The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses subprocess to execute SSH commands locally and bash commands on remote cluster nodes. This is the primary mechanism for all cluster operations, including training management and system monitoring.
[REMOTE_CODE_EXECUTION]: Deploys a persistent Python script (agent.py) to the remote cluster. This script acts as a high-performance execution bridge, receiving commands via JSON-Lines over stdin and returning results via stdout. This deployment is documented and essential for the skill's low-latency execution feature.
[REMOTE_CODE_EXECUTION]: Automatically configures and manages Mutagen file sync sessions, which involves the automated deployment of the Mutagen agent binary to the remote cluster via SSH.
[CREDENTIALS_UNSAFE]: Modifies the local ~/.ssh/config file to add host aliases for cluster nodes. To simplify workflows in dynamic environments, it utilizes options like StrictHostKeyChecking accept-new and UserKnownHostsFile /dev/null, which are documented and presented to the user during setup.
[DATA_EXFILTRATION]: All network activity is directed to user-specified SSH endpoints defined in the local configuration. No unauthorized external communication or data exfiltration patterns were identified.
[PROMPT_INJECTION]: The skill processes untrusted output from remote commands (e.g., logs, process lists) during health checks and status updates. It uses boundary markers (===SECTION===) to structure this data, which helps mitigate the risk of indirect prompt injection from malicious remote content.

remote-cluster-agent