reality-check
Pass
Audited by Gen Agent Trust Hub on Nov 9, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
- Trusted Source Download: The skill fetches the reality-check tool from Microsoft's official GitHub repository via uv. This is a routine procedure for developer tools from established vendors.
- Potential Indirect Prompt Injection: The validation command (check-software) uses natural language prompts to drive Agent SDKs. This represents a potential surface for indirect injection if the software being tested contains adversarial instructions.
- Evidence Chain for Indirect Injection: (1) Ingestion point: Natural language request input for software checks in SKILL.md. (2) Boundary markers: No delimiters defined in the instructions. (3) Capability inventory: Invokes external Agent SDKs (Copilot, Claude) to perform validation tasks. (4) Sanitization: No specific filtering or escaping of application content is described.
Audit Metadata