The Agent Skills Directory

[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection through the processing of untrusted web content.
Ingestion points: Untrusted data enters the agent context via page.goto(), page.content(), and page.accessibility.snapshot() as described in SKILL.md.
Boundary markers: No delimiters or specific instructions are provided to the model to ignore or escape instructions that might be embedded within the retrieved web pages.
Capability inventory: The skill can launch browser processes, write to the file system (screenshots and PDFs), intercept network requests via page.route(), and execute arbitrary JavaScript in the browser context via page.evaluate().
Sanitization: There is no evidence of sanitization, filtering, or validation of the content retrieved from external URLs before processing.
[COMMAND_EXECUTION]: The skill facilitates the creation and execution of Node.js scripts that control browser instances and interact with the host system's file system.
[DATA_EXFILTRATION]: Provides methods to extract data from the browser and save it locally, including the page.screenshot() and page.pdf() functions.

playwright-automation