The Agent Skills Directory

[COMMAND_EXECUTION]: The skill involves executing Node.js scripts (analyze-webpage.js, generate-path.js) and installing dependencies via npm. These scripts perform file system operations, including creating directories (fs.mkdirSync, mkdir) and writing files (fs.writeFileSync, writeFile) for metadata, screenshots, and downloaded images.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from external websites specified by the user or found during the scraping process. This content is then processed and returned to the agent context.
Ingestion points: The analyze-webpage.js script fetches external content using page.goto(url) and intercepts network requests for images.
Boundary markers: While the output is structured into JSON (metadata.json) and cleaned HTML, there are no specific markers instructing the agent to ignore potential instructions embedded in the scraped text.
Capability inventory: The skill possesses file-writing capabilities (saving HTML, JSON, and images) and executes JavaScript within a headless browser context via Playwright's page.evaluate.
Sanitization: The extractCleanedHTML function in analyze-webpage.js provides significant mitigation by removing <script>, <style>, and <noscript> tags, and stripping non-essential attributes from the HTML before it is saved or processed further.

scrape-webpage