The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes a wide range of system commands to survey disk usage and perform cleanup tasks. This includes file system operations (du, find, rm), container management (docker), package management (apt), and system logging (journalctl). These actions are standard for the skill's stated purpose of system maintenance.
[INDIRECT_PROMPT_INJECTION]: The skill has an attack surface for indirect prompt injection as it processes untrusted data from the filesystem and system services.
Ingestion points: Data enters the agent context through the output of survey commands such as du, find, lsof, and docker ps as described in SKILL.md.
Boundary markers: The skill establishes a 'Triage' procedure where the agent must present a list of targets to the user, providing a human-in-the-loop verification step before any potentially destructive commands are executed.
Capability inventory: The skill utilizes commands that modify the system state, including rm, docker rmi, git gc, and apt autoclean across its procedures in SKILL.md.
Sanitization: The skill does not perform automated sanitization of file paths or image tags, relying instead on the manual review process defined in the triage section to mitigate risks from maliciously named files or metadata.

disk-cleanup