The Agent Skills Directory

[PROMPT_INJECTION]: The skill fetches and processes content from arbitrary external URLs. This creates a surface for indirect prompt injection, where malicious instructions embedded in a web page could influence the agent's behavior.\n
Ingestion points: The tool accepts any URL provided as a command-line argument in scripts/main.ts.\n
Boundary markers: Output is delimited by a YAML frontmatter block and Markdown headers in scripts/html-to-markdown.ts, providing some separation for the agent.\n
Capability inventory: The skill can execute browser processes (scripts/cdp.ts), write files (scripts/main.ts), and create directories.\n
Sanitization: The conversion process in scripts/html-to-markdown.ts uses the turndown library and a custom sanitizeHtml function to strip scripts, styles, and other potentially dangerous tags, but the resulting Markdown still reflects the source page's content.\n- [DATA_EXFILTRATION]: There is no protocol validation for the input URL in scripts/main.ts. This allows the tool to potentially access and read local files on the host system if a file:// URI is provided, which is then converted to Markdown and returned to the agent.\n- [COMMAND_EXECUTION]: The script uses child_process.spawn in scripts/cdp.ts to launch system-installed web browsers such as Google Chrome, Microsoft Edge, or Chromium for rendering web pages.\n- [EXTERNAL_DOWNLOADS]: The skill depends on external Node.js packages including @mozilla/readability, linkedom, and turndown. These are well-known libraries from trusted sources.

baoyu-url-to-markdown