The Agent Skills Directory

[COMMAND_EXECUTION] (SAFE): The skill executes Git commands using user-provided input, but implements multiple layers of protection.
Input Validation: The script uses git check-ref-format --branch to validate names and bases, ensuring they conform to Git standards and preventing shell injection or command flag injection.
Path Safety: The assert_safe_worktree_path and assert_worktrees_dir functions check every component of a target path to ensure it is not a symbolic link, preventing attackers from using symlinks to write files or delete directories outside of the intended .worktrees/ directory.
Flag Injection Prevention: The script consistently uses the -- separator in Git commands (e.g., git worktree add -- "$target" "$name") to ensure that user-provided names are treated as positional arguments rather than command-line options.
Safe File Handling: The copy_env function copies environment files (.env*) using the cp -n (no-clobber) flag, preventing accidental overwriting of existing configuration. It also explicitly skips symbolic links to avoid copying files from outside the repository root.
Interactive Safety: The cleanup command requires explicit user confirmation before performing deletions and uses standard git worktree remove without the --force flag, minimizing the risk of accidental data loss.

worktree-kit