signals-scout-anomaly-detection

Pass

Audited by Gen Agent Trust Hub on Sep 10, 2026

Risk Level: SAFEDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [DYNAMIC_EXECUTION]: The skill utilizes Bash and python3 to execute statistical routines, specifically the bundled scripts/ks2.py script and inline Python heredocs. These components perform Kolmogorov-Smirnov tests and changepoint detection on data retrieved from the environment. This execution is confined to the agent's local sandbox and is a primary functional requirement for anomaly detection.\n- [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from external sources such as PostHog events, insight names, and dashboard descriptions. This ingestion of external content serves as a potential vector for indirect prompt injection. \n
  • Ingestion points: Data enters via scout-project-profile-get, dashboard-get, insight-query, and execute-sql as specified in SKILL.md.\n
  • Boundary markers: The instructions do not define explicit prompt delimiters for text data, but emphasize quantitative analysis which mitigates typical linguistic injection.\n
  • Capability inventory: The skill can emit reports, create notebooks, and execute SQL/Bash commands.\n
  • Sanitization: No specific text sanitization or filtering logic is documented for the data processed by the LLM.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 10, 2026, 04:00 PM
Security Audit — agent-trust-hub — signals-scout-anomaly-detection