codex-consult
This skill should be used when the user asks to "consult Codex", "ask Codex for advice", "get implementation suggestions from Codex", "codexに相談", "codexにきいて", "実装方針を相談", "アーキテクチャをcodexに聞く", or needs deeper analysis using OpenAI's Codex model for complex implementation decisions, architecture design, or code review that benefits from extended reasoning.
SKILL.md
| Name | codex-consult |
| Description | This skill should be used when the user asks to "consult Codex", "ask Codex for advice", "get implementation suggestions from Codex", "codexに相談", "codexにきいて", "実装方針を相談", "アーキテクチャをcodexに聞く", or needs deeper analysis using OpenAI's Codex model for complex implementation decisions, architecture design, or code review that benefits from extended reasoning. |
name: codex-consult description: This skill should be used when the user asks to "consult Codex", "ask Codex for advice", "get implementation suggestions from Codex", "codexに相談", "codexにきいて", "実装方針を相談", "アーキテクチャをcodexに聞く", or needs deeper analysis using OpenAI's Codex model for complex implementation decisions, architecture design, or code review that benefits from extended reasoning. version: 1.0.0
Codex Consult
Overview
This skill enables consultation with OpenAI's Codex CLI for deeper implementation analysis. Codex uses more powerful reasoning models (gpt-5.2-codex) that can provide thorough architectural guidance and implementation plans.
When to use:
- Complex architectural decisions requiring deep analysis
- Implementation planning for multi-file changes
- Code review needing extended reasoning
- Design pattern recommendations
- Refactoring strategy development
Execution Modes
Plan Mode (Default)
Provides analysis and implementation planning without code changes.
Output includes:
- Summary of the approach
- Implementation strategy
- Verification steps
Patch Mode
Provides analysis with concrete code change suggestions in unified diff format.
Output includes:
- Summary of proposed changes
- Unified diff patches
- Verification steps
Arguments
| Argument | Description | Default |
|---|---|---|
| Task description | Required. What to analyze or plan | - |
--model=<model> | Model to use | gpt-5.2-codex |
--mode=plan|patch | Execution mode | plan |
--scope=<path> | Target directory | current repo root |
--no-web | Disable web search | web enabled by default |
Usage Instructions
Basic Consultation (Plan Mode)
To consult Codex for implementation planning:
-
Execute the wrapper script with task description:
bash ${CLAUDE_PLUGIN_ROOT}/scripts/codex-exec.sh "Implement user authentication with JWT" -
Read the output from
/tmp/codex-consult.last.md -
If output is empty or error occurs, check
/tmp/codex-consult.run.log
Patch Mode Consultation
To get concrete code change suggestions:
bash ${CLAUDE_PLUGIN_ROOT}/scripts/codex-exec.sh --mode=patch "Refactor the database connection to use connection pooling"
Custom Model
To use a different model:
bash ${CLAUDE_PLUGIN_ROOT}/scripts/codex-exec.sh --model=o4-mini "Design the caching layer"
Scoped Analysis
To analyze a specific directory:
bash ${CLAUDE_PLUGIN_ROOT}/scripts/codex-exec.sh --scope=/path/to/src "Review error handling patterns"
Disable Web Search
To run without web search:
bash ${CLAUDE_PLUGIN_ROOT}/scripts/codex-exec.sh --no-web "Analyze this legacy code"
Output Interpretation
Summary Section
- What changes are proposed
- Assumptions and preconditions
- Impact analysis
Changes Section (Patch Mode)
- Unified diff format
- File-by-file breakdown
- Explanation of each change
Verification Section
- Test commands to run
- Manual verification steps
- Edge cases to check
Best Practices
- Be specific: Provide clear task descriptions with context
- Always specify
--scopeexplicitly: Use--scope=$(pwd)or--scope=/absolute/path/to/projectto ensure Codex analyzes the correct project context. This is especially important in multi-repository environments. - Start with plan mode: Use plan mode first, then patch mode if needed
- Verify suggestions: Always review Codex output before applying changes
- Consider execution time: Complex analyses may take longer. Avoid running as background tasks with strict timeouts.
Caveats
Scope Resolution in Multi-Repository Environments
When working in directories that contain multiple git repositories (e.g., nested subprojects or monorepos), Codex may analyze the wrong project context if --scope is not explicitly specified.
Problem scenario:
/home/user/projects/
├── main-project/
│ └── submodules/
│ └── other-repo/ ← Codex may analyze this instead
└── .git/
Solution: Always use explicit --scope parameter:
# From any directory, specify the target project explicitly
bash ${CLAUDE_PLUGIN_ROOT}/scripts/codex-exec.sh --scope=/home/user/projects/main-project "Analyze the codebase"
# Or use $(pwd) to ensure current directory is used
bash ${CLAUDE_PLUGIN_ROOT}/scripts/codex-exec.sh --scope=$(pwd) "Analyze the codebase"
Timeout Considerations
Codex CLI may encounter timeout failures (exit code 144) in the following scenarios:
- Background task execution: Running as a background process with strict timeout limits
- Complex analysis requests: Large codebases or architecturally complex queries requiring extended reasoning
- Network latency: Slow connections can cause API timeouts
Recommendations:
- Set timeout to at least 10 minutes (600 seconds): Complex analyses require extended processing time. Shorter timeouts will likely cause failures.
- Run Codex consultations in the foreground when possible
- Break down complex queries into smaller, focused questions
- For large codebases, use
--scopeto limit the analysis scope - Check
/tmp/codex-consult.run.logfor timeout-related errors
Error Handling
If Codex execution fails:
- Check
/tmp/codex-consult.run.logfor error details - Verify Codex CLI is authenticated (
codex login) - Check network connectivity
- Try with a different model if quota exceeded
Additional Resources
Reference Files
references/codex-cli-reference.md- Codex CLI commands and optionsreferences/output-format.md- Output format specification
Examples
examples/sample-consultation.md- Sample consultation outputs