What problem does it solve? Running commands on remote Ascend NPU servers often fails due to repeated SSH connections hitting MaxStartups limits, broken quoting across nested shells, CRLF line endings from Windows-edited scripts, and tasks dying when the SSH session drops. This Skill provides a single reusable execution channel that handles connection reuse, in-container execution, idle card selection, and file transfer safely. ## Core Features & Use Cases - Single-Connection SSH Execution: Reuse one paramiko connection for all remote commands via scripts/ssh_helper.py, with optional docker exec wrapping, concurrent stdout/stderr reads, and credential priority (MINDIE_SSH_PASSWORD env var > interactive prompt > --password). - Idle NPU Card Selection: Scan npu-smi HBM usage inside a container with scripts/pick_free_device.py to pick the least-loaded card after health prechecks. - File Transfer & Long-Task Discipline: SFTP over the reused connection, CRLF-to-LF conversion for text scripts only, nohup/docker exec -d backgrounding, and distinct destination directories to avoid same-name overwrites. - Use Case: You need to check CANN version, inspect NPU status, and create a working directory inside a remote container. Connect once with ssh_helper.py, chain the commands with semicolons, and avoid being rejected by the remote MaxStartups limit. ## Quick Start Ask the AI to connect to your remote Ascend server with the host IP, username, and container name, then run an npu-smi status check using the remote-access skill's ssh_helper script while reusing a single SSH connection.
