pg-graph
Pass
Audited by Gen Agent Trust Hub on Sep 17, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- Indirect Prompt Injection Surface: The skill is designed to process user-supplied data to derive ontologies and build knowledge graphs. This ingestion of external content represents a potential injection surface. The skill mitigates this risk by providing clear instructions to use safe interpolation and to never concatenate untrusted input into query strings. Specifically: (1) Ingestion points include user-controlled document tables and text columns. (2) Boundary markers are managed through strict query-level isolation. (3) Capabilities are scoped to database query execution via ag_catalog.cypher. (4) Sanitization is explicitly required for all external content interpolation.
- Secure Credential Handling: The skill includes procedures for configuring cloud AI services. It follows security best practices by instructing the agent to never hardcode, log, or commit subscription keys, and instead advocates for user-supplied input or the use of managed identities.
- Safe Usage of Database Extensions: The skill leverages established database extensions such as Apache AGE and azure_ai. It provides correct guidance on enabling these through managed platform parameters rather than unsafe local commands or unverifiable sources.
Audit Metadata