메뉴
Run tasks through the Gemini CLI and then review, verify, and evaluate the resulting changes. Use when the user asks to delegate work to Gemini (or says "use gemini") and expects a follow-up review of the edits, diffs, or outputs.
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
SOC 직업 분류 기준
| name | gemini |
| description | Run tasks through the Gemini CLI and then review, verify, and evaluate the resulting changes. Use when the user asks to delegate work to Gemini (or says "use gemini") and expects a follow-up review of the edits, diffs, or outputs. |
Delegate a task to Gemini CLI in yolo (auto-approve) mode, then validate and evaluate the resulting changes before reporting back.
Path A — running inside Claude Code (you ARE Claude Code): Do NOT spawn a nested claude process. Use the Bash tool to run the script directly:
skills/gemini/scripts/run.sh "Your task prompt here"
Optional model override:
skills/gemini/scripts/run.sh "Your task prompt here" --model gemini-2.5-pro
Or set GEMINI_MODEL in the environment to avoid repeating the flag.
Path B — running as an external agent (bash, CI, orchestrator):
scripts/run.sh "Your task prompt here"
The script validates that gemini is on PATH and runs gemini --yolo --prompt "<prompt>".
After Gemini finishes, inspect what changed:
git status -sb
git diff --stat
git diff
| Model | Best for |
|---|---|
gemini-2.5-pro (default) | Complex multi-file refactors, reasoning-heavy tasks |
gemini-2.0-flash | Fast, focused edits; simple one-file tasks |
Override via --model flag or GEMINI_MODEL env var.
--yolo skips all confirmation prompts. Gemini will write files immediately.git stash.scripts/run.sh: validate gemini is installed and invoke gemini --yolo --prompt "<prompt>".Run Claude CLI in non-interactive mode to review git diffs between branches or commits and return severity-ordered findings with file references. Use when a user asks for a "Claude review", wants a second-opinion code review from Claude, or needs fast triage of a large branch diff like master...feature.
Create strict JSON execution plans with Claude Code and return machine-readable handoff objects for planner-executor workflows. Use when a user asks for a 2-step planner to executor pipeline, wants Claude to produce a structured plan schema, or needs validated plan JSON before step-by-step execution.
Delegate a bounded local coding task to the Copilot CLI, either as a single non-interactive run or as multiple tmux workers with per-task logs. Use when the user explicitly wants local `copilot --yolo` execution with strict file ownership, prompt files, and post-run review.
Detect and remove dead/unused code using language-appropriate static analysis tools (ruff+vulture, knip, cargo clippy, golangci-lint, etc.). Use when cleaning up after vibe-coding, AI-generated code, or any session that may have left unused imports, dead functions, or unreachable exports.
Request or re-request GitHub Copilot review on the open PR for the current branch, or on a specified PR number, using gh CLI; verify gh auth first, trigger the reviewer request, and confirm GitHub recorded a fresh review request event.
Update a Docker-based Home Assistant deployment to the latest stable image and verify the service is healthy afterward. Use when a user asks to upgrade Home Assistant, check whether a Docker-based HA update is safe, or recover Home Assistant after an image refresh.