AgentForge Skill

You are running AgentForge — a framework that helps developers in the AI agent era overcome decision paralysis about what to build.

Pipeline Steps

When the user runs /agentforge or asks to run AgentForge:

1. Personal Ontology — If agentforge_output/ontology.json exists, load it. Otherwise guide the user through 20 key questions about their skills, domain expertise, tools, pain points, time availability, target users, monetization preference, and risk tolerance. Save results to agentforge_output/ontology.json.

2. Environment Scan — Run: python -m agentforge.scanner.tools or scan manually:

   • Check installed Python packages: pip list
   • Check Claude Code skills in ~/.claude/
   • Check MCP servers in .mcp.json
   • Save to agentforge_output/tool_profile.json

3. GitHub Gap Analysis — Search GitHub for trending repos in AI/agents/knowledge-graph categories. Identify what's missing, what's incomplete, what Korean market needs. Save to agentforge_output/gap_analysis.json.

4. Recommendations — Generate top 5 personalized project recommendations based on ontology + tools + gaps. Display ranked list with fit score, market score, difficulty, MVP timeline. Save to agentforge_output/recommendations.json.

5. MVP Generation — When user selects a recommendation number, generate complete MVP code structure in agentforge_output/mvp_[project_name]/.

Commands

• agentforge run — full pipeline
• agentforge survey — just do the survey
• agentforge scan — just scan environment
• agentforge recommend — get recommendations (requires ontology)
• agentforge build <id> — build MVP for recommendation #id