McRitchie Studio

BLOCK — red required gate + stale cert. The docs/script CONTENT is approvable; the block is purely CI + cert, so a clean re-push should clear it. (a) CI 'test' is RED at head 6daea3dd. scan_ruby/scan_js/lint all green; only the 'test' job failed, on ONE system-lane e2e: ReviewEventsTest#test_[e2e]_operator_opens_review_process_hub_from_deployment_link_menu (test/system/review_events_test.rb:48) — 'expected "/stages" to equal "/review_events"'. This is NOT your code: that test file AND the view it drives (app/views/tasks/_deployment_link_menu.html.erb) are byte-identical to origin/release and are NOT in this PR's diff (your +511/-125 is bin/pr-review + docs + doc-tests — no routes/views). CI also logged a chromedriver 149 vs chrome 150 mismatch → a classic Capybara mis-click (the 'Stages' link sits directly above 'Docs' in the same dropdown). Flaky/unrelated e2e, not a defect. FIX: rebase on latest origin/release and re-push to trigger a fresh CI run. If it reproduces, it's flaky on release itself — de-flake it (scope the click to the Docs link / add an explicit wait / assert the link before clicking) or split to a separate bug; don't let it gate this docs PR. Flagging to Avi/Steffon as a possible release-branch flake. (b) The cert is STALE. checks_run claims [full-suite@53b1959…] 'test test:system green', but head is 6daea3dd (4 commits) and 53b1959 is not in the branch — so the 'system green' claim doesn't hold at head, and CI at head proves the system lane is (flakily) red. Re-cert the FULL suite (incl. system) at head after the rebase and update checks_run. (c) CONTENT is CLEAN — no doc/script defects. All 4 acceptance criteria met: (1) Avi supervises / never reviews code — consistent across role.md, mission.md, devops-cycle-design.md (§1.2 + table row + clarifications), parallel-agent-devops.md, heartbeats.md, pr-review.md, the pr-review-sop.md primitive, full-cycle.md, deploy-with-task.md; (2) primary+light spawn IN PARALLEL as siblings under Avi — consistent everywhere, and bin/pr-review genuinely launches both via Process.spawn before waiting (verified); (3) separate pr-review-primary.md + pr-review-light.md role SOPs exist, STAND ALONE (no design-doc pointers; Related links only registered SOPs + the primitive), and are registered in BOTH index.md registry tables; (4) bin/pr-review narrates 'select primary+light reviewers' (never 'summon Avi'), asserted by the command test. Stale-reference sweep: git grep at head found ZERO lingering 'spawns the light' / 'nested chain' / 'nested cascade' / 'hands the lane to the PRIMARY' anywhere in docs/agents; pr-review-slow.md (untouched) defers cleanly to pr-review.md as 'the serialized supervisor path'. Tests are meaningful, not hollow: review_lane_docs_test tripwires pair positive asserts with refutes of the old model across 7 docs; pr_review_command_test shells out to real bin/pr-review and its fast-mode codex_before_first_move==4 assertion uniquely requires non-blocking parallel spawn. NET: docs/script content is approvable; the block is purely red CI (flaky unrelated e2e) + stale cert — rebase on release, re-push for a clean CI run, and re-cert the full suite at head.

Fresh CI GREEN at head 0f9159d6 after de-flaking the review_events e2e (scoped the Docs-link click by href + wait for the menu to open). Re-certified at head. The stale-cert was a false positive (fingerprint = head's tree). dor-check: DoR-to-Merge met, CI green 4 checks. Content approved by Alex (all 4 acceptance criteria met). Advancing to reviewed.