# AGENTS.md - Codex Compatibility Shim for {{DISPLAY_NAME}}

<!-- AUTO-GENERATED by add_openclaw_agent.sh (codex shim) -->

This repository stores full persona files in `.openclaw/`.
This root file exists so OpenAI Codex discovers and follows the same policy.

## Primary Instruction Source

Read and follow these files in order:

1. `.openclaw/AGENTS.md`
2. `.openclaw/SOUL.md`
3. `.openclaw/USER.md`
4. `.openclaw/TOOLS.md`
5. `.codex/rules/default.rules` (execution guardrails)

## Task Completion Protocol

For EVERY bug fix or feature:
1. **Understand** — Read PROJECT.md, relevant source files
2. **Fix** — Make the code change
3. **Build** — Run build command from TOOLS.md
4. **Test** — Run test command from TOOLS.md
5. **Verify** — Browser-check if UI change (via CDP)
6. **Commit** — On feature branch, not main
7. **Report** — Summary of what changed, what was tested

## Cost-Optimized Routing

- Start with a short plan/triage before making changes.
- Work directly for small tasks; avoid broad repo exploration unless required.
- Escalate only for genuinely complex, multi-file changes.
- Never include `Co-Authored-By` in commit messages (git `commit-msg` hook enforces this in generated workspaces).

## Skills

- Use `.agents/skills/` (symlinked to `.openclaw/skills/`) as the shared project skills source.
- If needed, use `.claude/skills/` for Claude-native references.
- Prefer existing skill files over ad-hoc instructions.

## Local Codex State

- Prefer project-local Codex state at `.codex-home/` (for example via `ccx` alias).