The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill provides patterns for browser automation that ingest and process content from external URLs, which could contain malicious instructions.
Ingestion points: The page.goto(url) and page.evaluate() methods in SKILL.md allow the agent to load and extract data from external websites.
Boundary markers: The provided code snippets do not include delimiters or specific instructions to the agent to ignore embedded commands within the scraped content.
Capability inventory: The skill utilizes network access for navigation (page.goto), script execution within the browser context (page.evaluate), and file system writes for document generation (page.pdf).
Sanitization: There is no evidence of sanitization or filtering of the external page content before it is processed by the agent.
[COMMAND_EXECUTION]: The skill provides examples of launching the browser with security-sensitive flags.
Evidence: The configuration in SKILL.md uses the --no-sandbox and --disable-setuid-sandbox arguments.
Context: These flags are standard requirements for running Puppeteer in restricted environments like Docker containers or certain CI/CD pipelines, but they reduce the security isolation provided by the Chrome sandbox.

puppeteer