[JacobStephens2]

Summary

bug Ticket run (fresh or Continuation)
  thirdshift-diagnosing-bugs before changing code
    reproduced → failing regression test → fix → review → PR delivery
    not reproduced → triage notes + needs-info → successful ending, comment link
    incomplete handoff → stop before delivery
other issues, Security fixes and Base fixes → existing implementation flow

Ship the headless diagnosis skill with the embedded Factory skills and publish its prompts on the Prompts and skills page. Document the workflow on the hand-written site pages. An unreproduced bug releases its Claim without requeueing, skips Security review and PR delivery, and links the triage comment in its outcome and notification. A Spec run reports an unreproduced Ticket without claiming it landed.

Continue the existing branch and reconcile current main, preserving review metrics, Security-finding close gates and Ticket run terminology. Compare triage links by parsed issue identity so accepted issue URL spellings, including /issues/007, accept canonical /issues/7#issuecomment-… links.

Closes #957

Evidence

• Before: bug issues received the ordinary implementation prompt; the original unreproduced-bug regression proceeded toward PR delivery and the Ticket regression reported a landed PR.
  After: retained CLI regressions cover fresh Ticket runs, Continuations with and without PRs, needs-info/Pickup, notifications, excluded issue types and incomplete handoffs. cargo test --test bug_run --test review_957_handoff_failure passes all 8 active tests.
• Before: the reviewer's exact command, cargo test --test bug_run not_reproduced_accepts_canonical_comment_for_leading_zero_issue_url -- --exact --nocapture, failed with exit 101: the Ticket run returned 1 instead of 0 because the canonical triage link did not match /issues/007.
  After: the identical command passes (1 test); the regression is retained. cargo test --test codex_sessions implement_sessions_and_their_resumes_request_fresh_reviewers -- --exact --nocapture also passes.
• Before: CI and cargo test --test pickup_run the_lower_numbered_of_two_ready_issues_is_run_with_lowest_first -- --exact --nocapture rejected a bug-labelled Pickup fixture that omitted the diagnosis completion line.
  After: the same command passes, and cargo test --test pickup_run passes all 96 active tests. A shared bug-success fixture now emits the completion marker for the ordering tests added on main, preserving the ordinary and failure fixtures.
• Validation on 6b65e7b1: formatting, clippy, prompt regeneration/verification and all 50 local site browser tests pass. CI passes the full Linux Rust suite, macOS, site and Console checks. CI's cargo test took about 11m47s, including 47s compilation. /usr/bin/time -p scripts/affected-tests main passed its 50 browser tests locally; its duplicate Rust phase was stopped after exact-head CI succeeded, following docs/testing.md (786s local elapsed at interruption, cached build; this is not a completed local full-suite timing).
• Pickup verification: claimed-dispatch safety passes, and both safeguard mutations still produce their expected counterexamples. The known pre-Claim dispatch counterexample remains documented; the literal dispatch property is not proven.

Test seams: the thirdshift CLI through fake Harness/GitHub adapters; Factory skill worktree linking and package contents; generated prompt/page output and browser DOM; the Pickup protocol's public Stateright Model interface.

Merge Danger

Door: two-way

Blast Radius: Diagnosis

Bug-labelled implementation sessions require an explicit diagnosis completion line. Valid needs-info handoffs end successfully without PR delivery; missing or invalid handoffs fail before delivery.

Unaddressed findings

• Standards: none. Both findings were accepted and fixed: diagnosis Claim policy uses the existing Claimed/Outside seam, and new prose uses the glossary's Ticket run terminology.
• Spec: none. The canonical-comment URL finding was reproduced with the supplied test, fixed and retained as a regression.
• Security: none. The diagnosis handoff, claim release, and skipped delivery were reviewed against the pinned merge base; no introduced finding.

Review coverage: Standards raised 2 findings and Spec raised 1; all were fixed. Both axes read all 35 changed files, including generated output, through ca410b23, and reviewed the findings' fixes in their single follow-up. The later CI repair 6b65e7b1 changed tests/pickup_run.rs to supply the diagnosis completion marker in a shared bug-success fixture; that final change remains unread by both reviewers. No other changed file is unread, and there are no generated exclusions. Reports and exact counts are retained beside this Session in .thirdshift-review-QE7HK5/standards.md, spec.md and summary.json.

The thirdshift-code-review skill requires: “Each axis gets at most one follow-up or replacement; remaining gaps are reported, not retried.” Both follow-ups preceded the CI fixture repair, so its unread change is disclosed here.

Built with codex codex-cli 0.164.0 · gpt-6.1-sol · xhigh

Security review outcome

No unaddressed introduced findings.