What problem does it solve? NanoClaw ships with only the Claude Agent SDK as its agent backend. This Skill wires in OpenAI's Codex (CLI + AppServer) as a full agent provider — planning, tool orchestration, native compaction, MCP tools, and session resume — so agents can run on a ChatGPT subscription or OPENAI_API_KEY instead of Anthropic. ## Core Features & Use Cases - Provider Installation: Copies Codex provider source files from the providers branch into both the host (src/providers/) and container (container/agent-runner/src/providers/) trees, registers them via barrel imports, and adds the Codex CLI install layer to the container Dockerfile. - Registration Guards: Ships structural tests that verify the barrel imports and Dockerfile edits stay intact, going red if the wiring drifts. - Flexible Auth: Supports ChatGPT subscription OAuth (codex login), API key via OPENAI_API_KEY, or BYO OpenAI-compatible endpoints via OPENAI_BASE_URL, selectable per group with agent_provider. - Use Case: A team wants some NanoClaw groups to run on GPT models through their ChatGPT subscription while others stay on Claude — set "provider": "codex" in the group's container.json after running this Skill. ## Quick Start Ask the agent to install the Codex provider into NanoClaw and rebuild the container image, then set a test group's provider to codex.