npu-docker-launcher
Warn
Audited by Gen Agent Trust Hub on Sep 16, 2026
Risk Level: MEDIUMCOMMAND_EXECUTIONPRIVILEGE_ESCALATIONDATA_EXFILTRATIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill dynamically constructs and executes complex
docker runshell commands using user-provided inputs for image names, volume mounts, and environment variables. - [PRIVILEGE_ESCALATION]: By default, the skill configures containers with the
--privilegedflag. This grants the container full access to the host's devices and kernel, effectively bypassing standard container security boundaries. It also provides instructions for adding users to privileged system groups (e.g., the 'huawei' group). - [DATA_EXFILTRATION]: The skill allows users to mount any host directory into the container via the
-vflag. This capability can be used to expose sensitive host files, such as SSH keys, credentials, or system configurations, to the container environment. - [INDIRECT_PROMPT_INJECTION]: The skill parses and displays output from local system tools like
docker images,ls, andnpu-smi. If an attacker can influence these outputs (for example, by creating a Docker image with a malicious name), they could potentially inject instructions into the agent's context.
Audit Metadata