llm-obs-session-classify

Warn

Audited by Gen Agent Trust Hub on Jun 12, 2026

Risk Level: MEDIUMCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill uses the pup CLI tool to query Datadog's LLM Observability, RUM, and Audit Log APIs via shell commands.\n- [COMMAND_EXECUTION]: The skill generates and executes Python scripts locally to process and summarize large JSON data payloads that are persisted to the file system.\n- [PROMPT_INJECTION]: The skill processes untrusted conversation content from traces to perform classification, introducing a surface for indirect prompt injection.\n
  • Ingestion points: Conversation messages and user queries retrieved via get_llmobs_agent_loop and get_llmobs_span_content in SKILL.md.\n
  • Boundary markers: None identified in the processing logic.\n
  • Capability inventory: Subprocess execution (pup) and file system access (open).\n
  • Sanitization: No specific sanitization or filtering of retrieved conversation content is mentioned before processing.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Jun 12, 2026, 08:21 PM
Security Audit — agent-trust-hub — llm-obs-session-classify