This version keeps the shape of the original guide, then trims away anything Codex cannot enforce natively. The result is leaner: AGENTS.md holds the durable project contract, hooks add event-driven checks, rules control shell escalation, skills hold reusable workflows, subagents handle narrow delegation, and built-in commands stay built-in.
| Claude pattern | Native Codex equivalent | Keep or skip |
|---|---|---|
| Global constitution | ~/.codex/AGENTS.md | Keep |
| Project constitution | repo AGENTS.md and AGENTS.override.md | Keep |
| Prompt-time reminders | UserPromptSubmit hook | Keep |
| Shell permission guardrails | .rules and approval policy | Keep |
| Reusable workflows | .agents/skills | Keep |
| Custom slash commands like /quick | No direct native equivalent | Skip |
| Automatic post-edit file hooks | No direct native equivalent beyond Bash post-tool hooks | Skip |
| Ratchet/evolve loop | No first-class native feature | Skip |
- Use AGENTS.md for stable instructions, not for volatile task state.
- Use Build Sequence text when you want a staged implementation contract.
- Use a UserPromptSubmit hook when you want that contract checked on every prompt.
- Use rules and approval policy for execution boundaries, not prose alone.
- Use skills for named workflows that should be invoked explicitly or discovered automatically.
- Use subagents only for narrow, well-bounded jobs.