app-docs
Pass
Audited by Gen Agent Trust Hub on Sep 15, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONDATA_EXFILTRATION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill navigates to and parses content from external web applications to generate descriptions and guides. This creates an attack surface where a malicious or compromised web app could include hidden instructions to manipulate the agent's documentation output or actions.
- Ingestion points: Page content, navigation menus, and metadata read from the target application URL in
SKILL.md. - Boundary markers: Absent. There are no instructions to the agent to disregard instructions embedded within the application being documented.
- Capability inventory: Browser automation (navigation, clicks), screenshot capture, and local file writing to the
docs/directory. - Sanitization: Absent. The agent is instructed to directly summarize the content it sees on the screen.
- [DATA_EXFILTRATION]: The skill is designed to take screenshots of live applications and save them to a local directory (
docs/screenshots/). The instructions specifically advise against using empty tables or "Test Client" data, preferring "realistic data." This behavior risks capturing personally identifiable information (PII), financial data, or session artifacts that appear in the UI, which are then permanently stored in the documentation files and potentially committed to a repository.
Audit Metadata