jinko-task-define-param-to-calibrate
Pass
Audited by Gen Agent Trust Hub on Oct 23, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: No security issues detected. The skill performs specialized parameter management using a local Python script for validation and an official SDK for model interaction.
- [INDIRECT_PROMPT_INJECTION]: The skill processes user-provided documentation and evidence to classify model parameters. This represents an indirect prompt injection surface where instructions could be hidden in external evidence. However, the risk is minimized by the use of a validation script (
apply_calibration_labels.py) that strictly checks the generated mutation plan for type safety and consistency with the model structure before any modifications are applied. - Ingestion points: User-supplied evidence documentation (SKILL.md workflow).
- Boundary markers: None explicitly mentioned in the prompt instructions for external text.
- Capability inventory: Reads model components and creates/modifies tags via the
jinkoSDK (scripts/apply_calibration_labels.py). - Sanitization: The validation script performs strict type and value checking on the mutation plan JSON and verifies component identifiers against the live model state.
- [EXTERNAL_DOWNLOADS]: The skill depends on the
jinko-sdk(pip) and refers to a prerequisite skill located in thenovainsilico/jinko-skillsrepository. These references are legitimate and consistent with the vendor's provided infrastructure.
Audit Metadata