Groundrule documentation
packs, your own standards, and what your repos and docs already say
an inbox of proposals, routed to the owners of each category
AGENTS.md, CLAUDE.md, Cursor and Copilot rules, MCP
deterministic checks on every change, promoted with evidence
Who are you?
Every page says who it was written for. Start where you are; the words are the same everywhere, only the depth changes.
Six pages that answer most questions
If you read nothing else, read these. Together they take under an hour and cover what a team needs on its first day.
- 01Get startedTen steps from an empty workspace to rules in every repository and checks in CI.
- 02How Groundrule worksStandards, packs, scopes, stages and proposals: the eleven words the product is built on.
- 03Rollout stagesObserve, Teach, Advise, Enforce: what each stage does in agent files and in checks.
- 04CLI quickstartSign in, connect a repository, sync agent files and run your first check.
- 05The rule formatEvery field of a standard, with examples, scope, checks and agent wording.
- 06AI and your dataWhat is sent to the model, what is removed first, what is kept, and what it costs.
Every engineering team has rules: how money is stored, which client calls the payment provider, what never goes in a log. They live in people's heads, a wiki nobody reads and a thousand review comments. Coding agents now write much of the code, and they only follow what they are told. Groundrule gives the team one rulebook: start from maintained packs, import what your repositories, documents and pull-request reviews already say, and review every proposal in an inbox. Each rule then reaches every coding agent as instructions in AGENTS.md, CLAUDE.md, Cursor and Copilot, and every change is checked against it, locally and in CI. Rules roll out in stages, Observe, Teach, Advise, Enforce, and move forward only when the evidence from your own code says they are ready. AI helps find and write rules, but a person accepts every one.