codex-guide-introspect

name	codex-guide-introspect
description	Audit a repository's Codex readiness and propose the next high-value improvements. Use when the user wants to review AGENTS.md, commands, verification, docs quality, CI, MCP usage, worktrees, or exercise ideas before deciding what to improve next.

Codex Guide Introspect

Inspect the repository first, then produce a structured improvement plan. Do not make repo changes until the user explicitly asks you to apply part of the plan.

Workflow

1. Ground in the repo

Inspect the current repository before giving advice.

Look for:

• AGENTS.md quality and coverage
• build, test, lint, and site commands
• docs structure and internal consistency
• local skills and whether their descriptions still match reality
• CI workflows and validation coverage
• MCP, worktrees, review, and automation signals where relevant

2. Evaluate the current Codex workflow

Assess whether the repo supports a clean inspect-plan-implement-verify loop.

Focus on:

• whether Codex can discover the right commands quickly
• whether validation is documented and runnable
• whether instructions are specific instead of generic
• whether site and docs structure are legible to both humans and Codex
• whether there are obvious missing exercises or stories

3. Produce a plan-first report

Return a concise audit with:

• strengths already present
• gaps or risks
• the next 3 to 5 improvements in priority order
• which changes are small enough to do now versus issue-worthy

Use this shape:

## Codex Readiness Audit

### Strengths
- ...

### Gaps
- ...

### Recommended next steps
1. ...
2. ...
3. ...

### Verification
- ...

4. Only apply after approval

If the user explicitly asks you to implement one of the recommended changes, do that work next. Keep the implementation scoped to the approved items.

Response rules

• Ground the audit in files and commands from the repo.
• Prefer specific fixes over abstract advice.
• Treat hooks as documented but experimental. Do not assume they replace skills, automations, or broader policy enforcement.
• Do not invent Claude-only features such as plugins or sub-agents.
• If the user wants deeper explanation of a specific Codex feature, suggest $codex-guide-ask.
• If the user wants guided practice, point them to the missions under docs/exercises/.