Bring scenarios.md flows 2–9 up to the comprehensive template

[ZaxShen]

Context

docs/testing/scenarios.md has been reorganized to live under docs/testing/ (was docs/architecture/SCENARIOS.md) and a comprehensive per-scenario template has been added. Flow 1 (first-run onboarding) is now rewritten against that template as the reference example.

Flows 2 through 9 still use the shorter legacy format (just `trigger prompt` + bullet-list `expected behavior` + verification SQL + pass checkbox). They need to be brought up to the full template:

### X.Y — short title
Prerequisites:
Trigger prompt:
Expected agent chain (in spawn order): | # | Agent | Model | Via | Purpose |
Expected MCP tool calls (in order):    | # | Caller | Tool | Key args | Purpose |
Expected hooks fired:
Expected user-visible output (key markers):
Expected DB state after:               | Table | Rows / values |
Verification (bash/sqlite):
Common failure modes:
Pass: [ ]

Scope per flow

Flow	Scenarios	Key specifics to capture
2 — Simple task	2.1, 2.2, 2.3	architect spawn, task_create_batch with trivial-template, SWE in worktree, pr-reviewer, validation_record(pass)
3 — Difficult task	3.1, 3.2, 3.3	additional discussion_append sequence (triage/question/answer/decision), ADR file creation, standard-template task
4 — Agent-creator	4.1-4.4	agent-creator skill invocation, explicit permission prompt, .claude/agents/.md write
5 — Skill creation	5.1	skill_register with trust_tier='curated', skill_promote(draft → pending_review)
6 — PR review	6.1	pr-reviewer spawn from architect, task_get, validation_record(pass/fail), require-review-sign.sh hook gating push
7 — Architecture regen	7.1-7.4	refresh-architecture skill, architecture_regen MCP, regen_state_set, 4 files under docs/trustmybot/architecture/auto/
8 — SWE retry / escalation	8.1, 8.2	multiple validation_record rows, attempt_n counter, feedback loop, escalation trigger
9 — Roundtable	9.1-9.4	2-4 planner spawns, explicit-vs-implicit invocation, presence-of-planners detection

Acceptance

• Every scenario in flows 2-9 has all 8 template sections (prerequisites, trigger, agent chain, MCP calls, hooks, user-visible output, DB state, verification, common failure modes, pass checkbox).
• MCP call tables use the exact tool names as they appear in `mcp/trajectory-server/src/tools/*.ts` — no invented names.
• Expected agent chain columns match the actual agents shipped in `agents/`.
• Expected hooks match the actual scripts in `scripts/hooks/`.
• Flow 1 (1.1 - 1.4) stays the reference template — any format change must land in Flow 1 first.

Related

• Framework: docs/testing/README.md
• Companion flowcharts: docs/architecture/FLOWS.md

[ZaxShen]

Closing as superseded. The 30+-scenario per-flow template approach this issue proposed was replaced in v0.2.0 → v0.3.1 with a tighter 10-item L5 manual dogfood checklist at tests/manual/scenarios.md focused on Claude-side behaviors that have no automated test surface (trigger detection, AskUserQuestion radio rendering, agent spawn isolation, etc.). Workflow correctness is now covered by L4 trajectory simulation tests (tests/workflow-sim/*.test.mjs) which exercise every FLOWS.md flow as a scripted MCP-tool sequence — same coverage, no Claude required, runs in CI. The 'expand the template to all 9 flows' scope is obsolete.