sales-hiver
Pass
Audited by Gen Agent Trust Hub on Jul 13, 2026
Risk Level: SAFEPROMPT_INJECTION
Full Analysis
- [PROMPT_INJECTION]: The skill implements a self-improving mechanism by reading from and writing to
references/learnings.md. This creates a surface for indirect prompt injection where untrusted data could be persisted and later read as authoritative context. - Ingestion points: The skill reads
references/learnings.mdat the start of the interaction in Step 1. - Boundary markers: No specific delimiters or safety instructions are present to differentiate the accumulated 'learnings' from the skill's instructions.
- Capability inventory: The skill has instructions to append data to the learnings file and routes requests to other specialized skills.
- Sanitization: There is no evidence of validation or sanitization of the content before it is stored in the local file.
Audit Metadata