simulink-curating-library-kg
Pass
Audited by Gen Agent Trust Hub on Aug 30, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONDYNAMIC_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from .satk/library-cache/*.json files, which contain block metadata like names and descriptions. This content is used by the agent to generate new library descriptions, creating a surface for indirect prompt injection if a library contains malicious strings.
- Ingestion points: The agent reads metadata from .satk/library-cache/*.json.
- Boundary markers: None. Instructions do not specify delimiters or ignore-instructions for the processed data.
- Capability inventory: The skill has access to evaluate_matlab_code (manifest.yaml) which executes arbitrary MATLAB commands, and file-writing capabilities through the library.LibraryCuration.save API.
- Sanitization: The skill does not perform validation or escaping of ingested strings.
- [COMMAND_EXECUTION]: The skill relies on the evaluate_matlab_code tool to perform its primary functions, including saving curation data and regenerating the knowledge graph using MATLAB APIs (library.LibraryCuration, library.kg.Populate). This execution is limited to the defined workflow and uses the vendor's provided toolkit.
- [DYNAMIC_EXECUTION]: The skill generates MATLAB code at runtime to populate data structures and call API methods. This includes constructing containers.Map objects and setting fields based on agent-inferred descriptions and categories. This represents low-risk script generation from known templates for task automation within the MATLAB environment.
Audit Metadata