gemini-delegation-patterns

# Gemini Delegation Patterns

## 🚨 MANDATORY: Invoke gemini-cli-docs First

> **STOP - Before providing ANY response about Gemini CLI capabilities:**
>
> 1. **INVOKE** `gemini-cli-docs` skill
> 2. **QUERY** for the specific feature topic
> 3. **BASE** all responses EXCLUSIVELY on official documentation loaded

## Overview

Meta-skill for determining when and how Claude should delegate to Gemini CLI. Documents decision criteria, execution patterns, and result handling.

## When to Use This Skill

**Keywords:** delegate to gemini, use gemini, gemini vs claude, which agent, hand off, gemini better, claude or gemini, second brain

**Use this skill when:**

- Deciding whether a task should go to Gemini
- Planning cross-agent workflows
- Understanding Gemini CLI integration points
- Optimizing task distribution between Claude and Gemini

## Decision Matrix

| Scenario | Delegate to Gemini? | Rationale |
| --- | --- | --- |
| Interactive TUI needed (vim, git rebase -i) | YES | Claude cannot handle PTY |
| Risky shell command | YES | Gemini sandbox isolation |
| Large codebase analysis (100K+ tokens) | YES | Gemini 1M+ context window |
| GCP/Firebase/Vertex auth required | YES | Native Google integration |
| Need instant rollback capability | YES | Gemini checkpointing |
| Quick code edit | NO | Claude is faster |
| Multi-turn conversation | NO | Claude maintains context better |
| Complex reasoning with files | NO | Claude's Edit tool is superior |
| Need persistent session | NO | Claude Code has better UX |
| Security-sensitive analysis | MAYBE | Gemini sandbox + Claude reasoning |

## Execution Patterns

### Pattern 1: Fire-and-Forget (Headless)

Best for: Quick queries, analysis, code generation

```bash
gemini "{prompt}" --output-format json
```

**When to use:**

- One-off questions
- Code analysis without modification
- Documentation generation

### Pattern 2: Sandboxed Execution

Best for: Risky commands, untrusted code

```bash
gemini -s "Execute: {command}" --out