The Agent Skills Directory

[Automated Command Execution]: The skill is designed to automatically detect project types and run installation commands such as npm install, pip install, and cargo build. While beneficial for workflow automation, these commands can trigger execution of arbitrary scripts defined in project configuration files (like preinstall scripts in package.json), which warrants caution when used in untrusted repositories.
[Indirect Prompt Injection Surface]: The skill operates by reading and acting upon data found in project files like requirements.txt or package.json. This creates a potential surface where malicious project content could influence the agent's actions, although the skill mitigates this by instructing the agent to ask for permission if setup steps are ambiguous.
[Command Parameter Handling]: Shell commands in the skill incorporate variables such as $BRANCH_NAME. While the agent is directed to generate these names, there is a theoretical consideration for command injection if the input used to generate branch names is not properly handled or contains shell metacharacters.
[Automated Test Execution]: The skill automatically runs project-specific test commands (e.g., npm test, pytest). Similar to installation scripts, running tests involves executing code from the local environment, which is an important consideration for security-conscious workflows.

using-git-worktrees