ワンクリックで
gemini-delegate
Delegate a coding task to Google Gemini agent for autonomous execution
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。
メニュー
Delegate a coding task to Google Gemini agent for autonomous execution
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 | gemini-delegate |
| description | Delegate a coding task to Google Gemini agent for autonomous execution |
Created: 2026-02-09-00-00 Last Updated: 2026-02-16-00-00
You are delegating a coding task to Gemini CLI. Gemini will analyze and execute the task.
Review conversation context. Restate the task clearly: what to do, which files, what "done" looks like. If unclear, ask the user.
Read CODEX.md and relevant code. Build ONE prompt with this structure:
# Project Context
[Relevant sections from CODEX.md]
# Task
[Clear description from conversation context]
# Relevant Files
[Key files the model should know about]
# Requirements
- [specific requirements]
# Success Criteria
- [how to verify it's done correctly]
Please analyze this task and provide:
1. Your implementation approach
2. The code changes needed
3. Any risks or considerations
Rules: Include CODEX.md context directly. Be specific about requirements. Do NOT tell the model how to implement or over-constrain the solution.
gemini <<'EOF'
YOUR_PROMPT_HERE
EOF
Alternative if heredoc fails:
gemini "YOUR_PROMPT_HERE"
Unlike Codex, Gemini does NOT auto-apply changes. YOU must:
## Delegation Report
### Task: [what was delegated]
### Context Provided: [summary of CODEX.md context included]
### What Gemini Suggested: [analysis, code changes, risks]
### Changes Applied: [files created/modified]
### Verification: [did it meet success criteria? issues?]
### Follow-up: [remaining tasks, things to review]
If Gemini fails with auth error:
Please set GEMINI_API_KEY in your environment:
export GEMINI_API_KEY="your-key-here"
# Add to ~/.zshrc for persistence
If Gemini times out or returns partial output, report what was received and suggest retry.
Process autonomous task queue from do-work/ folder
Enter plan mode for complex tasks (pour energy into the plan for 1-shot implementation)
Initiate autonomous PR review process with Codex agent
Run comprehensive repo assessment with Codex (every 3 PRs)
Comprehensive security audit covering 10 threat categories
Install all automation for a new repo (git hooks + GitHub Actions)