The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill performs automated downloads of software components during its initialization phase.\n
Fetches the playwright package from the NPM registry.\n
Downloads the Chromium browser binary from official Playwright sources, with a fallback to the well-known npmmirror.com service.\n- [COMMAND_EXECUTION]: The script uses execSync to manage its execution environment and dependencies.\n
Runs npm install and npx playwright install commands to set up the browser environment in the user's home directory (~/.cache/axure-extractor).\n
Launches the Chromium browser with specific flags including --allow-file-access-from-files, --disable-web-security, and --no-sandbox to facilitate broad data extraction.\n- [INDIRECT_PROMPT_INJECTION]: The skill possesses an attack surface for indirect prompt injection by processing arbitrary third-party web content.\n
Ingestion points: Content is ingested from user-provided URLs through browser-injected scripts such as extract-markdown.js (text and headings) and extract-links.js (interactive elements).\n
Boundary markers: The skill does not implement delimiters or safety instructions to distinguish between extracted web content and the agent's instructions.\n
Capability inventory: The skill can write files to the local disk (e.g., theme.json, content.md) and perform network operations via the automated browser.\n
Sanitization: Extracted content is converted to Markdown but does not undergo filtering to remove potential malicious instructions intended for the LLM.\n- [DYNAMIC_EXECUTION]: The skill uses dynamic JavaScript evaluation to execute its extraction logic.\n
In the library modules, the script reads local JavaScript files and uses the new Function() constructor to wrap and return functions for execution within the Playwright page.evaluate() context.\n
The lib/browser.mjs module performs a dynamic import() of the playwright package from a computed path in the local filesystem cache.

extract-page-data