| name | repo-harness-chatgpt-bridge |
| description | Use when setting up or operating the repo-harness ChatGPT Controller, GitHub Issue Launcher, local Agent Runs, or GitHub Copilot cloud sessions. |
repo-harness-chatgpt-bridge
You are operating inside a repo-harness adopted repository.
Primary model
Treat ChatGPT as the project controller. Repository-backed Issue/Task/Run state is authoritative; chat memory is not.
Use this chain for new work:
understand implementation
-> create/update Issue
-> inspect launch readiness
-> optionally publish to GitHub Issues/Project
-> launch narrow Tasks
-> review local diff or GitHub pull request
-> record Verification Gate evidence
-> accept or request changes
The legacy PRD -> checklist Sprint -> Codex Goal flow remains available for large product work and compatibility, but it is not the default execution path for every task.
First reads
docs/repo-harness-chatgpt-mcp-setup.md
docs/repo-harness-chatgpt-controller.md
docs/repo-harness-github-issue-launcher.md when GitHub is involved
repo-harness mcp doctor --repo .
repo-harness controller board --repo .
Do not rely on prior chat history when durable state exists.
Controller responsibilities
- Inspect current code, documents, Git state, Issues, Tasks, and Runs before planning.
- Keep each Task narrow, dependency-aware, path-scoped, and independently reviewable.
- Run
inspect_issue_readiness or prepare_issue_launch before launch.
- Use local
codex/claude Runs for local-only tooling and github-copilot for work that should be visible in GitHub Agents and a draft PR.
- Use
append_task, split_task, supersede_task, and set_task_dependencies when execution changes the plan.
- Review every completed Run. Agent success is not acceptance evidence.
- Record named check results and criterion-level evidence with
verify_task before accept_task.
- Never automatically merge, push, close a remote Issue, or bypass repository protections.
Setup commands
repo-harness mcp setup chatgpt --repo .
repo-harness mcp setup codex --repo . --scope project
repo-harness mcp keepalive --repo . --profile controller \
--enable-dev-runner --dev-runner-agents codex,claude --tunnel quick
GitHub integration check:
repo-harness controller github-status --repo .
Execution visibility:
repo-harness controller board --repo .
repo-harness controller runs --repo .
repo-harness controller watch <RUN-ID> --repo . --log
Safety boundaries
- Do not expose arbitrary shell input through MCP.
- Do not reveal or commit OAuth passphrases, bearer tokens, tunnel tokens, cookies, or credentials.
- Do not weaken immutable secret, credential, Git-internal, lockfile, or protected CI denies to make a Task pass.
- Local Agent workers must not commit, merge, or push automatically.
- GitHub Copilot cloud sessions may create draft pull requests only when explicitly launched; repo-harness never merges them automatically.
- Local isolated Runs must be reviewed and integrated before verification.
- A Task cannot be accepted until the Verification Gate passes.
Compatibility handoff
For an existing .ai/harness/handoff/codex-goal.md, preserve the referenced PRD and checklist Sprint, execute one task card at a time, run focused checks, and leave exact handoff evidence. Do not convert an active Controller Issue back into one large Goal unless the user explicitly requests the legacy workflow.
Troubleshooting
- Missing Controller tools: restart the MCP server and rescan or recreate the ChatGPT Connector.
- Only PRD/Sprint tools are visible: ChatGPT has an old Planner tool snapshot.
- Local Run looks hidden: use
repo-harness controller watch <RUN-ID> --log.
- GitHub Session does not start: run
repo-harness controller github-status --repo ., update/authenticate gh, and confirm Copilot coding-agent access.
- Task cannot launch: inspect the readiness blockers rather than bypassing them.
- Task cannot be accepted: integrate local work when required and record passing
verify_task evidence.