The Agent Skills Directory

[COMMAND_EXECUTION]: The skill leverages MCP browser tools (e.g., mcp__playwright__browser_navigate, browser_take_screenshot) to interact with and inspect websites as part of its primary design review function.
[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection because it processes content from untrusted external websites.
Ingestion points: Untrusted website content enters the agent's context through browser navigation and inspection.
Boundary markers: There are no explicit delimiters or instructions provided to the agent to treat the website content as data rather than instructions.
Capability inventory: The skill uses browser automation tools including navigation, resizing, and screenshotting.
Sanitization: There is no evidence of content sanitization or filtering of the website text before the agent evaluates it.
[DATA_EXFILTRATION]: The instructions allow for opening local files (e.g., index.html) when provided via $ARGUMENTS. While this is intended for local web development review, it enables the agent to read content from the local filesystem.

design-jury