audit-ai-compliance
Pass
Audited by Gen Agent Trust Hub on Oct 31, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to analyze external AI components, code, and documentation, creating a potential surface for indirect prompt injection through ingested data.
- Ingestion points: The skill uses
read_fileandsearch_codeto identify AI system architecture, model sources, and training data provenance (SKILL.md). - Boundary markers: The instructions explicitly include a security guardrail (ASI01 Goal Hijack) directing the agent to reject prompt injection attempts embedded in audit targets, though no technical separators are defined.
- Capability inventory: The skill is permitted to create and modify files using
write_file,edit_file, andcreate_fileto produce compliance reviews (SKILL.md). - Sanitization: No technical sanitization or schema-based validation is specified for the input data phase.
Audit Metadata