hyperloom-workload-optimizer
Pass
Audited by Gen Agent Trust Hub on Aug 16, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill downloads the 'hyperloom-inference-optimizer' package from the Python Package Index (PyPI) during Phase 0 bootstrap. This is a vendor-owned resource (AMD) and is installed using standard package management. Evidence: 'pip install hyperloom-inference-optimizer' instructions in SKILL.md and setup.md.
- [COMMAND_EXECUTION]: The skill executes local shell scripts and hardware utilities to orchestrate the optimization process and monitor GPU health. Evidence: Execution of 'launch.sh', 'install.sh', and 'preflight.py', the latter of which uses 'subprocess.run' to interface with hardware tools like 'amd-smi' and 'rocm-smi'.
- [INDIRECT_PROMPT_INJECTION]: The skill includes an attack surface for indirect prompt injection as it processes external model paths and workload configurations. 1. Ingestion points: Workload parameter intake in SKILL.md Phase 2. 2. Boundary markers: Implements a mandatory confirmation gate requiring the user to approve a structured launch plan before any Phase 3 execution. 3. Capability inventory: Capability to execute background shell processes via 'setsid' and 'nohup'. 4. Sanitization: Uses quoted heredocs in scripts to prevent unexpected variable expansion when writing configuration files.
Audit Metadata