exploring-llm-traces
Pass
Audited by Gen Agent Trust Hub on Sep 25, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill instructs the agent to execute local Python scripts such as scripts/print_summary.py and scripts/extract_conversation.py to process and summarize large JSON trace data. These scripts are provided within the skill package and perform standard data parsing, formatting, and text truncation tasks.
- [INDIRECT_PROMPT_INJECTION]: The skill handles untrusted data in the form of AI traces containing LLM inputs and outputs. Ingestion points: Data retrieved via posthog:query-llm-trace tools and saved to local JSON files. Boundary markers: The scripts use truncation via the MAX_LEN environment variable to limit the amount of content processed at once. Capability inventory: Execution of Python scripts via shell and execute-sql for data analysis. Sanitization: Truncation of large content fields in the Python parsing scripts reduces the risk of context-based injection attacks.
- [SAFE]: The skill interacts with official PostHog MCP tools and refers to legitimate vendor infrastructure (app.posthog.com). The provided documentation and scripts are functional and directly support the stated goal of AI observability without any detected malicious patterns or unauthorized data exfiltration.
Audit Metadata