The Agent Skills Directory

[DYNAMIC_EXECUTION]: The script scripts/lib/browser.mjs utilizes dynamic loading techniques to resolve either playwright or playwright-core at runtime. It employs createRequire().resolve() and import() with file URLs computed based on the current working directory and common global installation paths.
[EXTERNAL_DOWNLOADS]: The scripts/inspect-metadata.mjs tool performs network operations to fetch HTML, robots.txt, and sitemap files from arbitrary external URLs provided by the user via the --url argument.
[INDIRECT_PROMPT_INJECTION]: The skill ingests and processes untrusted HTML content from external websites or local directories, presenting a potential vector for malicious instructions to reach the AI agent. Evidence: 1. Ingestion points: loadSite in scripts/inspect-metadata.mjs. 2. Boundary markers: Absent. 3. Capability inventory: Network access via fetch and file system read access via fs.readFileSync. 4. Sanitization: Absent.
[OBFUSCATION]: The HTML parsing library in scripts/lib/html.mjs contains a decodeEntities function capable of decoding hexadecimal and decimal HTML entities. This functionality, while necessary for parsing, can be leveraged to obscure malicious content from simple text-based security scanners.

social-sharing