The Agent Skills Directory

[COMMAND_EXECUTION]: The skill involves executing shell commands to measure disk usage and clear caches. Security is maintained through a mandatory 'approval first' rule, where the agent must display the exact command and its targets before execution.
[INDIRECT_PROMPT_INJECTION]: The skill parses project configuration files (e.g., .ruby-version, .nvmrc) using scripts/list-pinned-versions.sh to protect active toolchain versions.
Ingestion points: Files like .ruby-version, .nvmrc, and .tool-versions in user-specified workspaces are read by the script.
Boundary markers: Explicit user approval is required for all resulting cleanup commands.
Capability inventory: The skill can execute deletion commands (e.g., rm -rf, brew cleanup, docker builder prune) via the agent.
Sanitization: scripts/list-pinned-versions.sh uses structured awk parsing to extract specific version strings, stripping comments and whitespace to prevent instruction injection.
[PRIVILEGE_ESCALATION]: References to sudo in references/cache-locations.md are limited to standard system maintenance tasks (e.g., cleaning systemd journals or package caches) and are only presented as instructions for the user to approve, adhering to best practices for system cleanup utilities.
[DATA_EXPOSURE_AND_EXFILTRATION]: Instructions explicitly forbid the deletion of personal user data (documents, photos, source trees), and the skill performs no network operations. The analysis and execution are confined to the local file system.

freeing-disk-space