add-provider
Skill: Add Provider
Objective
Scaffold a new LLM provider in scripts/reviewer.py following the pattern set by the shipping references — AnthropicProvider for the chat-completions family or ClaudeCodeProvider / CursorProvider / CodexProvider for the agent-runner family. Adds the class skeleton, the build_provider() registry entry, the DEFAULT_MODELS mapping, any provider-specific action.yml inputs (including CLI install steps for agent-runners), and the corresponding doc updates.
The skill produces the scaffold and the doc trail; the actual message/response translation (chat-completions) or the CLI subprocess wiring specifics (agent-runner) are the user's job. See .agents/agents/provider-implementer.md for the deep dive on both families.
Family selection — ASK FIRST
Before scaffolding, ask the user which family the new provider belongs to:
- Chat-completions (
Provider) — the vendor exposes a raw HTTP messages/completions API and you own the tool-use loop (Anthropic, OpenAI, Gemini, Bedrock, vLLM/Ollama). - Agent-runner (
AgentRunnerProvider) — the vendor ships a headless coding-agent CLI you shell out to; the CLI writes.aiprr/findings.json(Claude Code, Cursor Agent, OpenAI Codex, Aider, Continue, ...).
The rest of this skill has two implementation branches; run the one matching the family.