The Agent Skills Directory

[COMMAND_EXECUTION]: The skill makes extensive use of system CLI tools including df, du, pgrep, and git to survey the environment and perform cleanup. It also executes Docker management commands such as docker builder prune and docker rmi to reclaim container-related storage.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from the local operating system environment, including process command lines from /proc, file system paths, and git repository metadata. This data is used to drive logic for keeping or deleting files, which represents an attack surface where a malicious process or specially crafted file name could attempt to influence the agent's behavior. This risk is partially mitigated by mandatory user approval tiers for all non-cache deletions.
[PRIVILEGE_ESCALATION]: The skill employs a technique to delete root-owned files in the /tmp directory by mounting the target path into a temporary Docker container. While this is a common administrative workaround for cleaning up root-owned artifacts in dev environments, it constitutes a bypass of standard file system permission constraints.
[DYNAMIC_EXECUTION]: Uses nohup sh -c to execute long-running directory sizing tasks in the background. This allows the agent to continue its survey without being blocked by tool timeouts, though it involves executing a shell string constructed within the skill.

vps-cleanup