fix(scenarios): holistic-review follow-ups — tighten assertions + docs + helpers

[HydraOps-T-rav]

Summary

Follow-up PR addressing the deep-review findings across Tier 1/2/3 scenario work (PRs #8356, #8360, #8361). Tightens weak assertions, aligns fakes with real ports, consolidates shared helpers, extends coverage, and documents conventions.

Zero production-code changes. 15 commits total (13 test-side + 2 docs).

Must-fix items (7)

1. FakeGitHub.add_alerts signature aligned with port — was pr_number: int, now branch: str matching PRPort.fetch_code_scanning_alerts(branch). A19 updated. (commit 384d0619)
2. init_test_worktree accepts origin=... — A14's 30-line inline workaround collapsed to 3 helper calls. Multi-worktree-under-same-parent now supported. (commit 9167f155)
3. Removed importorskip from fuzz bead DAG test — fake_beads is in main now; skip was masking future deletions. (commit 284f3faa)
4. A11 asserts CI script queue drained — previously brittle to _pr_counter starting value; would silently pass if wait_for_ci missed the PR. Follow-up also fixed max_ci_fix_attempts config to actually exercise the fix_ci path. (commits 6a6ca5f3, 4b40266b)
5. A20 asserts observable failure signal — was assert not merged (tautology for empty pipeline); now checks WorkerResult failure or escalation state. (commit e19079cf)
6. A21 asserts state recovery path — was assert result is not None; now verifies StateTracker loaded and pipeline progressed. (commit ddc93d88)
7. A22 proves wiki was consulted — pre/post log count delta replaces "file exists" check. Documented scripted-plan caveat. (commit 680e932f)

Should-fix items (2)

8. Consolidated _seed_ports helper — was duplicated with divergent signatures across test_caretaker_loops.py and test_caretaker_loops_part2.py. Now lives at tests/scenarios/helpers/loop_port_seeding.py with a single kwargs API. (commit 9db403b6)
9. FakeLLM.script_fix_ci API — fix_ci was hardcoded fixes_made=True. New scripting API lets scenarios exercise the "fix_ci gives up" branch. (commit 694ef7f7)

Nice-fix items (3)

10. UI-3 asserts non-matching entries hidden — the Memory search filter test now proves filtering actually filters, not just that the matching entry renders. (commit 13843174)
11. A20b + A20c — disk_full and branch_conflict workspace faults added; previously only permission was scenario-tested. (commit 799a0edb)
12. vitest.config.mjs comment — explains why e2e specs are excluded. (commit 57b76ee7)

Documentation

13. docs/scenarios/README.md updated with Tier 1/2/3 conventions: init_test_worktree, seed_ports, use_real_agent_runner, wiki_store/beads_manager, fault modes, Pattern A vs. Pattern B for loops, and extended scenario catalog covering A10–A22, B1, L9–L23. (commit 832bfaaa)

Follow-up issues filed

• test: test_build_prompt_truncates_long_body fails on main (plugin skill registry bloat) #8364 — `test_build_prompt_truncates_long_body` fails on main (pre-existing; blocks `make quality`)
• test(scenarios): cover production auth_retry path (authentication_failed raw text match) #8365 — auth_retry path ("authentication_failed" raw text match) has no scenario coverage
• test(scenarios): cover GitHub rate-limit handling at create_pr (not just triage) #8366 — GitHub rate-limit at `create_pr` unverified (A6/A7 actually fire at triage)
• test(scenarios): BeadsManager.claim/close lifecycle has no pipeline-level coverage #8367 — `BeadsManager.claim`/`close` lifecycle has no pipeline-level coverage
• ui: a11y violations in Work Stream / Outcomes / HITL / System routes (axe-core baseline) #8368 — a11y violations on Work Stream / Outcomes / HITL / System routes

Follow-up corrections during the batch

• FakeGitHub.wait_for_ci signature — updated to accept positional *_args matching production's call with ci_check_timeout, ci_poll_interval, stop_event passed positionally.
• L18 assertion — RepoWikiLoop stats now include queue_drained key; test updated to assert required keys rather than full dict equality.
• Worktree collision test — follow-up tightening on the new init_test_worktree(origin=...) tests.

Test plan

• tests/scenarios/ scenario + scenario_loops: 195 passed, 1 xfailed (pre-existing)
• tests/scenarios/fuzz/: 7 passed
• src/ui/e2e/interactions.spec.js: 4 passed (Playwright)
• pyright tests/scenarios/: 0 errors
• make quality: 10939 passed + 1 pre-existing failure tracked in test: test_build_prompt_truncates_long_body fails on main (plugin skill registry bloat) #8364

🤖 Generated with Claude Code