GitLab Transcend hackathon · Path B · Supervised

SafeRules

A governed rule engine, with AI agents in GitLab.

The safety rules of a cosmetics ingredient checker (medicines, allergies, pregnancy) are fixed and written by a person. Every change to them travels through a lifecycle: request, AI triage, safety review, tests and scans, and a human decision. Agents do the steps. People approve.

Open the governance dashboardCheck a product (demo)

Demo only. No account, no photo, no personal data leaves your browser.

How a rule change travels

  1. Step 1Person

    Request

    A person opens a rule-change request as a GitLab issue, with a source.

  2. Step 2AI agent

    Triage

    A GitLab Duo flow reads it, sets the rule area and risk class, and asks what is missing.

  3. Step 3AI agent

    Safety review

    Another flow reviews the merge request for forbidden wording and weakened warnings. It never approves.

  4. Step 4Pipeline

    Pipeline

    404 clinical tests, security scans, a container and an SBOM run on every change.

  5. Step 5Person

    Human decision

    A person merges or closes. Production deployment stays manual.

Rule governance

See the lifecycle with real data

The rules catalogue, change requests with the agent's labels, safety review verdicts, pipeline health and an audit timeline, read live from the public GitLab project.

Product check (demo)

What the rules do for a person

Pick a fictional profile and a sample ingredient list. The fixed rules answer Good match, Check first or Better to skip, with the exact reasons.

Demo only. Not medical advice. SafeRules does not diagnose or treat anything. When in doubt, ask your pharmacist or doctor. See how the rules work. Built on Counter Check (MIT) by someone who worked in community pharmacies for twenty years.