gemini-cli - SKILL.md Agent Skill

name: gemini-cli description: Wield Google's Gemini CLI as a powerful auxiliary tool for code generation, review, analysis, and web research. Use when tasks benefit from a second AI perspective, current web information via Google Search, codebase architecture analysis, or parallel code generation. Also use when user explicitly requests Gemini operations. allowed-tools: - Bash - Read - Write - Grep - Glob

Gemini CLI Skill Guide

When to Use Gemini

Use Case	Why Gemini
Current web information	`google_web_search` - real-time Google Search
Codebase architecture analysis	`codebase_investigator` - deep analysis tool
Second opinion / code review	Different AI perspective catches different bugs
Parallel code generation	Offload tasks while continuing other work

When NOT to use: Simple quick tasks (overhead not worth it), interactive refinement, context already understood.

Running a Task

Verify installation: command -v gemini
Select the mode required for the task; default to read-only (no --yolo) unless edits are necessary.
Always use AskUserQuestion before using --yolo or -s flags. These modes allow file writes or sandboxed execution - get explicit user approval first.
Assemble the command with appropriate options:
- -m, --model <MODEL> - Model selection
- -y, --yolo - Auto-approve all tool calls (enables writes)
- -s, --sandbox - Run in Docker isolation
- -o, --output-format <text|json> - Output format
Important: extract the JSON response with:
```
gemini "prompt" -o json 2>/dev/null | perl -0777 -pe 's/^[^{]*//s' | jq -r '.response'
```
The perl stage is required: gemini CLI (>= 0.37) prints status noise (e.g. MCP issues detected. Run /mcp list for status.) on stdout before the JSON object, so piping straight to jq fails with Invalid numeric literal. The perl filter strips everything before the first {.

Critical Note

YOLO mode does NOT prevent planning prompts. Use forceful language: "Apply now", "Start immediately", "Do this without asking for confirmation".

Quick Reference

EXTRACT below stands for perl -0777 -pe 's/^[^{]*//s' | jq -r '.response' (always write it out in full when running commands).

Use case	Mode	Command pattern
Read-only analysis	read-only	`gemini "..." -o json 2>/dev/null \| EXTRACT`
Apply local edits	write	`gemini "..." --yolo -o json 2>/dev/null \| EXTRACT`
Sandboxed write	sandbox	`gemini "..." --yolo --sandbox -o json 2>/dev/null \| EXTRACT`

Example Commands

# Read-only
gemini "Review src/ for bugs" -o json 2>/dev/null | perl -0777 -pe 's/^[^{]*//s' | jq -r '.response'

# Write mode
gemini "Fix bug in file.py. Apply now." --yolo -o json 2>/dev/null | perl -0777 -pe 's/^[^{]*//s' | jq -r '.response'

# If redirection fails, wrap in bash -lc
bash -lc 'gemini "prompt" -o json 2>/dev/null | perl -0777 -pe "s/^[^{]*//s" | jq -r ".response"'

Following Up

Resume: echo "follow-up" | gemini -r latest -o json 2>/dev/null | perl -0777 -pe 's/^[^{]*//s' | jq -r '.response'
List sessions: gemini --list-sessions

Error Handling

jq parse error (Invalid numeric literal): stdout noise before the JSON — make sure the perl -0777 -pe 's/^[^{]*//s' stage is in the pipe.
Rate limit: CLI auto-retries with backoff. Use -m gemini-2.5-flash for lower priority tasks.
Command failure: Check with gemini --version, use --debug for details.
Always validate Gemini's output for security vulnerabilities (XSS, injection) before using.