メニュー
Delegate planning to Codex/GPT. Returns dense, AI-optimized plan (not prose). Caller specifies reasoning effort (high/xhigh) based on complexity.
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
SOC 職業分類に基づく
| name | codex-planner |
| description | Delegate planning to Codex/GPT. Returns dense, AI-optimized plan (not prose). Caller specifies reasoning effort (high/xhigh) based on complexity. |
| context | fork |
| model | haiku |
| allowed-tools | ["Bash","Read","Glob","Grep"] |
Delegate architecture and design work to Codex/GPT-5. Return response verbatim.
The caller (main agent) should specify:
high or xhigh based on complexity
high: Most planning tasksxhigh: Complex architecture, critical systems, novel problemshigh)codex exec -m gpt-5.2-codex -c model_reasoning_effort="[high|xhigh]" --sandbox read-only -C "[PROJECT_DIR]" "[PROMPT]"
1. TASK: Design/plan [feature/system] for [goal].
2. EXPECTED OUTCOME: Implementation-ready plan in dense, AI-agent format.
3. CONTEXT:
- Architecture: [description or "explore codebase"]
- Relevant: [file paths/patterns]
- Goal: [what to build]
4. CONSTRAINTS:
- Must work with: [existing systems]
- Cannot change: [protected components]
- Patterns: [conventions to follow]
5. MUST DO:
- Explore codebase for existing patterns
- Ask clarifying questions if ambiguous
- Provide architecture decisions with rationale
- Break into actionable steps
- Identify edge cases
- Estimate effort (Quick/Short/Medium/Large)
6. MUST NOT DO:
- Write or edit files
- Over-engineer
- Assume unclear requirements
7. OUTPUT FORMAT:
Dense, AI-agent-optimized format. No prose paragraphs.
ARCHITECTURE:
- [bullet points, not sentences]
FILES:
- path/to/file.ts: [purpose, 5 words max]
STEPS:
1. [imperative, minimal]
2. [imperative, minimal]
EDGE CASES:
- [case]: [handling]
EFFORT: [Quick|Short|Medium|Large]
QUESTIONS: [if any, else omit section]
Return Codex response verbatim. Format is intentionally terse for token efficiency.
Review Qodo findings on the open GitHub pull request for the current branch using the persistent review as the canonical finding list and GitHub inline threads as discussion anchors, then fix, verify, refresh, or escalate with `/ask` using evidence and Qodo acknowledgment state.
Codex reviews a plan. Returns dense verdict + feedback (not prose). Caller specifies reasoning effort (medium/high/xhigh) based on review depth needed.
Semantic version control workflow: sem (structural truth) -> inspect (review prioritization) -> weave (coordination + merge semantics).
Use the Google Workspace CLI to work with Gmail, Drive, Calendar, Docs, and Sheets from the local host shell.
Guide for creating or updating agent skills. Use when users want to create a new skill, update an existing skill, or need guidance on SKILL.md structure and best practices across Claude Code, Codex, and other compatible agents.