What problem does it solve? Connecting a new OpenAI- or Anthropic-compatible model provider to the Catalyst Code harness requires knowing whether the provider's model-discovery endpoint matches a known response shape; this Skill makes that determination and performs either the config-only setup or the required code change in provider.rs. ## Core Features & Use Cases - Decision tree for config vs code: Classifies a provider's model-discovery JSON into four shapes (Umans /models/info, OpenAI /models, Anthropic /v1/models, or custom) to decide if a config-only setup is possible. - Config-only provider setup: Writes a providers entry into ~/.config/catalyst-code/config.json with kind, base_url, api_key_env, headers, and per-model models_override capability entries. - Code-change path for custom discovery: Implements the OpenCode-Go pattern in core/src/provider.rs with a host matcher, capabilities table, known-models table, wire-protocol router, and unit tests. - Use Case: A user says "add provider X" — the Skill reads the provider's API docs, samples its model-listing response, and either writes the config entry or adds a discovery branch with tests. ## Quick Start Ask the agent to add a new model provider by giving it the provider's base URL, API key environment variable, and a sample of its model-listing JSON response.
