talent-factory/analyze

Gemini Bridge: Analyze

Delegate analysis tasks to Gemini 2.5 Pro (up to 1M token context) when the scope exceeds what Claude should handle directly.

When to Use Gemini vs. Claude

| Situation | Route to | |---|---| | Codebase > 150K tokens | Gemini | | Screenshot, diagram, or PDF input | Gemini (gemini_analyze_image) | | Independent second opinion on a design decision | Gemini | | Normal-sized files, tool use, code writing | Claude directly |

Quick token estimate: ~500 tokens per 400 lines of code; ~750 tokens per 600 words.

Workflow

1. Verify the bridge

gemini_status()

If the bridge is unavailable, inform the user and fall back to Claude's native analysis (with a note about the context limitation).

2. Collect content

Use Read, Glob, and Grep to gather the relevant files or text into a single context string. Include only what is necessary for the analysis task.

3. Invoke the appropriate tool

Large codebase analysis:

gemini_analyze_codebase(code_content=<collected_code>, task=<analysis_task>)

Text / document analysis:

gemini_analyze_text(prompt=<task_description>, context=<collected_text>)

Image / screenshot / PDF:

gemini_analyze_image(image_path=<path>, question=<task_description>)

4. Present the result

Always attribute Gemini's output clearly:

**Analysis by Gemini 2.5 Pro**
[Gemini's response]

Then synthesise findings and suggest next steps using Claude's judgment.

Model Configuration

The model is set via the GEMINI_MODEL environment variable. Changing this variable does not require any workflow changes — all tools continue to work unchanged. This is the model-agnostic design principle: routing logic lives in configuration, not in code.