The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill automates web browser interactions to capture screenshots and sample CSS styles from external websites. This functionality introduces an attack surface for indirect prompt injection if a visited website contains malicious content designed to exploit the agent.
Ingestion points: The shoot.mjs and styles.mjs scripts visit external URLs provided via command-line arguments to the agent.
Boundary markers: The skill relies on Playwright's browser isolation; however, there are no specific prompt delimiters or instructions to the agent to ignore content found within the processed sites.
Capability inventory: The skill can write files to the local system (including reports and images) and execute JavaScript in the context of the visited web pages via page.evaluate (defined in scripts/lib/browser.mjs).
Sanitization: The tool focuses on structural and visual comparison (pixel diffs and CSS properties) rather than natural language interpretation, which significantly reduces the risk of traditional indirect prompt injection.
[COMMAND_EXECUTION]: The skill provides a set of Node.js command-line utilities for regression testing. These scripts perform network requests and file system operations based on user-supplied configurations, representing a standard functional surface for such tooling.

visual-review