database-testing
Pass
Audited by Gen Agent Trust Hub on Oct 6, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from external database sources, creating a potential surface for indirect prompt injection attacks.
- Ingestion points: The skill reads database schemas, table metadata, and row content through
db.rawqueries inSKILL.mdand fixtures defined inevals/database-testing.yaml. - Boundary markers: Absent. There are no instructions or delimiters provided to ensure the agent ignores or treats natural language instructions found within database contents (e.g., within string columns or table comments) as data rather than instructions.
- Capability inventory: The skill possesses high-privilege database capabilities, including arbitrary SQL execution via
db.rawand the ability to initiate new agent tasks using theTaskprimitive. - Sanitization: Absent. The provided instructions do not specify any validation, escaping, or filtering routines for the data returned from database queries before it is used in subsequent agent reasoning or task generation.
Audit Metadata