The Agent Skills Directory

[PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface by processing content from external web environments.
Ingestion points: The skill retrieves untrusted data from the web through page snapshots and the inspection of browser console logs as part of its verification workflow.
Boundary markers: There are no explicit delimiters or instructions defined in the skill to isolate content retrieved from the web from the agent's internal instructions.
Capability inventory: The skill possesses the capability to navigate to arbitrary URLs and perform interactions (clicks, fills, presses) via the act and open tools.
Sanitization: No sanitization, escaping, or validation mechanisms are described for the content retrieved before the agent uses it to determine pass/fail status or exercise UI paths.
[COMMAND_EXECUTION]: The skill utilizes browser automation capabilities to perform sequences of actions on target websites. Evidence: The workflow involves using open, act, and snapshot tools to navigate and interact with web applications.
[DATA_EXFILTRATION]: The skill performs network operations to non-whitelisted domains and local environments, capturing and returning UI and console data. Evidence: Workflow steps in SKILL.md involve navigating to localhost or dev UI targets and capturing evidence via screenshots and log inspection.

browse-qa