Testing Skills With Subagents
Pass
Audited by Gen Agent Trust Hub on Feb 20, 2026
Risk Level: SAFEPROMPT_INJECTION
Full Analysis
- [PROMPT_INJECTION] (LOW): The file contains documentation variants and scenarios that utilize high-authority markers and imperatives (e.g., 'IMPORTANT:', 'THIS IS EXTREMELY IMPORTANT') to force behavioral compliance. Evidence: Scenario 1-3 use 'IMPORTANT: This is a real scenario. Choose and act.' Variant C uses 'THIS IS EXTREMELY IMPORTANT. BEFORE ANY TASK, CHECK FOR SKILLS!' and 'If a skill existed... you failed.'
- [INDIRECT_PROMPT_INJECTION_SURFACE] (LOW): The file exposes a surface for indirect prompt injection by housing instructions intended to influence agent execution during data processing. 1. Ingestion point: examples/CLAUDE_MD_TESTING.md. 2. Boundary markers: Absent. 3. Capability inventory: Local shell execution of 'ls' and 'grep'. 4. Sanitization: Absent.
Audit Metadata