modelopt-quant-recipe-search
Pass
Audited by Gen Agent Trust Hub on Nov 13, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: The skill implements a modular design that delegates all resource-intensive or potentially risky operations—such as checkpoint generation (
modelopt-ptq), serving (modelopt-deployment), and job monitoring (modelopt-monitor)—to dedicated external skills. This separation of concerns limits the search strategy skill to high-level planning and state recovery. - [INDIRECT_PROMPT_INJECTION]: The design workflow involves reading external artifacts including result tables, recipe logs, sensitivity reports, and experiment notes. While this exposes an ingestion surface for indirect prompt injection from potentially attacker-controlled data, the risk is managed by the skill's reliance on structured ML-specific data and its focus on technical optimization rather than open-ended natural language processing.
- [COMMAND_EXECUTION]: The instructions guide the agent to interact with established tools within the vendor's ecosystem (e.g.,
MLflow,Slurm, and themodelopt-*suite). No evidence of arbitrary shell injection, unauthorized credential access, or suspicious network exfiltration was found.
Audit Metadata