ziwang-Physics/AgentChat

---

name: AgentChat-IndependentTasks

description: 并行分发 N 道独立任务给多个浏览器 AI，回收后由 Claude Code 合成权威答案并输出教科书式解答手册 PDF。触发词："分给多个 AI""并行问多个 AI""汇总成 PDF/解答手册"。不适用于需 AI 间协作或角色分工的任务。

---

name: AgentChat-OneWeb

description: Multi-provider CDP bridge with automatic fallback (Gemini->ChatGPT->Claude->Qwen->Kimi->MiniMax->Doubao->MiMo->DeepSeek). Use for AI provider failover, fallback chain, multi-provider routing, or "send to any available AI". MANDATORY EXECUTION - invoking this skill REQUIRES running `node ~/.claude/skills/AgentChat-OneWeb/index.js "<prompt>"` as the FIRST action and quoting its `[receipt] AGENTCHAT_RUN` line in the final answer; explaining the skill or answering from model knowledge without a receipt is a violation.

---

name: AgentChat-WebSubAgent

description: Sequential 6-step AI pipeline — Claude Code plans→Kimi searches→(complex? Gemini reasons)→Claude synthesizes→ChatGPT reviews→Claude fixes. Use for complex software engineering tasks needing web research + deep reasoning + quality review. MANDATORY EXECUTION - invoking this skill REQUIRES actually running `node skills/AgentChat-WebSubAgent/index.js` for Step 2 (--search) and Step 5 (--review) unconditionally (Step 3 --reason if complex), and listing each step's receipt.run_id from the returned JSON in the final report; narrating the pipeline or answering from model knowledge without receipts is a violation.

---

name: mcp-server

description: AgentChat MCP Server — expose 8 AI providers as MCP tools for any MCP-compatible client (Claude Desktop, Cursor, Continue.dev). Configuration/reference skill, not executable