Skip to main content
Groundrule logo

Groundrule

Live this week

Your engineering standards, taught to every coding agent and checked on every PR

Groundrule - Your engineering standards, taught to every coding agent and checked on every PR
Groundrule - Your engineering standards, taught to every coding agent and checked on every PR

About Groundrule

AI agents now write a large share of your code. Your engineering standards didn't come along: they live in wikis, review comments and the heads of a few senior people, and every agent file (AGENTS.md, CLAUDE.md, Cursor rules, Copilot instructions) drifts into its own copy. Groundrule turns those standards into one rulebook, teaches it to every coding agent, and checks every change against it. **One rulebook.** Write a standard once: the requirement, why it exists, how to fix a violation, a severity and a stage. Start from curated packs (security baseline, TypeScript/Node, Python, Go, Java/Spring, React, architecture, Docker, Kubernetes, Terraform, GitHub Actions, testing, agent hygiene) and add your own. **Every agent reads it.** One `groundrule sync` writes AGENTS.md, CLAUDE.md, Cursor rules and Copilot instructions from the same source, and a stop hook plus an MCP server let agents check their own work before they finish. **Start from what your code already does.** `groundrule scan` reads the rules a repository already keeps (layer boundaries, placement conventions, package gates, imported from your existing docs too) and sends proposals to a review inbox, each with its evidence. Accept, and it's a standard. **Checked on every pull request.** Deterministic checks run in CI and on commits. Each finding names what failed, which standard, the evidence (file and line) and the fix. Blockers stop the merge; everything else teaches. **Roll out on evidence.** Every rule moves through observe → teach → advise → enforce, with the data to justify each step. A compliance report shows pass rate, rules in effect, exceptions used and what commits declare about coding agents, exportable as CSV. The core (CLI, checks, packs, hooks, MCP server) is open source under Apache-2.0 and works with no account. The hosted workspace adds the rulebook, inbox, rollout stages, team and repository settings, and compliance reports. Free for small teams; open-source projects stay free.

Gallery

Groundrule screenshot 1 of 6
Groundrule screenshot 1 of 6
Groundrule screenshot 2 of 6
Groundrule screenshot 2 of 6
Groundrule screenshot 3 of 6
Groundrule screenshot 3 of 6
Groundrule screenshot 4 of 6
Groundrule screenshot 4 of 6
Groundrule screenshot 5 of 6
Groundrule screenshot 5 of 6
Groundrule screenshot 6 of 6
Groundrule screenshot 6 of 6

Demo video

Groundrule demo video

Overview

Who is it for?

Engineering teams that ship with AI coding agents: platform and staff engineers, tech leads and engineering managers who own code quality, architecture and security standards.

Problem

Engineering standards are scattered across wikis, lint configs, review comments and senior engineers' heads. Coding agents can't read any of that, so every agent file (AGENTS.md, CLAUDE.md, Cursor rules, Copilot instructions) becomes a hand-maintained copy that drifts, and the only enforcement is a human reviewer catching the same mistakes again. Teams either slow down agents with review, or let architecture and security rules erode one PR at a time.

Solution

Groundrule turns standards into one rulebook and makes it operational: every coding agent reads it (one sync writes all agent files; a stop hook and MCP server let agents check their own work), every pull request is checked deterministically in CI with findings that name the standard, the evidence and the fix, and every rule rolls out on evidence through observe, teach, advise and enforce. Teams start from curated packs or from what their code already does (a scan proposes the rules it finds, with evidence), and auditors get a compliance report built from the actual runs.

What makes it unique

- Agent-first: one source writes AGENTS.md, CLAUDE.md, Cursor and Copilot files, and agents can check themselves before they finish. - Rules with proof: every finding carries the standard, the file and line, and the fix; a compliance report is built from runs, not claims. - Evidence-based rollout: observe → teach → advise → enforce per rule, with the numbers to justify each step. - Discovery, not a blank page: `groundrule scan` reads layer, layout and package rules a repository already keeps, plus existing docs, and proposes them with evidence. - Deterministic by default: imports, layout, metrics, files, regex, dependencies and Semgrep checks run without an LLM; AI is labelled where it's used. - Open-source core (Apache-2.0) that works with no account; the hosted workspace is optional.

FAQ

What is Groundrule?
Groundrule turns engineering standards into one rulebook, writes it into every coding agent's instruction files (AGENTS.md, CLAUDE.md, Cursor rules, Copilot instructions), and checks every pull request against it with findings that name the standard, the evidence and the fix.
Which coding agents does Groundrule support?
Any agent that reads AGENTS.md, plus Claude Code (CLAUDE.md), Cursor (.cursor/rules) and GitHub Copilot (copilot-instructions.md). One `groundrule sync` writes all of them from the same source, and an MCP server lets agents look up rules and check their work.
Do I have to write all the rules myself?
No. Start from curated packs (security baseline, TypeScript/Node, Python, Go, Java/Spring, React, architecture, Docker, Kubernetes, Terraform, GitHub Actions, testing, agent hygiene), or run `groundrule scan`, which reads the layer, layout and package rules a repository already keeps, and your existing docs, and proposes them with evidence.
How are rules checked?
Deterministically by default: imports, layout, metrics, file, regex, dependency and Semgrep checks run in CI, on commits and in the agent loop without an LLM. Each finding states what failed, which standard, the file and line, and how to fix it. AI-answered checks are labelled as such.
Will Groundrule block my pull requests on day one?
Only if you want it to. Every rule has a stage: observe (collect evidence only), teach (in agent files), advise (reported, never blocks) and enforce (fails the check at its severity). You move a rule forward when the evidence shows it is clean.
Is Groundrule open source?
The core is: the CLI, the checks, the packs, the hooks, the GitHub Action and the MCP server are Apache-2.0 and work with no account. The hosted workspace (rulebook, review inbox, rollout stages, compliance reports, team settings) is a paid product with a free tier.
How much does Groundrule cost?
The Free plan covers 5 active contributors and 3 repositories and never expires. Team is $15 per active contributor per month billed annually ($18 monthly); Business adds auditor-grade evidence and hosted checks. Open-source projects stay on Team free of charge.
What is an active contributor?
Anyone who committed to a connected repository in the last 90 days, counted by the CLI from git history as a salted hash. Reviewers, managers and people who only read the rulebook are free.
Does Groundrule send my code anywhere?
Checks run in your CI or on your machine and upload results, not source. Findings carry one-line snippets. Hosted checks run in an isolated sandbox with no network except the configured LLM provider, and secrets are redacted before any LLM call.
Can Groundrule prove compliance to an auditor?
Yes. The compliance report shows, for a period, which rules were in effect, the pass rate per rule, which exceptions were used and by whom, and what commits declare about coding agents, built from the actual check runs and exportable as CSV.