The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses shell commands and a local bash script to perform Git operations such as adding/removing worktrees and deleting branches.
Evidence: SKILL.md defines templates for git worktree add, git worktree remove, and git worktree prune.
Evidence: SKILL.md invokes a local script scripts/prune-squashed.sh to handle branch cleanup.
[INDIRECT_PROMPT_INJECTION]: The skill processes repository state data, specifically branch names, which are interpolated into shell commands. This creates a theoretical surface for indirect injection if the repository contains maliciously crafted branch names.
Ingestion points: Git branch names retrieved via git worktree list and git for-each-ref in scripts/prune-squashed.sh.
Boundary markers: The skill requires a strict naming convention (<type>/<slug>) and forces a human authorization gate before any command is executed.
Capability inventory: File system modification (worktree directories) and repository modification (branch deletion).
Sanitization: The bash script properly quotes variables (e.g., "$branch", "$target") to prevent shell word splitting and basic injection attacks.
[SAFE]: The skill implements best practices for agent safety by using disable-model-invocation: true in the frontmatter and establishing a 'Pre-Mutation Gate' that halts execution until a human reviews and approves the planned changes.

worktree