ontology-evolution
Pass
Audited by Gen Agent Trust Hub on Oct 18, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to update core business definitions and rules based on information gathered from external sources, which could contain malicious instructions.
- Ingestion points: Data enters the agent's context through policy documents (制度), interface data (接口), code/data samples (样例), and direct user statements (用户说法).
- Boundary markers: The instructions lack specific requirements for using delimiters or 'ignore' instructions when parsing these external data sources to prevent the agent from following embedded malicious commands.
- Capability inventory: The skill involves writing to configuration files (ontology.yaml and delta templates) and performing version control operations (git commit), which updates the logic used by other skills or agents.
- Sanitization: While the skill mentions 'anonymization' (脱敏) for data privacy when moving assets to public repositories, it does not specify content sanitization or validation to prevent the propagation of injected instructions into the ontology.
Audit Metadata