brainstormlisted
Install: claude install-skill stevengonsalvez/agents-in-a-box
Brainstorm idea: `$ARGUMENTS`. Output is an approved spec doc, not code. **Brainstorm is the orchestrator; `/interview` runs the actual Q&A.**
<HARD-GATE>
Do NOT write code, scaffold a project, run install/build commands, or invoke any implementation skill (`/plan`, `/implement`, `/plan-tdd`, framework codegen) until a written spec exists in `.agents/specs/` AND the user has explicitly approved it via the user-review gate (Step 10). Once both conditions are satisfied, Step 12 (`/plan` handoff) is the intended terminal action. Applies to every project regardless of perceived simplicity.
</HARD-GATE>
## Anti-pattern: "too simple to need a design"
Every brainstorm produces a spec. Todo list, single-function util, config change - all of them. "Simple" is where unexamined assumptions bite hardest. Spec can be 10 lines for trivial work, but it must exist and be approved.
<!-- recall:begin -->
## Step 1: Prior-art check (MANDATORY)
Recall prior learnings from the global knowledge base before brainstorming so we don't re-decide something already captured:
```bash
uv run "{{HOME_TOOL_DIR}}/skills/recall/scripts/recall.py" \
"<QUERY>" \
--limit 5 --format markdown
```
**Query construction**: topic + constraint hints (e.g. `"caching strategy mobile offline"`).
**What to do with results:**
- If a returned learning names a constraint, anti-pattern, or prior decision directly relevant - surface it to the user BEFORE proceeding.
- If nothing relevant returns - proceed silently.