The Agent Skills Directory

[COMMAND_EXECUTION]: The skill launches a local Chrome or Edge browser instance to render the target URL. This is performed using the Node.js spawn method with an array of arguments, including a validated URL and standard Chrome flags, effectively preventing shell injection.
[EXTERNAL_DOWNLOADS]: The skill fetches content from arbitrary URLs provided by the user via Chrome CDP. This is the primary function of the tool and is executed within the browser's sandbox.
[PROMPT_INJECTION]: The skill processes untrusted third-party content from the internet, which creates a potential surface for indirect prompt injection if the agent follows instructions found within the converted markdown.
Ingestion points: External web content is ingested via scripts/main.ts using Chrome CDP to navigate to user-supplied URLs.
Boundary markers: The output is structured with a YAML frontmatter block and markdown headings, providing clear separation between page metadata and the fetched content.
Capability inventory: The agent's primary capability in this context is reading the generated markdown file and using its content for information retrieval.
Sanitization: The skill implements robust sanitization by using linkedom and turndown to process the HTML, explicitly removing executable and interactive elements such as <script>, <iframe>, <style>, and <svg> tags before the content is presented to the agent.

url-to-markdown