LLM Supervisor

**Quick judgement:** A focused orchestration skill that adds real value in rate-limited environments by transparently handling cloud throttling and offering a well-governed fallback to a local Ollama model. It is narrow in scope but thoughtfully designed for developer workflows and hybrid cloud/local setups. **What it does well:** - Immediately surfaces cloud rate-limit/overload events instead of silently failing. - Offers a local fallback (Ollama with `qwen2.5:7b`) and tracks session state (current provider, last rate limit, confirmation for code). - Enforces an important safety/usability rule: **no automatic switch to a local model for code tasks without explicit confirmation**, while being more relaxed for simple chat/summarization once the user has opted in. - Provides clear, low-friction commands: `/llm status`, `/llm switch local`, `/llm switch cloud`. **Key risks / limitations:** - **Environment dependency:** Requires a correctly installed and configured Ollama environment; misconfigurations will degrade the fallback path. - **Model quality variance:** Fallback to `qwen2.5:7b` may produce lower-quality or slower code generation than the primary cloud model, which users must understand. - **Scope constraints:** Focuses specifically on Anthropic/OpenAI + Ollama; does not appear to generalize to other providers or advanced routing policies. - **State visibility reliance:** Users must rely on `/llm status` and prompts to stay aware of which provider is active; confusion is possible if they overlook these indicators. **Recommended scenarios:** - Heavy coding or tooling sessions where OpenAI/Anthropic rate limits are common, and a controlled, explicitly confirmed local fallback is desirable. - Hybrid LLM setups where users already run Ollama locally and want a simple way to reduce downtime and cloud token usage. - Development, teaching, or demo environments that want to showcase responsible model switching (especially around code generation) and transparent rate-limit handling.