dig-deeper-core
Pass
Audited by Gen Agent Trust Hub on Sep 14, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted user input describing a technical problem. This input is then processed by multiple agent roles to generate a diagnosis, which creates a surface for indirect prompt injection attacks.
- Ingestion points: SKILL.md (Step 1) receives the 'human problem' data, which is then passed to the Troubleshooter and Interrogator roles.
- Boundary markers: None are explicitly defined to separate user input from agent instructions within the shared artifacts.
- Capability inventory: The agent's capabilities (tools, file access) are governed by an external 'authority profile' provided at runtime, rather than being hardcoded in this core skill.
- Sanitization: The skill lacks explicit sanitization or escaping of the input problem text, relying instead on the 'Interrogator' role to audit the logic and ensure no unauthorized remediation or authority expansion occurs.
Audit Metadata