simulatelisted

# /simulate — Reaction Simulator 12 personas. 3 rounds. They react to your announcement AND to each other — camps form, opinions shift, objections emerge. **Usage:** `/simulate [paste your announcement, or describe the situation]` --- ## How It Works Inspired by social simulation research (OASIS engine). The key insight: agents that interact with each other produce fundamentally different output than 12 independent prompts, because opinions shift when personas see each other's arguments. 1. **Context** — read product data, identify audience and channel 2. **Personas** — generate 12 with detailed profiles (MBTI, influence, bias) 3. **Round 1** — all 12 react independently to the announcement (parallel, each does 1 web search) 4. **Round 2** — each persona reads ALL Round 1 reactions, can shift stance, reply, form camps (parallel) 5. **Round 3** — final positions after seeing the full debate (parallel) 6. **Report** — opinion trajectory, camps, ranked objections, suggested rewrite ~36 agent calls total across 3 sequential rounds. --- ## Phase 1: Context ### Extract from input - **Text** — the announcement (`$ARGUMENTS`, or ask) - **Product** — what product/company is this about? - **Audience** — who sees this? - **Channel** — where is it published? (email, Twitter, blog, HN, internal) ### Read product context silently - Check for `README.md`, `CLAUDE.md`, or docs in the current project for product details - If a URL is provided, fetch it with web tools - Look for: cu