brainstormlisted
Install: claude install-skill taipt1504/claudehut
# Brainstorm (phase 2 of 7)
Turn a grounded problem into **≥2 genuinely distinct, well-reasoned approaches** and a recommendation. This is
**general-purpose ideation** — feature, bug, refactor, performance, design, or non-code decision. It does NOT
explore the codebase or run a reuse-scan: that is **Discover** (phase 1), whose context + reuse DECISION this
phase consumes. Decoupling ideation from discovery is deliberate (v0.4 reversal) — forcing explore+reuse here
narrowed the option space; freeing it widens creative breadth.
Run **inline on the main thread** (it owns the state write and the user gate; a forked subagent cannot spawn
subagents).
## Flow
```mermaid
flowchart TB
s(["entered after Discover<br/>(context map + reuse DECISION)"]) --> disp["dispatch claudehut-brainstormer<br/>(problem + Discover context)"]
disp --> val{"return conforms to pipeline?<br/>≥2 distinct + scores tied to criteria +<br/>both premortems + option 0 if reuse candidate"}
val -- "no (missing piece)" --> disp
val -- "yes" --> enf["assemble enforcement set (1% rule):<br/>scan skills + .claude/rules/ tree;<br/>set-enforcement --skills --rules"]
enf --> write["write brainstorm.md from template<br/>(main thread writes; agent has no Write)"]
write --> gate{"set-brainstorm accepts?<br/>≥2 scored rows + Premortem + Recommendation"}
gate -- "no (freeform / thin)" --> write
gate -- "yes" --> mode{"interactive run?"}
mode -- "yes" --> ask(["AskUserQuestion: scored opt