The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to execute a Node.js one-liner to parse and filter Lighthouse audit reports locally. Evidence: The node -e script in SKILL.md used to extract failing audits from report.json without loading the full file into context.
[DYNAMIC_EXECUTION]: The skill utilizes the evaluate_script tool to execute custom JavaScript logic within the browser context for specialized accessibility checks. Evidence: JavaScript snippets in references/a11y-snippets.md used for measuring tap targets, checking color contrast, and identifying orphaned form inputs.
[EXTERNAL_DOWNLOADS]: The skill references and fetches accessibility documentation from a well-known service (web.dev) to research specific guidelines. Evidence: Instructions in the 'Core Concepts' section of SKILL.md for retrieving raw markdown versions of web.dev articles.
[INDIRECT_PROMPT_INJECTION]: The skill processes data from untrusted sources (web page content and browser console messages) while possessing capabilities to interact with and execute scripts on the page. Ingestion points: list_console_messages and take_snapshot in SKILL.md. Boundary markers: The skill does not provide instructions to treat console output or accessibility tree snapshots as untrusted data. Capability inventory: Access to evaluate_script, press_key, take_screenshot, and lighthouse_audit. Sanitization: No sanitization or filtering of external browser data is performed before processing.

a11y-debugging