The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The skill instructs the agent to handle sensitive user authentication data, such as session cookies and JWT tokens, by passing them as plain-text command-line arguments to the inject-auth.js script. This pattern can result in the exposure of credentials in system logs, process lists, or shell history.
[COMMAND_EXECUTION]: The workflow involves executing multiple external scripts via node (e.g., inject-auth.js, navigate.js, screenshot.js) with arguments derived from user input or the target website's environment.
[EXTERNAL_DOWNLOADS]: The skill performs network operations to browse and audit arbitrary external websites based on user-supplied URLs, which are untrusted remote sources.
[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection. Ingestion points: Untrusted content from target URLs ($URL) is ingested during browsing. Boundary markers: Absent; the skill does not use delimiters or instructions to ignore embedded commands in web content. Capability inventory: Shell command execution via node and file system writes for screenshots and reports. Sanitization: No evidence of sanitizing or escaping website content before it is processed by subagents or AI multimodal analysis.

test-ui