The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides standard shell commands for installing development dependencies (npm i -D @axe-core/playwright), executing test suites (npx playwright test), and running Lighthouse CI (npx @lhci/cli autorun). These are typical for frontend testing environments.
[EXTERNAL_DOWNLOADS]: The instructions involve downloading well-known, trusted packages like Playwright and Axe-core from the official NPM registry. These are standard industry tools and do not represent a security risk.
[INDIRECT_PROMPT_INJECTION]: As a QA toolkit, the skill inherently processes external data (the DOM and content of web pages being audited). While this is a potential attack surface for indirect prompt injection, the skill's focus on technical diagnostics, metrics, and specific audits (like WCAG or Core Web Vitals) rather than following natural language instructions found on pages mitigates the risk.
Ingestion points: Playwright scripts and browser console helpers in SKILL.md and EVAL.md read and interact with the DOM of the target website.
Boundary markers: No explicit instruction delimiters are used, but the audit scripts are programmatic and diagnostic in nature.
Capability inventory: The skill utilizes Playwright to automate browser interactions and execute diagnostic JavaScript within the browser context. It also provides shell commands for CI integration.
Sanitization: The manual-check helpers in SKILL.md use CSS.escape for handling element IDs, showing awareness of basic data handling safety.
[DYNAMIC_EXECUTION]: The Playwright scripts use page.evaluate and page.addInitScript to execute JavaScript within the target browser's context. This is the standard method for web automation and testing to extract metrics and modify the test environment state (e.g., seeding random values for determinism).

frontend-qa-audit-toolkit