The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to navigate to and scan external web applications, which are untrusted data sources that could contain malicious instructions designed to influence the agent.
Ingestion points: During 'Phase 4: Explore', the agent navigates to target URLs and performs visual scans, interacts with form elements, and monitors console/network logs, all of which represent untrusted input channels.
Boundary markers: The instructions do not define explicit boundary markers or provide the agent with specific guidance to ignore or isolate commands found within the web pages it is testing.
Capability inventory: The agent has access to Bash, Write, Read, and WebSearch tools. This combination of capabilities allows for potential local system actions if an indirect injection successfully overrides the agent's primary instructions.
Sanitization: There are no instructions for sanitizing or filtering the content harvested from external web pages before the agent processes it for reporting.

qa-only