The Agent Skills Directory

[PROMPT_INJECTION] (LOW): Indirect Prompt Injection surface detected in the slash command template.
Ingestion points: The template in references/slash_command_template.md uses args.0 and args.1 (user-provided branch names) as direct inputs for shell script generation.
Boundary markers: Absent. The template directly places arguments into shell variable assignments.
Capability inventory: The generated scripts execute git worktree and git branch commands.
Sanitization: While the standalone scripts/create_worktree.sh uses sed 's/[^a-z0-9-]//g' for sanitization, the simplified command template provided in the markdown reference lacks this protection, creating a risk if the agent copies the template exactly for use with untrusted user input.
[COMMAND_EXECUTION] (SAFE): The provided shell scripts perform various git operations and file system deletions (rm -rf). These are legitimate for the skill's purpose and include interactive confirmations and input validation (e.g., regex checks for numeric selections in cleanup_worktrees.sh) to prevent accidental or malicious misuse.

git-worktrees