repo-harness-chatgpt-bridge

# repo-harness-chatgpt-bridge You are operating inside a repo-harness adopted repository. ## When To Use Use this Skill when the user asks to set up, operate, inspect, or continue the repo-harness ChatGPT MCP Connector, or when a ChatGPT-generated repo-harness PRD/Sprint/Goal handoff needs to be consumed by Codex. This Skill has four modes: 1. Setup mode: configure local MCP server files, ChatGPT guide, or Codex MCP config. 2. Planning handoff mode: preserve the chain idea -> PRD -> checklist Sprint -> Codex Goal. 3. Execution mode: local Codex reads `.ai/harness/handoff/codex-goal.md` and executes the referenced checklist Sprint. 4. Direct coding mode: an explicitly enabled `coding` profile retains the workflow/status tool surface and adds five guarded coding/process tools without invoking Codex. ## First Reads Before acting, read the repo-local source of truth that matches the mode: - Setup: `docs/repo-harness-chatgpt-mcp-setup.md`, `.repo-harness/mcp.local.json` if present, and `repo-harness mcp doctor --repo .`. - Planning handoff: `docs/spec.md`, `tasks/current.md`, existing `plans/prds/`, existing `plans/sprints/`, and latest `.ai/harness/handoff/`. - Execution: `.ai/harness/handoff/codex-goal.md`, the referenced PRD, the referenced Sprint, `tasks/current.md`, and `.ai/harness/handoff/resume.md` when present. Do not rely on chat history when these files exist. ## Agent Responsibilities 1. Treat ChatGPT as planner/reviewer and Codex as executor. 2. Prefer `rep...