The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to fetch content from external websites and provide it to the AI agent, creating a surface for indirect prompt injection.
Ingestion points: The scripts scripts/playwright-simple.js (line 42) and scripts/playwright-stealth.js (line 145) extract document.body.innerText from arbitrary URLs provided at runtime.
Boundary markers: The extracted content is returned within a JSON object, but the text itself lacks internal delimiters or instructions to help the agent distinguish between data and potential malicious commands embedded in the scraped page.
Capability inventory: The skill possesses network access via Playwright and the ability to write files (screenshots and HTML) to the local filesystem using fs.writeFileSync in scripts/playwright-stealth.js (line 173).
Sanitization: The captured content is truncated (to 1000 or 5000 characters) but is otherwise returned as a raw string without any escaping or filtering of potential injection payloads.
[COMMAND_EXECUTION]: The skill utilizes shell commands and Node.js processes to launch the Playwright browser and execute scraping logic. This includes the use of browser arguments like --no-sandbox and --disable-setuid-sandbox which, while common for headless environments, reduce the security isolation of the browser process. This behavior is documented and aligns with the skill's primary purpose.

playwright-scraper-skill