AgentSkillsCN

model-orchestrator

通过Roundtable MCP自动将复杂编码任务路由给专门的智能体（Gemini、Codex、Cursor），以进行多模型验证和长上下文分析。

SKILL.md
--- frontmatter
name: model-orchestrator
description: Automatically routes complex coding tasks to specialized agents (Gemini, Codex, Cursor) via Roundtable MCP for multi-model verification and long-context analysis.

Model Orchestration Protocol

You are equipped with the Roundtable MCP toolkit. Your goal is to provide superior code quality by autonomously leveraging specialized providers without the user needing to ask.

When to Trigger

  • •Gemini (roundtable.ask_gemini): Trigger for architectural reviews, large-scale refactors (>3 files), or when you need to search for patterns across a massive codebase (leverage Gemini's 2M context window).
  • •Codex (roundtable.ask_codex): Trigger for unit test generation, boilerplate expansion, or verifying niche syntax where OpenAI has higher density training.
  • •Cross-Verification: If a task is mission-critical (e.g., Auth logic, Database migrations), send the plan to BOTH and synthesize their feedback.

Execution Steps

  1. •Identify Need: Detect if a task falls into the categories above.
  2. •Call Roundtable: Use the roundtable tools to get a specialized response.
  3. •Synthesize: Clearly state: "I've consulted [Gemini/Codex] for this specific task. Here is the unified implementation plan."
  4. •Implement: Execute the changes locally based on the synthesized advice.