The Agent Skills Directory

[COMMAND_EXECUTION]: The skill is designed to execute remote commands. It constructs an SSH command string and pipes a multi-line bash script to the remote server to collect system metrics (Docker status, disk usage, memory, uptime, and logs).
[DATA_EXFILTRATION]: The skill accesses sensitive local file paths to establish connectivity. Specifically, it reads connection details from deploy/deploy.env and uses a private key located at deploy/vps.key for authentication. This is standard behavior for an SSH-based management tool.
[PROMPT_INJECTION]: The skill has an indirect prompt injection surface because it ingests and analyzes untrusted data from remote sources.
Ingestion points: Remote server command output and container logs (docker compose logs) are read into the agent context in SKILL.md.
Boundary markers: None are present; the agent is instructed to search the raw log output for specific patterns like 'ERROR' or 'Traceback'.
Capability inventory: The skill performs subprocess calls via ssh to retrieve data and performs local file reads.
Sanitization: There is no evidence of sanitization or filtering of the remote content before it is processed by the agent.

server-check