customer-research
Pass
Audited by Gen Agent Trust Hub on Jul 23, 2026
Risk Level: SAFEPROMPT_INJECTIONNO_CODE
Full Analysis
- [PROMPT_INJECTION]: The skill's core functionality involves processing untrusted external data from multiple sources, making it vulnerable to indirect prompt injection.
- Ingestion points: Processes local files (transcripts, surveys, tickets) and remote web content (Reddit threads, G2 reviews, forums) to extract signal and themes.
- Boundary markers: No instructions are provided to the agent to use delimiters or specific framing to prevent it from following instructions found within the ingested data.
- Capability inventory: The skill is designed to have the agent read diverse local and remote text assets and synthesize findings.
- Sanitization: There are no sanitization steps defined for the external content before it is processed by the model's reasoning engine.
Audit Metadata