so each needs the reason on its own. No issue filed: this is a style choice, not one for the Day shift.
• Finding 2 ("In short" / "thirdshift Pass"): declined.
  • The body itself asks for "use ready-for-human for anything a scheduled Pass should not take yet".
  • The broader "Pass" is accurate, because Pickup also skips ready-for-human (src/labels.rs:73, src/ready.rs:146).
  • "In short" sums up the bullet list directly above it.
  • No issue filed.

Spec: none.

Review coverage

• Changed files: docs/agents/triage-labels.md, docs/guide.md and site/docs/index.html. Both reviewers read the first two.
• Generated file excluded from full reading: site/docs/index.html, which is built from docs/guide.md by site/generate-docs.mjs.
  • npm run docs:check --prefix site exited 0, after npm ci --prefix site exited 0. A first check failed because the site packages weren't installed.
  • Both reviewers accepted the exclusion and read the one changed paragraph. The new #triage-runs link resolves to id="triage-runs".
• Unread changed files: none on either axis. Each reviewer's full list of files read is in standards.md and spec.md.
• Reports: written to .thirdshift-review-TMQQWk/standards.md, spec.md and summary.json ({"standards":3,"spec":0}).

Validation

• scripts/affected-tests main, first run: exit 1. My untracked validation logs in .review-logs/ made the selector run the full Rust suite. npm test --prefix site passed. cargo test failed: 237 passed and 2 failed in --test process:
  • findings::closing_rechecks_each_merged_base_and_leaves_a_failing_finding_for_the_day_shift
  • findings::fixing_dispatches_each_reproduced_finding_and_keeps_unreproduced_findings_private
  • Both stop with "the Security fix publishing session created public issue(s) outside the staged path: Keeping the PR mergeable: merge the Base branch and conflict Repair #8".
• cargo test --test process findings::: exit 101. The same 2 tests failed again (16 passed). This branch changes nothing under src/, tests/, Cargo.toml or Cargo.lock, and no Rust file reads the changed docs. So these tests run code identical to the fixed point, and the failure is not this PR's. I did not run them at the fixed point to prove it was already failing there.
• scripts/affected-tests main, re-run after moving the logs into the git-ignored .affected-tests/: exit 0.
  • npm test --prefix site: PASS.
  • cargo test --test share_card --test site_deploy: PASS.
  • cargo test --bin thirdshift prompts_page: PASS.
  • docs/agents/triage-labels.md selects no checks of its own.
• Extra commands: the list was empty, so none ran.

Summary: Standards had 3 findings; the most significant was the unclear #1965 sentence, now fixed. Spec had 0 findings.

Security

No unaddressed introduced findings.

Delivery failed

pull request changed during Review run