@coro-ai/llm-openai

OpenAI Responses API phase executor for Coro.

This package implements the PhaseExecutorRuntime contract from @coro-ai/plugin-sdk and is designed to be loaded by the runner as plugin id openai.

Capabilities

• Uses the OpenAI Responses API.
• Replays stateless session state through Job.conversationHistory.
• Maps the runner's in-process Coro MCP server into OpenAI function tools named mcp__<server>__<tool>.
• Enforces per-phase allowed-tools and write-root hook policy before every tool call.
• Reports normalized token usage and derives cost from the package model catalogue.

Configuration

{
  "plugins": {
    "installed": {
      "openai": {
        "enabled": true,
        "config": {
          "apiKey": "sk-...",
          "defaultModel": "gpt-5.4"
        }
      }
    }
  },
  "llm": {
    "defaultProvider": "openai",
    "aliases": {
      "planning": { "provider": "openai", "model": "gpt-5.5", "reasoningEffort": "high" },
      "coding": { "provider": "openai", "model": "gpt-5.3-codex", "reasoningEffort": "medium" }
    }
  }
}

apiKey may be omitted when OPENAI_API_KEY is set in the runner environment.

Optional config fields:

• baseURL — OpenAI-compatible endpoint override.
• organization — OpenAI organization id.
• project — OpenAI project id.
• defaultModel — fallback model when a phase resolves no explicit model.

Notes

External plugin MCP servers using stdio, http, or sse transports are not directly dialed by this package yet. The runner's in-process coro MCP server is fully supported and includes SCM/tracker proxies plus file and skill tools when the active executor lacks native equivalents.