Skills
The four skills that teach a coding agent agent-qa's workflows, and how to install and keep them current.
Skills are short instruction packs that teach a coding agent how to work with agent-qa: which tools to call, in what order, and what never to do. agent-qa ships four of them inside the agent-qa package, matched to the installed version.
The four skills
| Skill | Use it to | What it teaches |
|---|---|---|
agent-qa-authoring | Write, edit and run tests. | Writing tests with agent_qa_write_test, editing with agent_qa_refine_test, adding hooks and targets with agent_qa_change_config, running with agent_qa_start_run, and the rules for IDs, secrets, remote browsers, plugins and devices. |
agent-qa-result-triage | Sort out why a run failed. | Reading the digest, overview and evidence in order, then naming a cause with its confidence, the evidence and the next action. |
agent-qa-debug-fix | Fix the cause of a failure. | Treating the digest as a hypothesis, making the smallest fix, and rerunning the narrowest test. |
agent-qa-knowledge | Read runs and what agent-qa learned. | Which command or tool answers which question, and never editing agent-qa-knowledge/ by hand. |
The triage skill sorts failures into timeout, appium_startup, browser_disconnect, element_not_found, assertion_failure, hook_failure, infrastructure and unknown_failure.
Install them
npx agent-qa init installs the four skills for the coding agents it sets up, into .claude/skills/ for Claude Code and .agents/skills/ for Codex, together with the entry for the MCP server the skills call. After you upgrade agent-qa, run npx agent-qa init again and it refreshes them, so there is nothing to delete or copy by hand.
npx agent-qa skills uninstall removes them, with the MCP entries and the AGENTS.md block init wrote. Add --dry-run to see what it would remove, and --agents to pick the hosts.
Whether you commit the installed folders is up to you. Coding agents lists everything init writes for each host.
Use them
Name a skill in your request, or let the agent pick one when your request fits:
/agent-qa-authoring Write a test for the checkout flow and run it.
$agent-qa-result-triage Why did the latest run fail?The first form is Claude Code's, the second Codex's.
Find them
npx agent-qa skillsThe command prints a Path: line with the folder the skills ship in, Source: package, and one line per skill with its description. --json prints the same as path, source and skills. It installs nothing.
Capabilities for agents
npx agent-qa init --capabilities prints one JSON document that describes everything agent-qa can do. Its rows hold every function with its inputs and an example call, the chat shortcuts and the skills, and its posture says whether this project's config is absent, reads or fails to read. It writes nothing. An agent can read it before choosing what to call.