What problem does it solve?
Interactive requirement discovery before planning starts. Use when the user has an idea or loose requirement and wants to sharpen the goal, scope, and approach before anyone plans tasks or writes code. Produces a single markdown document that survives context reset.
Core Features & Use Cases
- Preflight — new doc: confirm the path, derive a slug. Resume: read the existing doc in full, investigate everything it references, identify the next open question. Brownfield sweep either way: README, manifest, docs, directory tree, recent commits.
- The Doc — create docs/discovery/discovery_<slug>.md with the skeleton.
- Discovery conversation — section-by-section: Goal → Scope (in) → Scope (out) → Approach / Hypothesis → Open Questions. One question at a time; write the user's answer in immediately.
- Finalise — user signals done → honest review → flip Status to Ready for Planning. Hand off to /plan.
- Brownfield discovery: spawn sub-agents for research in parallel exactly as the main loop requires for brownfield discovery. Wait for all to complete.
- Research rigor: verify claims, spawn sub-agents in parallel; surface context the user may not know and surface evidence in the doc.
- Finalising: when the user signals discovery is done, review the doc honestly before accepting the call. Surface remaining vagueness and let the user decide how to address it.
Quick Start
Create a new discovery doc for your idea and follow the Goal, Scope in, Scope out, Approach, and Open Questions steps.