bdi-mental-states
Pass
Audited by Gen Agent Trust Hub on Apr 30, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: The skill is designed for knowledge representation and cognitive modeling using semantic web technologies.
- [EXTERNAL_DOWNLOADS]: The code references legitimate libraries such as 'rdflib' for graph manipulation and 'fipa_acl' for agent communication. It does not perform any unauthorized remote code downloads or silent script executions.
- [PROMPT_INJECTION]: The skill utilizes Logic Augmented Generation (LAG) as a safety and consistency measure, forcing the model to constrain its output within a predefined ontological schema, which effectively limits the potential for instruction bypass.
- [INDIRECT_PROMPT_INJECTION]: The skill ingests external context to construct cognitive representations. This potential surface is mitigated by the inclusion of a validation pipeline in 'references/framework-integration.md' that parses and verifies generated RDF triples against the BDI ontology before they are accepted (e.g., verifying that every intention is supported by a belief).
- [DATA_EXFILTRATION]: There is no evidence of the skill accessing sensitive environment variables, credentials, or local files for exfiltration. Network operations are limited to standard SPARQL endpoint interactions for mental state storage.
Audit Metadata