adversarial-brainstormlisted

# Adversarial Brainstorm ## What it is A multi-agent dialectic. Instead of one model emitting a list, ideas are **generated by diverse personas, attacked by a red team, defended by champions, and synthesized into a ranked path**. Ideas that survive are stronger because they had to survive — the synthesis is decision-ready, not a brainstorm dump. ``` FRAME (1 agent, blocking) └─ sharpen the real problem → success criteria, constraints, non-goals, unknowns DIVERGE (N persona ideators, parallel) └─ each lens emits K ideas; dedup by title ATTACK → DEFEND (pipeline, per idea) ├─ M red-team critics try to KILL each idea on a distinct lens ├─ kills ≥ killThreshold → dropped (with recorded objections) └─ survivors get a champion: strengthened pitch + rebuttal + cheapest first test SYNTHESIZE (1 agent, blocking) └─ rank shortlist (effort/risk/score/first-test) + 1-3 hybrids + one recommendation + open questions ``` ## When to use vs alternatives | Need | Use | |---|---| | Generate + pressure-test ideas for a wide-open design question | **this skill** | | Pin down vague requirements into a spec via Q&A | `deep-interview` | | Turn a chosen direction into an implementation plan | `planner` / `ralplan` | | Compare a few already-known options on trade-offs | `architect` | | Audit an existing codebase | `multi-agent-codebase-audit` | Brainstorm explores *what to build*; deep-interview clarifies *what you already decided*; planner sequences *how*. Chain them: **adversarial-