| Function | State | What the run proved | Still open |
|---|---|---|---|
| F1 ConnectLoom reads a design system from where your team keeps it, and links every component, token and rule to the file it came from. | Passes in partSuite run on public sources, 1 November 2026. |
|
|
| F2 Audit, gaps, plan and setupLoom scores the system, shows the gaps, drafts a change for each gap and asks the setup questions that configure the working tool. | PassesReal run, 1 November 2026. The draft pull requests are in our private test repository, so we describe them instead of linking them. |
| Nothing open. |
| F3 Build a componentA person asks for a component. An agent on your model drafts it on-system and opens a draft pull request where your checks run. | Built, real run pendingSuite run, 1 November 2026. The real run needs approved paid model calls. |
|
|
| F4 Review every pull requestLoom comments on each pull request. It names what breaks the system and what to use instead. | PassesReal run, 1 November 2026. The pull requests are in our private test repository, so we describe them instead of linking them. |
| Nothing open. |
| F5 GovernA web application where people file requests and proposals, the right owner decides, and an accepted change becomes a pull request. | Built, real run pendingSuite run, 1 November 2026. The real run needs a hosted instance and a second person to sign in. |
|
|
| F6 SyncAfter a merge, the Library, Storybook and the Figma library update from the repository, and drift between them is reported. | Passes in partSuite run, 1 November 2026, with real commits on a cloned repository. |
|
|
| F7 Agents on-systemA coding agent connected to Loom builds a screen with your components. The same task runs with and without Loom, and the difference is scored. | Built, real run pendingSuite run, 1 November 2026. The real run needs approved paid model calls. |
|
|
| + MonitoringOnce set up, Loom keeps watching health, drift, unmet needs, expiring exceptions and review times, and reports to the owners named in setup. | Built, real run pendingNo test of its own. | Nothing has passed for real yet. |
|
How a function earns its state
Passes means the acceptance test passed in a real run, on the date shown. Passes in part means some of the test passed for real, and the row says which part is still open. Built, real run pending means the code exists and is tested on its own, but the real run has not happened yet. We do not describe a function as working until it passes.
Monitoring runs across the other functions. It has no acceptance test of its own, so it never shows as passing on this page.
Some runs are described, not linked
The real runs for the audit and for the pull request review used our own test repository on GitHub. That repository is private, so a link would not open for you. We describe what each run proved instead. In a conversation, we can run the same tests on a repository you choose.
The suite runs without access too
Anyone with the code can run the whole suite. Without access tokens or approved model calls, every check that needs them reports as blocked and names what is missing. The last run of that kind, on 1 November 2026, finished with 30 checks passing, 18 blocked and 1 failing. The failing check is an offline Figma export that exists only as sample data, so it cannot pass.
The audit also ran on public systems
We ran the audit and plan on pinned public releases of two open-source design systems, GitLab UI and Carbon. The scores, the evidence, the gaps and the plan Loom would propose are on Open audits. Nothing was applied to those projects.
No client has used Loom yet
These results come from our own systems, public open-source systems and our test repository. No client has used Loom yet. When a design partner joins, their results are theirs to publish or keep.
Open audits of public systems Every capability with its state How we keep our claims honest
Coming next
- Build a component, for real. An agent on a real model drafts a component and opens a pull request where the system's checks run. This needs approved paid model calls.
- Agents with and without Loom. The same tasks run twice on a real model and scored. The result will appear here, whatever it shows.
- Govern on a hosted instance. A second person signs in, files a request, and the owner's decision becomes a pull request.
- Figma. Reading a real Figma file, updating a Figma library after a merge, and the full drift report.