tilegym-cutile-autotuning
Pass
Audited by Gen Agent Trust Hub on Jun 18, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: The skill provides comprehensive technical documentation for optimizing GPU kernels through the CuTile framework. No malicious patterns such as prompt injection, exfiltration, or persistence mechanisms were detected.
- [SAFE]: Includes an integrity signature file (
skill.oms.sig) using a Sigstore bundle and in-toto statement to verify the authenticity and provenance of the skill's contents, which is a recognized security best practice. - [SAFE]: Python code examples in the
assets/directory use established libraries (torch,cuda.tile) for hardware-specific performance tuning and do not introduce unverifiable dependencies or remote execution risks. - [SAFE]: Environmental variables like
DISABLE_AUTOTUNEare used legitimately for CI/CD determinism and profiling isolation, with no evidence of hidden conditional attacks.
Audit Metadata