The Agent Skills Directory

[COMMAND_EXECUTION]: The suite is designed to execute arbitrary shell commands on remote servers and inside Docker/Kubernetes containers through the ssh-exec.sh and ssh-job.sh scripts. This is a core feature of the ssh-connection module.
[EXTERNAL_DOWNLOADS]: The skill downloads source code and dependencies from several external repositories during the installation process, including official repositories from OpenMMLab and Facebook Research on GitHub, as well as the Ascend DrivingSDK on GitCode.
[DATA_EXFILTRATION]: The ssh-download.sh script enables the retrieval of files and directories from remote servers to the local machine, providing a mechanism for data transfer that could be used for exfiltration if mismanaged.
[CREDENTIALS_UNSAFE]: The ssh-connection/config.yaml file contains hardcoded internal IP addresses and a specific local file path to a private SSH key (C:/Users/qian/.ssh/id_ed25519). Exposure of these environment-specific details in the skill configuration is a minor security concern.
[DYNAMIC_EXECUTION]: The verify_mmlab_imports.py script uses the __import__ function to dynamically check for the presence of specific Python packages. Additionally, ssh-job.sh uses base64 encoding and decoding to wrap commands for execution via bash -c to handle quoting, which technically constitutes dynamic code execution.

ascend-model-migration