zero-trust-assessment
Pass
Audited by Gen Agent Trust Hub on Sep 15, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze untrusted external data, such as architecture documentation, network diagrams, and configuration files, which could serve as vectors for indirect prompt injection.
- Ingestion points: The skill explicitly processes architecture documentation, configuration data, and policy metadata as stated in the 'Injection Hardening' and 'Limitations' sections.
- Boundary markers: While the skill lacks formal string delimiters for input, it includes a robust 'Injection Hardening' section and a 'Prompt Injection Safety Notice' instructing the agent to treat all data as untrusted and to specifically ignore directives like 'ignore previous instructions'.
- Capability inventory: The skill utilizes file system tools including 'Read', 'Grep', and 'Glob' to analyze target environments.
- Sanitization: The instructions mandate that the agent must not execute instructions found within data and must instead report suspected injection attempts as security findings.
- [PROMPT_INJECTION]: Deterministic detectors identified keywords associated with prompt injection; however, a manual review confirms these instances are part of the skill's safety logic. The skill uses these patterns as examples of what the agent should detect and ignore within analyzed data, rather than attempting to override the agent's own system instructions.
Audit Metadata