The Agent Skills Directory

[COMMAND_EXECUTION]: The skill invokes several shell commands to manage and audit the developer workspace.
Evidence: Executes sociuu env list for environment inspection and glab mr list to query GitLab merge request statuses.
[PRIVILEGE_ESCALATION]: The skill performs maintenance tasks requiring elevated permissions within a virtualized environment.
Evidence: Uses colima ssh -- sudo fstrim -v /var/lib/docker to reclaim host disk space by trimming the filesystem inside the Docker VM.
[INDIRECT_PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection through the processing of untrusted external data.
Ingestion points: Ingests data from local worktree inventories and remote merge request information via glab in SKILL.md.
Boundary markers: No delimiters or warnings are used to separate external data from instructions.
Capability inventory: The skill has the capability to execute shell commands and remove environments via the sociuu CLI and colima.
Sanitization: Branch names and MR data are interpolated directly into shell commands (e.g., glab mr list --source-branch "$branch") without explicit escaping or validation shown in the instructions.

sociuu-cleanup