Overview
What the plugin does with an accessibility audit, where it stops, and how this documentation is organised.
barrierlab-agent-plugin adds two commands to Claude Code. /barrierlab:page audits one page with
the barrierlab rule set and writes a report the agent then
works with: it explains each finding, proposes fixes for the source of the project and keeps
everything that was not checked visible.
The engine decides
Outcomes, rule ids and counts come from
barrierlab-cli. The agent never adds, drops or
re-grades a finding. A script builds the frame of the report from the engine’s JSON, and a second
pass of the same script verifies that the finished report still names every finding.
Where it stops
- One page per run, analysed statically. What needs a browser is reported as not checked.
- No score, and no statement about conformance or law.
- Fixes are proposed as diffs, not applied.
The full list is on Limits.
How the docs are organised
- Getting started: install the engine and the plugin, then run a first audit.
- Reference: the commands, the report and its guarantees, the limits, and how to work on the plugin.