The Agent Skills Directory

[SAFE]: The skill serves as a documentation and methodology framework for testing. It does not contain any executable malicious payloads or obfuscated code.
[COMMAND_EXECUTION]: The framework uses standard, non-malicious commands for software verification, such as npm test, npm run build, and grep. These commands are contextually appropriate for a development-focused agent and are used for deterministic pass/fail grading.
[DATA_EXPOSURE]: No hardcoded credentials or access to sensitive system directories (e.g., .ssh, .aws) were detected. The skill interacts exclusively with project-specific directories like .claude/evals/ and standard source code paths.
[REMOTE_CODE_EXECUTION]: There are no patterns involving the download and execution of remote scripts or the installation of unknown third-party packages.

eval-harness