knowledge-graph-construction
Pass
Audited by Gen Agent Trust Hub on Sep 15, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill defines a workflow for ingesting and processing unstructured data from external sources such as papers, clinical notes, web pages, and logs to build structured knowledge. This introduces a surface for indirect prompt injection, where malicious instructions hidden in the source data could attempt to influence the agent's behavior during extraction or validation.
- Ingestion points: Input data sources cataloged in SKILL.md (Step 1) include various external document types.
- Boundary markers: The skill recommends using few-shot examples and multi-round verification (SKILL.md Step 4, resources/methodology.md Section 1), which can act as a partial mitigation, although explicit security delimiters are not mandated.
- Capability inventory: The workflow includes populating graph databases (e.g., Neo4j, Neptune) and performing automated validation checks as specified in SKILL.md (Step 6) and resources/methodology.md (Section 4).
- Sanitization: The methodology emphasizes data quality through schema enforcement, entity normalization, and multi-round verification in resources/methodology.md (Sections 4, 5, and 6), though these are primarily aimed at accuracy rather than security hardening.
Audit Metadata