jinko-task-cmaes
Pass
Audited by Gen Agent Trust Hub on Oct 23, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes external data tables and fitness sources as part of the calibration workflow.
- Ingestion points: The skill reads fitness-ready data tables and output sets containing objectives as described in the Inputs section of SKILL.md.
- Boundary markers: No explicit prompt delimiters or boundary markers are defined for the processed data content.
- Capability inventory: The skill utilizes scripts for creating and running calibrations, fetching status, and retrieving performance results (documented in SKILL.md).
- Sanitization: The workflow includes a mandatory "Review its dry-run output before applying it" step to inspect the configuration before execution.
- [SAFE]: References to external repositories (e.g.,
novainsilico/jinko-skills) and thejinko-sdkdependency are consistent with the skill's stated purpose and identify as resources belonging to the vendor (Nova In Silico). - [SAFE]: The use of session codes for SDK authentication is a documented operational requirement of the Jinkō platform for managing secure connections.
Audit Metadata