The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill automatically triggers package managers including npm, pip, poetry, go, and cargo to fetch and install dependencies from well-known public registries when relevant configuration files (e.g., package.json, requirements.txt) are detected.
[COMMAND_EXECUTION]: The skill executes multiple shell-based commands for environment setup, including git worktree add for workspace creation, git commit for persisting .gitignore updates, and arbitrary test runners such as npm test, cargo test, pytest, and go test to verify the baseline state of the worktree.
[PROMPT_INJECTION]: The skill identifies a potential surface for indirect prompt injection by extracting configuration preferences from the CLAUDE.md file using grep. If an attacker provides a repository with a maliciously crafted CLAUDE.md, they could attempt to influence the file paths or command parameters used by the agent during worktree setup.
Ingestion points: The CLAUDE.md file via the command grep -i "worktree.*director" CLAUDE.md.
Boundary markers: Absent; there are no delimiters or instructions provided to the agent to disregard instructions embedded within the configuration file.
Capability inventory: The skill possesses the capability to modify the file system, alter repository history, download external code via package managers, and execute project-defined scripts and binaries.
Sanitization: No validation or sanitization logic is present to verify the integrity or safety of the directory paths or strings extracted from the configuration file before they are incorporated into subsequent shell operations.

using-git-worktrees