The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests and processes untrusted HTML content from target websites or local directories to inventory links and metadata.\n
Ingestion points: The scripts/check-links.mjs script fetches and parses content from user-provided URLs or directories.\n
Capability inventory: The skill performs network requests (fetch), reads local files, and can drive a headless browser via Playwright.\n
Boundary markers: The agent's instructions do not specify the use of delimiters or safety wrappers when presenting crawled content (link labels, titles) to the model.\n
Sanitization: Content is parsed for structure, but there is no mechanism to filter or escape potential prompt injection payloads embedded in link text or page headings.\n- [DYNAMIC_EXECUTION]: The utility scripts use dynamic loading to resolve optional browser automation dependencies.\n
Evidence: scripts/lib/browser.mjs uses createRequire and dynamic import() to load playwright or playwright-core from the host project or global environment.\n- [COMMAND_EXECUTION]: The skill instructions require the agent to execute a bundled Node.js script to perform its primary function.\n
Evidence: The SKILL.md file explicitly instructs the agent to run node scripts/check-links.mjs with various arguments to audit the site.

legal-navigation