| name | rig-developer |
| description | Use when implementing or reviewing Rig itself, including changes to the CLI, runners, run artifacts, worktree isolation, MCP adapter, environment diagnostics, or agent harness documentation. |
Rig Developer
Use this skill for changes inside the Rig repository.
Boundaries
- Rig is a local AI coding harness, not an agent-to-agent protocol, workflow
engine, PTY automation framework, or package manager for agent assets.
- Keep the core model stable:
Task -> Run -> AgentAdapter -> Artifacts.
- Prefer declarative
.rig/config.yaml agent definitions over vendor-specific
Python branches.
- Keep skills, hooks, prompts, and MCP client config as external assets that Rig
can detect or document but should not silently install.
Implementation Rules
- Preserve run artifacts:
task.md, command.json, stdout.log,
stderr.log, result.md, and status.json; worktree runs also write
diff.patch.
- Keep CLI and MCP behavior aligned. MCP tools should call the same run store,
orchestrator, worktree, and artifact paths as the CLI.
- Keep safety defaults explicit.
RIG_MCP_ROOT constrains accepted cwd
values, and RIG_MCP_ALLOW_APPLY=1 gates MCP patch application.
- Use fake commands for tests. Real agent CLIs, credentials, or network access
belong only in explicit opt-in smoke checks.
- For new runners or prompt modes, add focused config parsing tests, CLI tests,
artifact assertions, and docs examples.
Verification
Run from the repository checkout:
uv run pytest -q
uv run ruff check .
uv run mypy rig tests
Use uv run rig ... when manually exercising the CLI from this checkout.