What problem does it solve? When running agent skills across different LLM providers (OpenAI, Anthropic, DeepSeek, Gemini, local runtimes, and more), each provider has a different API base, auth header, tool-calling format, and capability set. This Skill standardizes that pre-flight configuration so every other skill knows which model is active and what it can do. ## Core Features & Use Cases - Provider Detection & Pinning: Auto-detects the active harness (Pi, Codex, Cursor, Claude Code) and optional 9router gateway, or pins a specific provider via --provider, --probe-env, or --harness flags. - Canonical Provider Table: Maps 13+ providers (GPT, Claude, DeepSeek, GLM, MiniMax, Mistral, Gemini, Qwen, xAI, OpenRouter, Ollama, vLLM, llama.cpp) to their API bases, auth headers, and tool-call formats. - Capability Flags: Writes .skills-config.yaml declaring context window, reasoning, vision, audio, prompt caching, and output token caps so downstream skills adapt their behavior. - Use Case: After switching your harness from Claude to a local Ollama model, run this Skill to regenerate the config, verify tool calling works, and warn skills that vision is no longer available. ## Quick Start Ask the agent to run the model-provider-config skill to detect the current harness and write a fresh .skills-config.yaml for the project.
