local-inference-tuning
Pass
Audited by Gen Agent Trust Hub on Jul 10, 2026
Risk Level: SAFE
Full Analysis
- [COMMAND_EXECUTION]: The skill uses standard, read-only system diagnostic commands (
sysctl,system_profiler,df,ollama list) to audit hardware capabilities such as CPU, RAM, and GPU. These are legitimate and necessary for the skill's purpose of hardware-specific tuning.- [DATA_EXFILTRATION]: No evidence of data exfiltration was found. The skill emphasizes binding local services to127.0.0.1(loopback) only and warns against external exposure (0.0.0.0) without a security review.- [EXTERNAL_DOWNLOADS]: While the skill assists in configuring model downloads, it contains explicit safety guidelines to verify storage capacity and user-approved volumes before initiating any large downloads. It does not perform automated remote code execution.- [PROMPT_INJECTION]: The instructions are focused on technical diagnostics and do not contain patterns typical of prompt injection or safety filter bypasses.
Audit Metadata