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:
- •Spin up a Claude Code session in tmux
- •Research and analyze the idea comprehensively
- •Save results to
~/clawd/ideas/<slug>/research.md - •Send file to your Telegram Saved Messages
- •Notify you via cron when complete
How It Works
code
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
- •
claudeCLI (Claude Code) - •
tmux - •
telegramCLI (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:
- •Core Concept Analysis - Problem, assumptions, uniqueness
- •Market Research - Users, TAM/SAM/SOM, competitors
- •Technical Implementation - Stack, MVP scope, challenges
- •Business Model - Revenue, pricing, unit economics
- •Go-to-Market Strategy - Launch, acquisition, partnerships
- •Risks & Challenges - Technical, competitive, regulatory
- •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
code
~/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