The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses dynamic context injection (the !command syntax) to execute shell commands (git branch, git rev-parse, git worktree list, and test) at load time to populate its environment variables.
[COMMAND_EXECUTION]: The instructions direct the agent to execute chmod +x on local scripts located in ${CLAUDE_PLUGIN_ROOT}/scripts/ if they are found to be non-executable.
[PROMPT_INJECTION]: The skill contains an indirect prompt injection attack surface where untrusted metadata from the local git environment is ingested into the agent's context.
Ingestion points: SKILL.md (via the branch and worktrees dynamic context variables)
Boundary markers: Absent; untrusted data is processed without delimiters or warnings
Capability inventory: Shell command execution via dynamic context and local script execution (git-clean-audit, git-clean-apply)
Sanitization: None detected; the skill assumes git metadata like branch names and paths are benign

git-sweep