setting-up-warehouse-properties
Pass
Audited by Gen Agent Trust Hub on Sep 4, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes external schema metadata from data warehouses, which represents a surface for indirect prompt injection if column names or data types contain malicious instructions.
- Ingestion points: Schema and column data are ingested via
external-data-schemas-retrieveandview-getas described inSKILL.md. - Boundary markers: The skill includes a mandatory manual review step, instructing the agent to "show it to the user for approval" before any creation tools are invoked.
- Capability inventory: The agent utilizes
custom-property-sources-createandcustom-property-sources-backfillto write data and trigger historical processing within the PostHog environment. - Sanitization: The instructions explicitly direct the agent to filter out identity properties (such as
email,name, or$prefixes) and to identify and flag sensitive columns like password hashes or tokens to prevent accidental exposure.
Audit Metadata