barrierlab-agent-pluginv0.1.0

First audit

Run /barrierlab:page on a file or URL and read the four sections of the report.

Run it

Name a local HTML file or a URL:

/barrierlab:page dist/contact/index.html

A local file that links stylesheets needs to know where they live, otherwise contrast stays unchecked and the report says so first:

/barrierlab:page dist/contact/index.html --base-url https://example.com/contact/

What you get

The run creates .barrierlab/runs/<UTC timestamp>/ in the project with two files: report.json, the engine’s output as it was printed, and report.md, the report the agent wrote on top of it. Add .barrierlab/ to .gitignore unless you want to keep the runs.

The report has four sections, always in this order:

  1. Failed: what the engine found to be a violation, with the reason it matters and a proposed change to the source as a diff.
  2. Needs review: what a person has to decide. The agent adds an assessment, labelled as one, with the lines it read.
  3. Not checked: what the engine could not decide statically, the step that would decide it, and the rules that did not run with the engine’s reason.
  4. Plan: the proposed order of work, by severity and by how many findings a change removes.

Two frames built from real engine output are in the showcase.

Confirm a fix

Apply the change, run /barrierlab:page again and compare the engine summary line at the top of both reports. The plugin does not compare runs for you yet.

Edit this page on GitHub · Docs for v0.1.0