codex-collaboration
Codex CLI integration for code review and consultation. Use when: (1) code needs review before commit, (2) need expert consultation on implementation approach, (3) debugging assistance needed. This skill provides Codex interaction patterns only - workflow orchestration is defined in project CLAUDE.md.
When & Why to Use This Skill
This Claude skill integrates the Codex CLI to streamline the software development lifecycle through automated code reviews, expert implementation consultation, and intelligent debugging. By establishing a 'Review Gate' pattern, it ensures code quality, security, and correctness before commits, while providing interactive guidance for complex architectural decisions and full-auto task execution.
Use Cases
- Pre-commit Quality Assurance: Automatically analyzing uncommitted changes to detect bugs, security vulnerabilities, and edge cases before they reach the repository.
- Architectural Decision Support: Comparing multiple implementation approaches against specific constraints to determine the most efficient and scalable solution.
- Automated Debugging and Fixing: Leveraging the 'Full Auto' execution mode to identify root causes of issues and implement fixes directly within the project workspace.
- Technical Consultation: Providing developers with expert-level guidance and interactive sessions when stuck on complex implementation logic or unfamiliar frameworks.
| name | codex-collaboration |
|---|---|
| description | "Codex CLI integration for code review and consultation. Use when: (1) code needs review before commit, (2) need expert consultation on implementation approach, (3) debugging assistance needed. This skill provides Codex interaction patterns only - workflow orchestration is defined in project CLAUDE.md." |
Codex Collaboration Skill
Provides Codex CLI interaction patterns for code review and consultation.
Codex CLI Commands
Code Review (Primary Use)
# Standard review
codex review --uncommitted
# Review specific changes
codex exec -m gpt-5.2 "
Review this implementation:
$(git diff)
Check:
1. Correctness
2. Bugs
3. Security
4. Code quality
5. Edge cases
Verdict: PASS or FAIL with issues
"
Consultation
# Ask for guidance
codex exec -m gpt-5.2 "
Context: [CONTEXT]
Question: [SPECIFIC_QUESTION]
"
# Interactive session
codex "Help me with [TOPIC]"
Full Auto Execution
# Let Codex make changes
codex --full-auto "Implement [TASK]"
# With workspace access
codex -C /path/to/project --full-auto "Task"
Review Gate Pattern
Code Change
│
▼
codex review
│
├── PASS → git commit
│
└── FAIL → Fix issues → Re-review
Review Result Handling
PASS:
git add .
git commit -m "type(scope): description"
FAIL:
- Parse issues from output
- Fix each issue
- Re-submit:
codex review - Repeat until PASS
Model Options
gpt-5.2- Latest flagship (recommended)gpt-5.1-codex-mini- Faster, cheaper
Usage Examples
Before Commit
# After implementing a feature
codex review
# If PASS, commit
# If FAIL, fix and re-review
When Stuck
codex exec -m gpt-5.2 "
I'm implementing [TASK].
Current code: [CODE]
Issue: [PROBLEM]
What's the best approach?
"
Architecture Decision
codex exec -m gpt-5.2 "
Option A: [APPROACH_A]
Option B: [APPROACH_B]
Constraints: [CONSTRAINTS]
Which is better and why?
"