The Agent Skills Directory

[REMOTE_CODE_EXECUTION] (MEDIUM): The script create_worktree.sh automatically performs project setup by running npm install, pip install, or flutter pub get depending on detected files.
Evidence: scripts/create_worktree.sh lines 70-78.
Risk: These commands can execute arbitrary code defined in the repository's configuration files (e.g., preinstall scripts in package.json). If the user uses this skill on an untrusted or compromised repository, the system could be compromised during the "setup" phase.
[COMMAND_EXECUTION] (LOW): The script executes shell commands using variables derived from user input.
Evidence: scripts/create_worktree.sh lines 50 and 70 utilize $NEW_BRANCH, $BASE_BRANCH, and $WORKTREE_PATH inside shell commands.
Risk: While the variables are double-quoted to prevent simple word splitting, highly craft input or path traversal attempts in branch names could lead to unintended file system operations or git errors.
[PROMPT_INJECTION] (LOW): The skill is susceptible to Indirect Prompt Injection (Category 8) because it processes untrusted data from the local repository.
Ingestion points: package.json, requirements.txt, and pubspec.yaml files within the target repository.
Boundary markers: None present; the script blindly trusts the existence and content of these files for decision making.
Capability inventory: Subprocess execution of git, npm, pip, and flutter commands.
Sanitization: No validation is performed on the content of the repository files before processing.
[EXTERNAL_DOWNLOADS] (LOW): The skill triggers the download of external packages from public registries.
Evidence: Execution of npm install and pip install -r requirements.txt in scripts/create_worktree.sh.

git-worktree