test(sim): pipeline simulation — engine driven end-to-end as a black box

[leeovery]

Summary

• New permanent suite tests/scripts/test-pipeline-simulation.cjs (registered in npm test): drives the engine CLI as a black box through the call sequences the skill prose prescribes, in prose order — extracted per-phase from the entry/process skills.
• Thirteen scenarios: feature/bugfix/quick-fix/epic mainlines, quick-fix promotion, grouping supersession (terminal guards asserted), topic cancel/reactivate (map order round-trip), reopen/going-backwards, work-unit complete/reactivate/cancel loops, pivot, absorption, cross-cutting spec promotion, the implementation task loop (fix-attempt, analysis-cycle, gate-mode session reset, completed_tasks dedupe), skipped-review completion, and mid-pipeline guard refusals.
• After every mutation the full state is audited: manifests parse and are schema-valid, discovery items stay status-less, no phase-named shadow roots, every derivation (lifecycle, phaseStatus, unit phase state) computes, the bridge + all six navigation gateways discover AND format the state, and render surfaces (entry/resume/early-completion/revisit/epic-all-done/phase-completed/phase-note) hold at checkpoints. Refusal paths assert clean {ok:false} JSON and untouched state.
• The suite already earned its keep during authoring: it caught that entry-gate renders empty on pass and that workunit promote refuses incomplete specs — both correct behaviours my draft scenarios had wrong, exactly the feedback loop the suite exists to give.

Test plan

• npm test — 1590 pass (suite adds 13 scenarios, ~2s runtime).
• Conventions lint, typecheck green.

🤖 Generated with Claude Code

Stack

1. feat(engine): discovery-map add-batch — the harvest persists in one call (D7) #505
2. feat(engine): manifest apply — batched set/delete across one work unit (D7) #506
3. refactor(prose): task-map family folds onto existing batch forms (D7, stage 3) #507
4. feat(engine): manifest set — two grammars, never mixed #508
5. refactor(prose): dump/read folds — subtree gets replace per-topic loops (D7, stage 5) #509
6. feat(engine): commit --plan — plan storage stages from recorded state (D7, stage 6) #510
7. fix: batching-stack review round — re-resolve rewired, guards symmetric, hardening pinned #512
8. fix: deep-dive round 2 — backbone rule flipped, pre-upgrade paths routed, flow bugs closed #513
9. fix(review-r3): engine hardening + crash-resume seams #514
10. feat(scoping): promotion routes through first-phase choice + bridge #515
11. fix(spec): pin the bugfix source name to the topic #516
12. fix(skills): gate-skip sentence accurate in gate-less phases #517
13. fix(review-entry): name the real upstream — specification and plan #518
14. fix(shared): strip triage-landing's unreachable superseded-research arm #519
15. feat(engine): refuse work-unit-level fields that shadow the phases tree #520
16. fix(discovery): first-written briefs join the full brief lifecycle #521
17. test(sim): pipeline simulation — engine driven end-to-end as a black box #522 👈 current
18. fix(round-4): review findings — backfill keying, ratchet evasion, pull guard #523