The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill uses puppeteer and cuimp which download external binaries (browser engines and curl-impersonate) to function.
Evidence: package.json dependencies include puppeteer and cuimp. scripts/fetch_special_sites_http.mjs configures cuimp with autoDownload: true.
[COMMAND_EXECUTION]: The skill executes browser processes locally to render and extract content from web pages.
Evidence: scripts/fetch_special_sites.mjs and scripts/fetch_generic_fallback.mjs use puppeteer.launch with the --no-sandbox flag to execute Chromium, Chrome, or Edge.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data (web page content) which could contain malicious instructions targeting the agent.
Ingestion points: External URLs provided to scripts/url_to_markdown.mjs are fetched via fetch, cuimp, or puppeteer in scripts/fetch_generic_fallback.mjs and scripts/fetch_special_sites.mjs.
Boundary markers: None. The cleaned markdown content is returned directly to the agent's context.
Capability inventory: The skill has network access and can execute browser processes.
Sanitization: Content is processed via Mozilla Readability and Turndown, which provides structural cleaning but does not filter for semantic prompt injection attacks.

web-to-markdown