Codex agent
Use when you want a second-opinion review via Codex CLI, cross-verification after another agent implements changes, debugging help, or alternative implementation proposals. Requires Codex CLI to be installed and authenticated.From its SKILL.md
npx -y skills add majiayu000/spellbook --skill codex-agentAssembled from the repository path, not quoted from the project. Check it against their README if it does not work.
One thing to look at
- runs commandsInstructs the agent to run 8 commands, including `REPORT="$(mktemp -t codex-review.XXXXXX.md)"` and 7 more.
SKILL.md
7.4 KB, ~1.8k tokens by cl100k_base, as published. Nobody here has run it
Codex Agent Collaboration Skill
This skill enables Claude Code to collaborate with OpenAI's Codex CLI agent for second-opinion review, cross-verification, debugging analysis, and alternative implementation proposals.
Default posture: Codex reviews in read-only; the primary agent applies changes only when the user asked for fixes or approved them after reading the review.
Optional Codex Review Workflow
Use this workflow when the user asks for Codex review, wants a second opinion, or needs cross-verification from a separate coding agent.
Step 1: Call Codex and Read Feedback
REPORT="$(mktemp -t codex-review.XXXXXX.md)"
codex exec -C <project_path> -s read-only -o "$REPORT" \
"Review the code in <file_or_directory>. Check for:
- Security vulnerabilities
- Performance issues
- Code quality and best practices
- Potential bugs and edge cases
- Naming and readability
Provide specific, actionable feedback with file paths and line numbers."
cat "$REPORT"
Keep the write and read in the same Bash call, or pass a concrete report path between calls; shell variables do not persist across tool calls.
Step 2: Apply Fixes Based on Codex Feedback
When the user asked to apply fixes, handle each issue identified by Codex:
- Read the relevant file
- Apply the fix using Edit tool
- Verify the fix addresses Codex's concern
If the user only asked for a review or second opinion, report findings without editing files.
Step 3: Re-verify with Codex (Optional)
codex exec -C <project_path> -s read-only \
"Verify the fixes applied to <files>. Confirm issues are resolved."
Workflow Examples
Example 1: Review and Fix a Single File
# Step 1: Get Codex review
REPORT="$(mktemp -t codex-review.XXXXXX.md)"
codex exec -C /project -s read-only -o "$REPORT" \
"Review src/auth/login.ts for security vulnerabilities and code quality issues. Provide specific line numbers and fixes."
# Step 2: Read the feedback
cat "$REPORT"
Then the primary agent reads the feedback, applies fixes with Edit tool, and optionally re-verifies.
Example 2: Review Recent Changes
# Get diff of recent changes
DIFF_REPORT="$(mktemp -t recent-changes.XXXXXX.diff)"
git diff HEAD~1 > "$DIFF_REPORT"
# Step 1: Have Codex review the diff
REPORT="$(mktemp -t codex-review.XXXXXX.md)"
codex exec -C /project -s read-only -o "$REPORT" \
"Review the changes saved at $DIFF_REPORT. Check for bugs, security issues, and improvements needed."
# Step 2: Read and apply fixes
cat "$REPORT"
Example 3: Full Project Review
# Step 1: Comprehensive review
REPORT="$(mktemp -t codex-review.XXXXXX.md)"
codex exec -C /project -s read-only -o "$REPORT" \
"Perform a comprehensive code review of src/. Focus on:
1. Security vulnerabilities (OWASP Top 10)
2. Error handling patterns
3. Performance bottlenecks
4. Code duplication
Prioritize issues by severity (critical/high/medium/low)."
# Step 2: Read prioritized feedback
cat "$REPORT"
Review Request Format
When asking Codex for review, include:
Review <target_files_or_directory>.
