Codex CLI (Beta)

devcoach runs the full coaching loop in OpenAI's Codex CLI: the MCP server, automatic lesson cues after each task (Codex's Stop/UserPromptSubmit hooks use the same wire protocol as Claude Code's), and the devcoach agent skill. Requires Node.js ≥ 24 and Codex CLI ≥ 0.114 (hooks).

:::info Beta The Codex CLI integration is beta: the coaching engine is the same battle-tested core used with Claude Code, but the Codex hook wiring is newer. Please report anything odd on GitHub issues. :::

One command

npx -y devcoach install --codex

This registers everything devcoach needs:

• MCP server — via codex mcp add when the codex CLI is on your PATH. (Codex stores MCP servers in ~/.codex/config.toml; if codex isn't found, devcoach prints the exact [mcp_servers.devcoach] snippet to paste rather than editing your TOML.)
• Hooks — Stop (lesson cue after each task) + UserPromptSubmit (priming) in ~/.codex/hooks.json
• Skill — the devcoach coaching skill into ~/.agents/skills/devcoach/ (Codex's user-level Agent Skills directory)

:::note Trust the hooks once Codex asks you to trust newly configured hooks the first time they run — approve the devcoach hooks on your next codex session or they stay inert. :::

Check the wiring anytime with npx -y devcoach doctor, and re-run the install after upgrading devcoach to refresh the skill.

How pacing works

Identical to Claude Code: lessons are rate-limited (default max 2/day, 4h gap) and paced every N interactions per session. Codex sends the same stop_hook_active loop guard and permission_mode field Claude Code does, so hook-forced retries never re-cue and plan-mode turns stay quiet.

→ Next: Coaching in your agent.