signals-scout-ai-observability
Pass
Audited by Gen Agent Trust Hub on Sep 20, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from LLM traces and evaluation results, which presents a surface for potential indirect prompt injection attacks.
- Ingestion points: Untrusted data enters the agent context through
query-llm-trace,execute-sql(reading$ai_generation,$ai_trace, and$ai_evaluationevents), andscout-scratchpad-searchas described inSKILL.md. - Boundary markers: The skill does not define specific delimiters or instructions to ignore embedded commands when processing the content of LLM traces or generation events.
- Capability inventory: The agent has the capability to write to a durable scratchpad via
scout-scratchpad-rememberand to author or edit inbox reports throughscout-emit-reportandscout-edit-report. - Sanitization: There is no evidence of specific sanitization, filtering, or validation logic applied to trace content before it is included in reports or stored in the agent's memory.
Audit Metadata