The Agent Skills Directory

[SAFE]: The skill implements a robust security model that treats all repository metadata and project files as untrusted input. It utilizes a shared trust boundary (node_environment.py) to sanitize the execution environment, preventing the execution of malicious binaries that might be placed within the project directory (binary shadowing).
[SAFE]: The test runner (run_vitest.py) employs a strict allowlist for environment variables and utilizes a specific regex pattern to ensure that package.json scripts are only auto-executed if they are direct, safe Vitest invocations without shell chaining or redirection.
[SAFE]: The skill includes explicit defenses against various obfuscation techniques, such as right-to-left override (RLO) characters and zero-width characters in command output, ensuring that the agent and the user are not misled by the displayed commands.
[SAFE]: While the skill accesses local project files and executes Vitest via subprocess, all command construction is carefully vetted, uses shell=False (implicitly by passing lists to subprocess.run), and resolves binaries against a filtered PATH that excludes the project itself, significantly mitigating the risk of arbitrary code execution from malicious repository content.

vitest