The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses Git commands through subprocesses to manage worktrees. This is the core functionality and is implemented with multiple safety checks:
It verifies paths to ensure they stay within the expected parent directory (preventing directory traversal).
It uses check-ref-format to validate Git branch names before use.
It enforces strict base-commit validation using specific SHA hashes from origin/develop to ensure a consistent state.
[DATA_EXFILTRATION]: While the skill interacts with the remote via git fetch and git ls-remote, it does not transmit sensitive data. It only retrieves metadata about branch states and commit SHAs to ensure local workspace consistency.
[PROMPT_INJECTION]: The instructions include clear boundaries and constraints (e.g., "not modify business code", "not submit, push, or create Issue/MR"). It specifically restricts the use of sub-skills like zc-bug-fix to read-only operations, preventing unauthorized downstream actions.

task-worktree