The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: Fetches the turndown and readability libraries from the well-known JSDelivr CDN at runtime in scripts/markdown_converter.js and scripts/readability_loader.js.
[REMOTE_CODE_EXECUTION]: Dynamically loads and executes JavaScript from the JSDelivr CDN into the browser context using script tag injection.
[REMOTE_CODE_EXECUTION]: Employs the eval() function in scripts/readability_extractor.js and scripts/readability_extractor_local.js to execute the Readability library code, which is stored as an inlined string literal.
[COMMAND_EXECUTION]: Documentation in SKILL.md suggests shell commands that configure the chrome-devtools MCP server with flags that weaken browser security, such as --disable-web-security and --disable-features=IsolateOrigins,site-per-process.
[DATA_EXFILTRATION]: scripts/save_with_images.js utilizes Node.js http and https modules to download images from arbitrary external URLs to the local filesystem, which is part of its core functionality.
[PROMPT_INJECTION]: The skill is susceptible to Indirect Prompt Injection because it processes untrusted data from external websites.
Ingestion points: Extracts text, titles, and metadata from arbitrary URLs provided at runtime.
Boundary markers: No protective delimiters or instructions to ignore embedded commands are added to the extracted content.
Capability inventory: The skill can execute arbitrary JavaScript in the browser, write files locally, and initiate network requests.
Sanitization: While it removes HTML formatting tags (like <script>), it does not sanitize the resulting text for malicious natural language instructions aimed at the agent.

web-article-extractor