Install the agent skill

Install

aix finds the coding agents installed on your machine and asks which one gets the skill. Installing at user scope makes it available to that agent in every project.

npx -y @a1st/aix install https://github.com/hypothesi/a11ied/blob/master/packages/skills/ai.json --only skills --user

Add --target claude-code, --target codex, or another supported agent name to install without being asked.

First run

Install the a1 command if it is not already on the machine.
Restart the coding agent so it loads the new skill.
Start the app you want to test.
Give it the route and the behavior you want checked.

npm install --global a11ied

a1 doctor

Use the a11ied skill to test http://localhost:3000/checkout.
Run the automated checks first, then test the field labels and the
error messages with the screen reader. Keep the two results separate.

Rules in the skill

Looks up the criterion first When the task includes a criterion from the Web Content Accessibility Guidelines (WCAG), the agent runs wcag_show and reads the criterion's test method before it runs anything else.

Keeps the requested scope The development skill uses relevant criteria to focus a requested component check. The complete audit skill uses the coordinator's required procedures and coverage issues; relevance hints do not remove WCAG obligations.

Picks the test method Mapped scanner checks provide partial coverage. Hybrid criteria also need interaction or semantic assessment; manual criteria need human or agent judgment.

Moves from simulated to real The simulated screen reader for fast local runs, VoiceOver or NVDA when the question is how the real software behaves.

Reports results separately Automated results, transcripts, and a person's judgment stay apart in the report. A passing axe run is not reported as meeting WCAG.

Skill and MCP server

The skill tells the agent how to plan a test. The MCP server gives it the tools to run one. Either works without the other. With both, the agent has the tools and the instructions for using them.

The package installs a11ied for component development and full-site-audit for complete desktop assessments. The audit skill uses runtime state, evidence validation, and recovery rather than hand-editing inventories.

Follow use with an AI agent to connect the MCP server. Use the MCP reference when you need exact tool inputs.

Update

The same aix install command pulls the current skill from the repository and reinstalls it at user scope.