querying-posthog-data

Pass

Audited by Gen Agent Trust Hub on Sep 14, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONDATA_EXFILTRATION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill proactively addresses the risk of indirect prompt injection by instructing the agent to treat user-authored metric definitions as untrusted data. It explicitly forbids the agent from obeying any instructions embedded in that data, such as tool calls or attempts to override system prompts.
  • [DATA_EXFILTRATION]: The skill identifies potentially sensitive analytics data, specificallyconversation histories, system prompts, and application state stored in AI observability tables. It provides a safety mandate to never output these heavy fields directly into the conversation, requiring them to be written to files for local analysis instead.
  • [SAFE]: The skill implements a secure 'discovery-before-query' workflow, using information_schema to verify data structures before execution. It also promotes resource safety through progressive exploration (counting rows before large aggregations) and limiting the size of select statements to prevent out-of-memory errors.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 14, 2026, 10:27 PM
Security Audit — agent-trust-hub — querying-posthog-data