The Agent Skills Directory

EXTERNAL_DOWNLOADS (LOW): The skill instructs the agent to fetch live data from unpkg.com (specifically web-features and browser-compat-data JSON files) to ensure browser support information is current. While these are legitimate data sources, they represent external dependencies.
Evidence: references/live-mdn-fetch.md directs the agent to fetch from https://unpkg.com/web-features/data.json and https://unpkg.com/@mdn/browser-compat-data/data.json.
COMMAND_EXECUTION (LOW): The skill contains rules encouraging the agent to use browser automation MCP servers (Playwright, Puppeteer, etc.) to visually verify CSS implementations.
Evidence: references/rules/browser-verification.md specifies a workflow using Playwright/Puppeteer to navigate to pages and take screenshots for rendering verification.
PROMPT_INJECTION (LOW): The behavioral rules include strong directives to prioritize CSS over JavaScript, which acts as a domain-specific override of the agent's default logic.
Evidence: references/rules/css-only-enforcement.md states: 'ALWAYS prioritize CSS-only solutions. NEVER suggest JavaScript when CSS can solve the problem.'
INDIRECT PROMPT INJECTION (LOW): The skill possesses a vulnerability surface where external data is ingested and processed alongside high-capability browser tools.
Ingestion points: references/live-mdn-fetch.md (fetches JSON from unpkg.com).
Boundary markers: Absent for external JSON data.
Capability inventory: Browser automation (Playwright, Puppeteer) and network fetching.
Sanitization: Absent; the agent is expected to parse and act on the external JSON content directly.

css-first