The Agent Skills Directory

[COMMAND_EXECUTION]: The remote-exec.sh script enables the execution of arbitrary commands on remote hosts. The exec subcommand passes arguments directly to ssh, allowing the AI agent to perform any action the user's SSH credentials permit on the target server.
[DATA_EXPOSURE]: The skill reads and parses the user's private ~/.ssh/config file. While intended to simplify setup by auto-importing host aliases, this exposes the user's remote infrastructure details, including hostnames, IP addresses, and usernames, to the agent's context.
[INDIRECT_PROMPT_INJECTION]: The skill creates a surface for indirect prompt injection by processing data from untrusted remote systems.
Ingestion points: Content from remote files (e.g., via cat) and stdout/stderr from executed commands are fed directly back to the agent.
Boundary markers: The instructions do not define delimiters or warnings to help the agent distinguish between command output and new instructions.
Capability inventory: The skill provides full remote shell access and file manipulation capabilities.
Sanitization: There is no evidence of filtering or sanitizing remote output before it is processed by the AI.

remote-exec