cuda-skill
Pass
Audited by Gen Agent Trust Hub on Mar 28, 2026
Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [SAFE]: The skill is primarily a documentation repository for NVIDIA CUDA technologies, including PTX ISA 9.1, Runtime API 13.1, and Driver API 13.1. Extensive review of the 492 files did not reveal any malicious code, obfuscation, or data exfiltration patterns.
- [COMMAND_EXECUTION]: The skill provides legitimate instructions for the agent to interact with well-known development tools such as
nvccfor compilation,nsysandncufor profiling, andcompute-sanitizerorcuda-gdbfor debugging. These tools are used appropriately for the skill's stated purpose of assisting with GPU code development. - [PROMPT_INJECTION]: The skill presents an attack surface for indirect prompt injection (Category 8) as it processes local documentation and allows for shell command execution. This is a characteristic of the skill's intended functionality as a developer reference tool. 1. Ingestion points: Documentation files in the
references/directory are searched and read by the agent (SKILL.md). 2. Boundary markers: None present; the content is technical documentation. 3. Capability inventory: Access to shell commands (nvcc,nsys,ncu, etc.) through the agent's toolset. 4. Sanitization: None. Despite this surface, no injection content was detected in the provided documentation files.
Audit Metadata