The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from the browser DOM of the target application.
Ingestion points: Probes in probes/target-size.md, probes/failure-injection.md, and probes/focus-walk.md scrape text content including innerText, aria-label, and bodyText from the application being tested.
Boundary markers: Absent. Scraped text is interpolated into finding blocks without delimiters or instructions to the agent to treat the content as untrusted.
Capability inventory: Full browser automation via Playwright is used across all probe files, including navigation (page.goto), user interaction (page.click, page.keyboard), and request interception (page.route).
Sanitization: Absent. Scraped content is recorded and reported without validation, escaping, or filtering.
[DYNAMIC_EXECUTION]: Measurement probes in axe-scan.md, target-size.md, focus-walk.md, layout-shift.md, and web-vitals.md utilize page.evaluate() and page.addInitScript() to execute JavaScript logic within the browser context.
[COMMAND_EXECUTION]: The skill executes local shell commands to boot the target application (e.g., npm run dev or npm start) based on the repository's package.json as described in references/session-setup.md.

ui-verification