carto-find-spatial-data
Pass
Audited by Gen Agent Trust Hub on Sep 17, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted metadata (dataset names, descriptions, and provider details) from the CARTO Data Observatory. This metadata is retrieved through search tools and incorporated into the agent's context, creating a surface where a maliciously crafted dataset entry could attempt to influence the agent's behavior during discovery.
- Ingestion points: Metadata is ingested via the
search_data_observatorytool andcarto do searchCLI command as detailed inreferences/data-observatory.md. - Boundary markers: There are no explicit boundary markers or instructions provided in the skill to isolate or ignore instructions that might be embedded within the external metadata.
- Capability inventory: The skill allows the agent to manage subscriptions (materializing data into the user's warehouse) and execute SQL queries on the destination warehouse.
- Sanitization: No sanitization or validation of the external strings is documented before the agent evaluates the fit of the datasets.
Audit Metadata