[JacobStephens2]

Summary

editing Steps
  → security-review (flag/config; Harness fallback)
  → delivery (publish outcome; repair conflicts and CI)
  → self-merge (held by any refused/incomplete review)

Ticket runs and Custom processes share the review Step and Delivery's hold policy.

Closes #1461

Test seams

• CLI (thirdshift process <name>) with real Git, fake GitHub and fake Harnesses, as agreed in Spec: the Ticket run as a Built-in process file #1456.
• Existing Delivery unit seam retained while extracting the shared Security review Step.

Evidence

• Before: the Custom process regression failed because security-review was not a catalogue value.
  After: CLI coverage exercises config/flag selection, skips, placement refusals, configured and explicit Harnesses, fallback, and merge holds.
• Before: the Spec review's exact regression merged after a refusal followed by a clean review. A follow-up show regression returned one review trace instead of two.
  After: both exact commands pass; both tests are retained, every outcome is published, and each review remains discoverable.
• Before: the new flag words made existing Processes named security-review and no-security-review return usage errors.
  After: the reviewer's exact regression passes; bare words remain Process names, and dashed flags work before the name.

cargo test --test process security_review::later_clean_security_review_does_not_clear_an_earlier_refusal_hold -- --exact --nocapture
cargo test --test process security_review::show_lists_every_custom_security_review_trace -- --exact --nocapture
cargo test --test process security_review::security_flag_words_remain_usable_custom_process_names -- --exact --nocapture

• cargo fmt --check and cargo clippy --all-targets -- -D warnings pass.
• After reconciling the updated base: 268 focused integration tests passed, including the unchanged Ticket-run Security review, merge-run and Failed-run suites. All 12 new Security review CLI tests pass, including setup failure and untrusted-edit gates on an optional Step.
• /usr/bin/time -p scripts/affected-tests origin/issue-1456: 51 site tests passed. The local Rust run stopped after 310 passes on github::history_freezes_do_not_wake_an_unchanged_inbox (unchanged read woke the Inbox); that unchanged test passed directly and in five further consecutive runs. The selector took 725.44s wall, including compilation.
• Full Rust validation on exact head a03c9d68: CI passed, 3,293 passed, 74 skipped; 655.102s execution plus 35.43s compilation. This successful CI result is reused rather than repeating the full suite on the same head. The CI run also passed formatting, clippy, site and Console checks.

Merge Danger

Door: two-way

Blast Radius: Delivery

Shared review and hold handling affects Ticket runs and Custom processes; no data migration.

Unaddressed findings

Standards

None. The one finding (duplicated Security-hold policy) was addressed with a shared review-aware Delivery operation.

Spec

None. All three findings (lost earlier holds, missing review traces in show, and shadowed Process names) were reproduced with the reviewers' exact commands, fixed, and kept as regression tests.

Security

None.

Security review outcome

No unaddressed introduced findings.

Review coverage

Reviewed against issue-1456, including an independent reconciliation review after #1459 landed, pinned at f21175a26b908f5daf76b578687618a57cb7db76. Cumulative findings: Standards 1, Spec 3. All are addressed.

Both axes' final files-read lists cover all 17 changed files. Unread changed files: none. No generated output was excluded from reading.

Reports and final files-read lists are retained in .thirdshift-review-s3KxT3/standards.md, spec.md, and summary.json in the Session worktree.

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