Codex Skill Guide

Use Codex proactively for complex architectural decisions, deep code analysis, refactoring, security/performance audits, code reviews before commits/PRs, when stuck after 2+ failed attempts, or when the user explicitly requests GPT/Codex assistance.

Models

Role	Model
Primary	gpt-5.4
Fallback	gpt-5.3-codex

Default to gpt-5.4. Fall back to gpt-5.3-codex if the user requests it or if gpt-5.4 fails.

---

Running codex exec

1. Ask the user (via AskUserQuestion) which model (gpt-5.4 or gpt-5.3-codex) AND which reasoning effort (xhigh, high, medium, or low) in a single prompt. Default suggestion: gpt-5.4 + high.

2. Always use -s read-only — codex is primarily a validator/analyzer. Only escalate sandbox if the user explicitly asks codex to edit files or needs network access:

   • User asks to edit/refactor → -s workspace-write --full-auto (confirm via AskUserQuestion first)
   • User needs network → -s danger-full-access --full-auto (confirm via AskUserQuestion first)

3. Assemble the command:

   • -m <MODEL> — model selection
   • -c model_reasoning_effort="<effort>" — reasoning effort
   • -s read-only — default sandbox (see step 2 for exceptions)
   • -o /tmp/codex-last.txt — always capture full response
   • -C <DIR> — working directory
   • --skip-git-repo-check — always include

4. Append 2>/dev/null to suppress thinking tokens (stderr). Only show stderr if debugging.

5. Run the command, then read /tmp/codex-last.txt with the Read tool to get the full response (Bash output may truncate). Summarize for the user.

6. After completion: "You can resume this session at any time by saying 'codex resume'."

Quick Reference - exec

Use case	Command
Analysis/validation (default)	codex exec --skip-git-repo-check -m MODEL -s read-only -o /tmp/codex-last.txt "prompt" 2>/dev/null
Apply local edits	codex exec --skip-git-repo-check -m MODEL -s workspace-write --full-auto -o /tmp/codex-last.txt "prompt" 2>/dev/null
Network/broad access	codex exec --skip-git-repo-check -m MODEL -s danger-full-access --full-auto -o /tmp/codex-last.txt "prompt" 2>/dev/null
Different directory	codex exec --skip-git-repo-check -C <DIR> -m MODEL -s read-only -o /tmp/codex-last.txt "prompt" 2>/dev/null

---

Resuming Sessions

resume works both as top-level command and as exec subcommand:

Mode	Command
Interactive resume (picker)	codex resume
Interactive resume (last)	codex resume --last
Non-interactive resume	echo "prompt" | codex exec --skip-git-repo-check resume --last 2>/dev/null

• Don't use config flags on resume unless explicitly requested.
• For codex exec resume, flags go between exec and resume.

---

Running codex review

No need to ask for model/reasoning — review uses sensible defaults.

1. Ask the user (via AskUserQuestion) what to review:

   • Uncommitted changes — all local modifications
   • Against a branch — compare to base branch (e.g., main)
   • Specific commit — review a commit SHA

2. Optionally ask for custom review focus (security, performance, conventions, etc.)

3. Run the appropriate command and summarize findings.

Quick Reference - review

Use case	Command
Uncommitted changes	codex review --uncommitted
Against main branch	codex review --base main
Against specific branch	codex review --base feature/xyz
Specific commit	codex review --commit abc123
Commit with title context	codex review --commit abc123 --title "Add user auth"
With custom focus	codex review --uncommitted "Focus on security"

Note: --title requires --commit (not --base).

Configuration Overrides

codex review --uncommitted -c model="gpt-5.3-codex"

---

Decision Guide: exec vs review

Situation	Use
"Review my changes"	codex review --uncommitted
"Review this PR"	codex review --base main
"Analyze this architecture"	codex exec
"Help me debug this"	codex exec
"Refactor this code"	codex exec with workspace-write
"Is this secure?"	codex review for diffs, codex exec for deep analysis

---

Following Up

• After every command, use AskUserQuestion to confirm next steps
• For exec: offer to resume with codex resume --last
• For review: offer to help address specific findings

Error Handling

• Stop and report failures on non-zero exit; request direction before retrying
• Summarize warnings or partial results and ask how to adjust