gemini-exploration-patterns

# Gemini Exploration Patterns

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

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

## Overview

This skill provides strategic guidance for leveraging Gemini CLI's large context window for codebase exploration. It covers when to delegate exploration to Gemini, which model to use, and how to structure outputs for Claude to consume.

## When to Use This Skill

**Keywords:** explore codebase, analyze architecture, large context, token limit, gemini exploration, codebase analysis, when to use gemini, model selection

**Use this skill when:**

- Deciding whether to explore with Claude or Gemini
- Planning a large codebase analysis
- Choosing between Flash and Pro models
- Structuring exploration output for cross-CLI consumption
- Optimizing exploration for cost vs quality

## Token Threshold Decision Matrix

| Codebase Size | Tokens | Recommended Agent | Rationale |
| --- | --- | --- | --- |
| Small | <50K | Claude native | Claude's tools are faster |
| Medium | 50K-500K | Gemini Flash | Good balance of speed/cost |
| Large | 500K-1M | Gemini Flash + chunking | Stay within Flash limits |
| Very Large | 1M-2M | Gemini Pro | Need extended context |
| Massive | >2M | Gemini Pro + progressive | Multi-pass exploration |

### Token Estimation

```bash
# Quick estimation: 1 token ~ 4 characters
chars=$(find . -name "*.ts" -o -name "*.py" | xargs wc -c | tail -1 | awk '{print $1}')
tokens=$((chars / 4))
echo "Estimated tokens: $tokens"
```

### Decision Rule

```text
IF estimated_tokens < 50,000:
    USE Claude's native Explore agent

ELIF estimated_tokens < 1,000,000:
    USE Gemini Flash via /gemini-explore

ELIF estimated_tokens < 2,000,000:
    USE Gemini Pro via /gemini-explore --pro

ELSE:
    USE Progressive exploration (chunk by module)
`