Context:
- Project type: <TypeScript/Python/etc>
- Framework: <Express/React/etc>
- Focus areas: <security/performance/quality>
Check for:
1. Security vulnerabilities
2. Performance issues
3. Error handling
4. Code quality
5. Edge cases
Output format:
For each issue:
- File: <path>
- Line: <number>
- Severity: critical/high/medium/low
- Issue: <description>
- Fix: <specific code change>
Applying Fixes
After receiving Codex feedback, apply fixes systematically:
- Parse the review - Extract each issue with file, line, severity
- Prioritize - Fix critical/high issues first
- Read file - Use Read tool to see current code
- Apply fix - Use Edit tool with precise old_string/new_string
- Track progress - Mark each issue as fixed
Prerequisites
Codex CLI must be installed and authenticated:
# Install via npm
npm install -g @openai/codex
# Or via Homebrew (macOS)
brew install --cask codex
# Authenticate
codex login
Command Reference
Basic Command Pattern
codex exec [options] "<task_description>"
Core Options
| Option | Description |
|---|---|
"<task>" | Task description (positional, must be quoted) |
-C <dir> | Working directory (use absolute path) |
-s read-only | Read-only sandbox (use for reviews) |
-o <path> | Save output to file |
--json | Output as JSON Lines |
AI-to-AI Communication
When communicating with Codex, PRIORITIZE ACCURACY AND PRECISION:
- Use structured data and exact technical terms
- Provide full file paths and precise details
- Include relevant context from the current codebase
- NO conversational formatting needed
Other Use Cases
Cross-Verification (after Claude implements)
codex exec -C /project -s read-only \
"Verify the implementation in src/feature/. Check correctness and edge cases."
Get Alternative Implementation
REPORT="$(mktemp -t codex-alternative.XXXXXX.md)"
codex exec -C /project -s read-only -o "$REPORT" \
"Propose an alternative implementation for the caching in src/cache/manager.ts"
cat "$REPORT"
Debugging Assistance
codex exec -C /project -s read-only \
"Debug: tests in tests/auth.test.ts failing with timeout. Analyze root cause."
Session Management
For multi-turn reviews:
# Initial review
codex exec -C /project -s read-only "Review src/api/ for security issues"
# Note session ID from output
# Follow-up after fixes
codex exec resume <session_id> "I've applied the fixes. Please re-verify."
Helper Scripts
scripts/check-codex.shchecks whether the Codex CLI is installed and authenticated.scripts/codex-wrapper.shis optional. Use it only when you need a small CLI wrapper; it executes Codex through shell arrays and must not useeval.
Gotchas
- Do not write reviews to fixed paths such as
/tmp/codex-review.md; usemktempor a project-specific private report path so concurrent projects cannot overwrite or read stale feedback. - Keep Codex review commands in
read-onlyunless the user explicitly asked Codex itself to edit. - Treat
danger-full-access,--dangerously-bypass-approvals-and-sandbox,--dangerously-bypass-hook-trust, and--skip-git-repo-checkas high-impact flags. Ask before using them. - Do not blindly apply every Codex suggestion. Re-read the target file, confirm the root cause, and verify the final behavior.
Troubleshooting
Authentication Issues
codex logout
codex login
Check Installation
codex --version
which codex
See Also
- scripts/check-codex.sh - Local Codex CLI availability check
- scripts/codex-wrapper.sh - Wrapper for repeatable Codex CLI invocation
- sandbox-modes.md - Sandbox security levels
- examples.md - More usage examples
- advanced.md - Advanced configuration
What ships with it: 6 files
19.5 KB alongside SKILL.md, 2 of them executable
scripts/
- check-codex.shruns1.0 KB
- codex-wrapper.shruns3.5 KB
- advanced.md3.8 KB
- examples.md4.3 KB
- README.md4.9 KB
- sandbox-modes.md2.0 KB
Gives 0 of the 12 instructions most context ai engineering skills give in ~1.8k tokens
Counted across 1,328 of the 2,349 authors here whose files we hold, read 2026-09-06
- Dispatch a fresh subagent for each taskin 76 of 1328, across 59 files
- Perform spec compliance review before code quality reviewin 44 of 1328, across 34 files
- Dispatch a final code reviewer after all tasksin 38 of 1328, across 26 files
- Answer subagent questions before allowing implementationin 36 of 1328, across 26 files
- Use the least powerful model capable of the taskin 33 of 1328, across 26 files
- Create a TodoWrite list for all tasksin 32 of 1328, across 22 files
- Perform a task review after each implementationin 31 of 1328, across 24 files
- Extract all tasks and context from the planin 29 of 1328, across 20 files
- Provide full task text to subagentsin 28 of 1328, across 20 files
- Use git worktrees for isolated workspacesin 25 of 1328, across 20 files
- Specify the model explicitly when dispatching a subagentin 23 of 1328, across 18 files
- Execute all tasks from the plan without stoppingin 21 of 1328, across 16 files
Said here and by no other author read
- run codex in read-only mode for reviews
- use mktemp for report file paths
- apply fixes using the edit tool
- include project context in review requests
- ask user before using high-impact flags
- confirm root cause before applying suggestions
Grouped from the skills themselves: near-identical wordings counted once, and counted by distinct author, so one author publishing three of these counts once. Length counted with cl100k_base; the agent that loads this file may tokenize it differently.