idea

# Idea Exploration Skill Launch autonomous Claude Code sessions to explore business ideas in depth. Get market research, technical analysis, GTM strategy, and actionable recommendations. ## Quick Start **Trigger phrase:** Say `Idea: [description]` and the assistant will: 1. Spin up a Claude Code session in tmux 2. Research and analyze the idea comprehensively 3. Save results to `~/clawd/ideas/<slug>/research.md` 4. Send file to your Telegram Saved Messages 5. Notify you via cron when complete ## How It Works ``` User: "Idea: AI calendar assistant" ↓ ┌─────────────────────────────────┐ │ 1. explore-idea.sh starts │ │ 2. Creates tmux session │ │ 3. Runs Claude Code │ │ 4. Claude analyzes & writes │ │ 5. notify-research-complete.sh │ │ → Sends file to "me" │ │ → Queues notification │ │ 6. Cron checks queue (1 min) │ │ 7. Notifies user in chat │ └─────────────────────────────────┘ ``` ## Setup ### Prerequisites - `claude` CLI (Claude Code) - `tmux` - `telegram` CLI (supertelegram) - Clawdbot with cron enabled ### 1. Create Scripts See `~/clawd/scripts/explore-idea.sh` for the full implementation. Key components: - Creates idea directory with prompt and runner script - Unsets OAuth env vars to use Claude Max - Runs claude with `--dangerously-skip-permissions` - Calls notify script on completion ### 2. Set Up Cron Job ```bash # Cron job to check notification queue every minute { name: "Check notification queue", sessionTarget: "isolated", wakeMode: "now", payload: { kind: "agentTurn", message: "Check ~/.clawdbot/notify-queue/ for .json files...", deliver: true, channel: "telegram", to: "YOUR_CHAT_ID" }, schedule: { kind: "every", everyMs: 60000 } } ``` ### 3. Add AGENTS.md Instructions ```markdown **When user says "Idea: [description]":** 1. Extract the idea description 2. Execute: `CLAWD_SESSION_KEY="main" ~/clawd/scripts/explore-idea.sh "[idea]"` 3. Confirm: "Idea exploration started. You'll be notified when complete." ``` ## Analysis Framework The exploration covers: 1. **Core Concept Analysis** - Problem, assumptions, uniqueness 2. **Market Research** - Users, TAM/SAM/SOM, competitors 3. **Technical Implementation** - Stack, MVP scope, challenges 4. **Business Model** - Revenue, pricing, unit economics 5. **Go-to-Market Strategy** - Launch, acquisition, partnerships 6. **Risks & Challenges** - Technical, competitive, regulatory 7. **Verdict & Recommendations** - Clear yes/no with action plan ## Verdict Types - 🟢 **STRONG YES** - Clear opportunity, pursue aggressively - 🟡 **CONDITIONAL YES** - Promising but needs validation - 🟠 **PIVOT RECOMMENDED** - Core insight good, execution needs work - 🔴 **PASS** - Too many red flags ## Example Output ``` ~/clawd/ideas/ai-calendar-assistant/ ├── metadata.txt ├── prompt.txt ├── run-claude.sh └── research.md # 400-500 line comprehensive analysis ``` ## Tips - Ideas typically take 3-5 minutes to analyze - Monitor progress: `tmux attach -t idea-<slug>-<timestamp>` - File goes to Saved Messages even if notification fails - Check `~/.clawdbot/notify-queue/` for stuck notifications