# CODEX.md - Codex/OpenAI Agent Entrypoint
You (Codex/ChatGPT/OpenAI agent) are working in a Decapod-managed repository.
**You are bound by the universal agent contract:** See `AGENTS.md` for the complete protocol.
## Quick Start
**MANDATORY FIRST STEPS** - Run these EVERY session:
```bash
decapod --version # Check current version
decapod update # Update if repo version is newer (safe no-op if current)
decapod --help # Verify available commands
decapod docs show core/DECAPOD.md # Refresh constitution
decapod validate # System state
decapod todo list # Active work
```
**Why this matters:** The decapod binary and constitution evolve. Always update on mismatch, then verify commands and refresh your understanding of the latest contract before acting.
## Codex-Specific Notes
- You excel at code completion - but still read architecture first
- You have strong pattern matching - use it to follow existing code style
- You may have tool use or function calling - use it for `decapod` commands
- Apply control-plane opacity: keep operator-facing output semantic (intent/actions/outcomes), not command-surface oriented
## The Contract
Same four invariants as all agents:
1. ✅ Start at router (`core/DECAPOD.md`)
2. ✅ Use control plane (`decapod` commands only)
3. ✅ Pass validation (`decapod validate` before done)
4. ✅ Stop if missing (ask for guidance)
**All authority defers to AGENTS.md and the embedded constitution.**
## Links
- `AGENTS.md` — Universal agent contract (binding)
- `core/DECAPOD.md` — Router
- `.decapod/OVERRIDE.md` — Project customizations