insights-clustering
Pass
Audited by Gen Agent Trust Hub on Aug 3, 2026
Risk Level: SAFEPROMPT_INJECTION
Full Analysis
- [PROMPT_INJECTION]: The skill instructions create a surface for indirect prompt injection by ingesting untrusted text from ServiceNow record fields such as short_description, description, and journal comments. Malicious instructions embedded in these fields could influence the agent during clustering analysis or when generating recommendations.
- Ingestion points: Data enters the agent context through the SN-Query-Table MCP tool and REST API calls to ServiceNow tables in SKILL.md.
- Boundary markers: The procedure lacks delimiters or specific instructions to isolate retrieved data, making it harder for the agent to distinguish between valid data and embedded commands.
- Capability inventory: The skill possesses the ability to create records (SN-Create-Record) and access local shell environments (Bash), which increases the potential impact of a successful injection.
- Sanitization: There are no defined steps for escaping or validating the content retrieved from external records before it is processed by the agent.
Audit Metadata