pre-plan-brainstormlisted
Install: claude install-skill Canhada-Labs/ceo-orchestration
# Pre-Plan Brainstorm — SKILL
## Why this exists
Our Plan → Debate → Execute protocol (`PROTOCOL.md` §Session protocol
Gate 3) consumes cycles in **Round 1 debate** resolving ambiguity that
could have been resolved pre-plan. superpowers (MIT framework, 159k
stars) battle-tested the pattern of **9-step brainstorm** as a
prerequisite to drafting the plan itself. superpowers audit BORROW-1
(ADR-058 PROPOSED) adopts this pattern.
**Same-LLM honesty:** this skill does not give the CEO independent
"expertise" it lacks — all agents are the same underlying model. What
it does is **force a checklist pass** that the CEO would otherwise
skip under time pressure, and **produce a durable spec.md artifact**
that survives across sessions and debate rounds.
## When to invoke
**Mandatory** (opt-out via `CEO_BRAINSTORM_GATE=0`):
- L3+ plan (`level: L3` or higher in frontmatter)
- Plan crosses 3+ modules OR 2+ subsystems
- Plan touches a VETO-protected domain (auth, financial, compliance)
- First-of-kind feature (no close precedent in the repo)
**Skip** (L1-L2 or well-precedented):
- Bug fix in 1-2 files
- Typo, log message, config tweak
- Change with an exact precedent already committed
- Mechanical refactoring (rename, extract, inline) with tests green
**Ambiguity smell-tests (any one triggers brainstorm):**
1. The request uses "should" / "probably" / "maybe" / "I think"
2. Success criteria are not enumerable as pass/fail tests
3. The reviewer of the PR cannot be named up-front
4. Roll