The Forge

# Forge Setup — Run from your clawbuddy-kit directory # 1. Deploy the edge function supabase functions deploy forge-analyzer --no-verify-jwt # 2. Set the required secret supabase secrets set OPENAI_API_KEY=your-openai-api-key # 3. Run the migration (creates forge_analyses table) supabase db push # That's it. Forge is ready to analyze content.

# Forge — AI Content Analyzer You have access to the Forge analyzer, an AI-powered tool that takes any content (YouTube transcripts, API docs, MCP specs, URLs, or plain text) and identifies every buildable item inside it. ## Endpoint ``` POST ${CLAWBUDDY_API_URL}/functions/v1/forge-analyzer ``` Auth: `x-webhook-secret: ${CLAWBUDDY_WEBHOOK_SECRET}` ## Actions ### Analyze Content Feed content to Forge and get back a structured list of buildable items. **From a transcript (with optional YouTube URL for metadata):** ```json { "action": "analyze", "input_type": "transcript", "content": "Paste the full transcript here...", "url": "https://youtube.com/watch?v=VIDEO_ID" } ``` **From a URL (content fetched automatically):** ```json { "action": "analyze", "input_type": "url", "url": "https://docs.stripe.com/api" } ``` **From API docs or MCP spec:** ```json { "action": "analyze", "input_type": "api_docs", "content": "Paste the API documentation here..." } ``` **From any text:** ```json { "action": "analyze", "input_type": "text", "content": "Any text content to analyze..." } ``` Input types: `transcript`, `url`, `api_docs`, `mcp_spec`, `text` ### Response Format Each analysis returns: - `summary` — Brief overview of the content - `source_type` — What kind of content was analyzed - `items[]` — Array of buildable items, each with: - `name` — Kebab-case identifier (e.g., "stripe-webhook-handler") - `type` — One of: skill, edge_function, ops_app, automation, make_scenario, tool - `description` — What it does - `complexity` — low / medium / high - `recommended_agent` — Which agent should build it - `recommended_model` — Best model for the job - `build_steps[]` — Step-by-step instructions ### List Past Analyses ```json {"action": "list", "limit": 10} ``` ### Get a Specific Analysis ```json {"action": "get", "analysis_id": "uuid-here"} ``` ### Get YouTube Video Metadata ```json {"action": "video_info", "url": "https://youtube.com/watch?v=VIDEO_ID"} ``` Returns title, author, videoId via oEmbed (no API key needed). ## Workflow 1. User shares a video, doc, or URL 2. You call Forge to analyze it 3. Review the buildable items returned 4. Present them to the user with complexity and recommendations 5. If approved, dispatch builds via the queue or start building directly ## Rules - Always include the full transcript/content — partial input gives partial results - For YouTube videos, include both the transcript AND the URL (URL enriches with metadata) - Store analysis results — they're saved in `forge_analyses` table automatically - Present results clearly: group by type, show complexity, highlight quick wins (low complexity items first)