Help us improve
Share bugs, ideas, or general feedback.
From gemini-bridge
Background knowledge for using the Gemini Bridge MCP tools effectively. Teaches Claude when and how to delegate to Gemini in a model-agnostic multi-agent setup. Claude-invocable only.
npx claudepluginhub talent-factory/claude-plugins --plugin gemini-bridgeHow this skill is triggered — by the user, by Claude, or both
Slash command
/gemini-bridge:gemini-analystThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
Estimate token count before analysis:
Delegates tasks to Gemini CLI for large-context analysis like broad codebase reviews or long-document processing. Activates on explicit requests such as 'use gemini' or 'delegate to gemini'.
Guides large codebase exploration with Gemini CLI using token thresholds, Flash/Pro model selection, Claude vs Gemini decisions, and chunking strategies.
Consults Google's Gemini model for second opinions on coding tasks like debugging tricky issues, evaluating architecture trade-offs, reviewing code, and brainstorming alternatives.
Share bugs, ideas, or general feedback.
Estimate token count before analysis:
150K tokens → Delegate to Gemini
Quick estimation: ~750 tokens per ~600 words, ~500 tokens per ~400 lines of code.
For image analysis tasks where Gemini's vision capabilities or extended context
provide an advantage, prefer gemini_analyze_image. Claude Code can also read
images natively via the Read tool, so use Gemini when the analysis requires
Gemini-specific strengths (e.g., large diagrams, complex OCR, PDF parsing).
For architecture decisions, security reviews, or critical design choices:
Use gemini_compare_approaches to get an independent assessment.
This implements the "propose with Claude, validate with Gemini" pattern.
# Check bridge is working (do this once per session)
gemini_status()
# Short text prompts / second opinions
gemini_analyze_text(prompt, context=None, temperature=0.2)
# Large codebase analysis (up to 1M tokens with default model)
gemini_analyze_codebase(code_content, task, language=None)
# Image/screenshot/PDF analysis
gemini_analyze_image(image_path, question)
# Architecture/implementation comparison
gemini_compare_approaches(problem, approach_a, approach_b, criteria=None)
Always label Gemini's output clearly:
**Analysis by Gemini 2.5 Pro**
[Gemini's response]
This is important for:
The bridge is designed so that the routing logic lives in agent configuration, not in application code. To swap Gemini for a different model:
GEMINI_MODEL environment variableThis demonstrates to students: good architecture is model-agnostic.