Agent First Codebase · soulfaro bundle Diagnose and transform a codebase so coding agents (Claude Code, Codex, and similar) can work in it reliably and autonomously. Based on the practices OpenAI's Codex team used to ship a 1M-line internal product with zero human-written code. Use this skill whenever the user wants to set up or restructure AGENTS.md / CLAUDE.md, organize a docs/ directory for AI agents, diagnose why Claude Code keeps making mistakes in their repo, make their codebase "AI-friendly" or "agent-first", references "harness engineering" or Codex best practices, audits an existing repo's agent-readability, or is starting a new repo they plan to develop heavily with AI assistance. Also trigger if the user complains that their agent "forgets context", "reinvents patterns", or "ignores conventions" — those are symptoms this skill addresses at the structural level.
