The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes child_process.exec in src/lib.ts to perform test runs. This function invokes a system shell, which is inherently risky if the command string contains unsanitized input.
[COMMAND_EXECUTION]: The customCommand argument from the command line (process.argv[3]) is passed directly to the shell without any filtering or sanitization in src/index.ts, permitting arbitrary command injection.
[REMOTE_CODE_EXECUTION]: The logic in src/lib.ts automatically extracts and executes command strings from a project's package.json file. An attacker can craft a malicious repository with dangerous code in the test script field to achieve code execution when the agent analyzes the repo.
[DATA_EXFILTRATION]: The skill returns the full output of the executed commands. A malicious command could be used to read local sensitive files like ~/.ssh/id_rsa or .env, and the contents would be returned in the skill's response.

test-genie