How does the gate fail a build?

rubrkit test grades the instructions in your repo and exits non-zero when any artifact drops below the bar you set. A non-zero exit fails the CI step the same way a failing unit test does — no special integration required.

Does it run on every CI provider?

It is a Node CLI invoked with npx, so anywhere you can run a shell step — GitHub Actions, GitLab CI, CircleCI, Jenkins — you can run the gate.

Can I set a different bar per artifact?

Yes. The bar is configurable, so a stable skill can be held to a higher score than something still in draft.

Can it annotate the pull request?

The SARIF and JUnit output feed the surfaces that annotate PRs, while the exit code fails the check itself. You get both the red X and the inline detail.

How do I avoid flaky gate results?

A grade on a fixed version is reproducible, and the bar is a threshold you set, so the gate only flips when the artifact actually changes relative to the bar. If you want to gate on a change rather than a threshold, an advanced comparison measures the evaluator's own run-to-run spread first and only reports a difference that clears it — see statistical testing.

In the pipeline

Quality gates for the instructions your agents run.

rubrkit test in CI. Fail the build when a skill drops below your bar. JSON, JUnit, SARIF, stable exit codes.

Read the CLI docs

.github/workflows/ci.yml

- name: Rubrkit quality gate
  run: npx rubrkit test --min-score 80 --format junit

The gate

Fail the build, not the agent in production.

A non-zero exit fails the step like any other test. Drop a skill below the bar and the PR goes red — with JSON, JUnit, or SARIF output for the surface you already use.

rubrkit test — skill/code-review scored 71, below bar 80 · exit 1

Or call the SDK directly

import { gradeArtifact } from 'rubrkit';

const { score, passed } = await gradeArtifact(file, { minScore: 80 });
if (!passed) process.exit(1);

Wire it into any pipeline

A test step, not a new platform.

rubrkit test behaves like any other check: it reads the instructions in your repo, scores them, and exits non-zero when one drops below the bar. Nothing about your pipeline has to change to adopt it.

Stable exit codes

A non-zero exit fails the build the way a failing unit test does. Branch on the code in any CI.

Machine-readable formats

JSON for custom tooling, JUnit for test reporters, and SARIF for code-scanning surfaces that annotate PRs.

Per-artifact bars

Hold a stable, widely-used skill to a higher score than something still in draft.

Runs anywhere

A Node CLI over npx — GitHub Actions, GitLab CI, CircleCI, Jenkins, or a local pre-commit hook.

Developers

Bring Rubrkit into your toolchain.

Sync graded artifacts into the repos where your agents run, or connect any MCP client to the same audits, bundles, and compaction tools the web app uses.

CLI

rubrkit on npm

Pull approved artifact bundles into local projects and place them where Codex, Claude, or generic agents expect their instructions.

# Pull approved artifacts into your project
npx rubrkit pull

# Place them where your agent expects them
npx rubrkit pull all --agent claude --yes

View on npm API docs

MCP server

Public

Point any MCP client at Rubrkit and call the same artifact bundle, audit, and compaction tools, authenticated with your Rubrkit API key.

{
  "mcpServers": {
    "rubrkit": {
      "url": "https://www.rubrkit.com/api/v1/mcp",
      "headers": {
        "Authorization": "Bearer <your-rubrkit-api-key>"
      }
    }
  }
}

Create an API key API docs

FAQ

Answers before you start.

Know which instructions are ready to run.

Read the CLI docs

Newsletter

Follow the review loop as it ships.

Notes on AI artifact testing, prompt compaction, evals, and proof reports.