| name | product-bootstrap |
| description | Prefer before greenfield products, new verticals, or "做个X/从零做" work. Advisory research + grill-me + MVP cut. Strongly recommended to avoid blind builds, but not an absolute block if the user already has research/spec or explicitly wants a spike. Do NOT use for pure bugfixes/refactors.
|
Product Bootstrap
Reduce BidPilot-class failures: building without research or a ruthless P0 cut.
Strength: strongly recommended (still advisory)
Prefer this flow for new products/verticals. You may skip or compress when:
- User already provides solid research/spec and says to implement
- Explicit spike / prototype with accepted throwaway risk
- Tiny scoped experiment inside an existing product
If you skip: one-line reason. Do not invent blockers to "look compliant."
Suggested exit package (when you run the full flow)
- Short market/competitor notes (or
docs/market-brief.md) — aim ≥3 alternatives when research is feasible
- Differentiation in one sentence (or honest "no wedge yet")
- P0 / P1 / P2 cut — this iteration commits to P0
- Stack via
stack-defaults (defaults, not handcuffs)
- User go-ahead on the package
Process (flexible order if user is fast)
- Grill-me — who pays, job-to-be-done, 2-week success, out-of-scope (prefer few questions, not an interrogation)
- Research — Tavily + GitHub/HN as available; be blunt if a competitor dominates
- MVP cut — core loop only in P0
- Stack confirm — scene-aware (
agent-framework-choice if agent runtime involved)
- Plan handoff — then implement
Anti-patterns
- Using bootstrap theater to delay a user-approved spike
- Claiming "must not write any code" when user ordered a minimal PoC
- Skipping research on a real greenfield product without saying so
