The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The install.sh script downloads and executes the nvm installer from its official GitHub repository (nvm-sh/nvm) using a piped bash command. This is a common practice for developer tools but involves running remote code during setup.
[COMMAND_EXECUTION]: The installation script modifies file permissions using chmod +x for all Python and JavaScript files in the scripts directory to ensure they are executable.
[COMMAND_EXECUTION]: The documentation for handling authenticated sites recommends running Playwright with --no-sandbox and dangerouslyDisableSandbox: true. Disabling the browser sandbox is a security concern as it removes a primary layer of isolation when the agent navigates to arbitrary or potentially malicious external URLs.
[CREDENTIALS_UNSAFE]: The skill instructions facilitate a workflow where the agent captures, saves, and reuses session cookies (site_cookies.json) in the /tmp directory. While used for legitimate scraping of login-required sites, this involves the agent managing sensitive plain-text session data.
[PROMPT_INJECTION]: The skill has a high surface for indirect prompt injection as its primary function is to ingest untrusted content from the web into the agent's context.
Ingestion points: Untrusted web content enters the context through scrape_page.py, crawl_site.py, and scrape_docs.py using both requests and playwright.
Boundary markers: The skill does not implement specific delimiters or 'ignore' instructions when presenting the scraped Markdown to the agent, increasing the risk of the agent obeying instructions found on a webpage.
Capability inventory: The skill possesses the ability to execute subprocesses (calling Node.js), perform network operations, and write files to the local system.
Sanitization: HTML content is cleaned using BeautifulSoup to remove scripts, styles, and other non-content tags before being converted to Markdown.
[EXTERNAL_DOWNLOADS]: The skill downloads the Chromium browser binary via npx playwright install chromium as a functional requirement for rendering JavaScript-heavy websites.

web-scraper