The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill fetches and processes content from external URLs, creating a surface for potential indirect injection attacks from untrusted web data.\n
Ingestion points: The analyze-webpage.js script uses Playwright to navigate to and load content from user-provided external URLs.\n
Boundary markers: The SKILL.md file contains a dedicated 'External Content Safety' section that explicitly instructs the agent to treat all fetched content as untrusted and to ignore any embedded commands or directives.\n
Capability inventory: The skill has the capability to perform outbound network requests and write multiple files (cleaned HTML, metadata JSON, and images) to the local workspace via scripts/analyze-webpage.js and scripts/image-capture.js.\n
Sanitization: The extractCleanedHTML function in analyze-webpage.js removes executable elements such as <script>, <style>, and <noscript> tags and strips all but a strict whitelist of safe attributes. Additionally, scripts/generate-path.js includes functions to sanitize filenames and paths to prevent injection or traversal.\n- [EXTERNAL_DOWNLOADS]: The skill is designed to fetch and download images from remote servers during the scraping process.\n
Evidence: scripts/image-capture.js implements logic to intercept image responses and save the data to a local images/ directory, using sharp for format conversion where necessary.\n- [COMMAND_EXECUTION]: The skill instructions prompt the execution of Node.js scripts to perform the analysis tasks.\n
Evidence: SKILL.md provides specific bash commands for the user to run analyze-webpage.js with target URLs and output directories.

scrape-webpage