eval

To evaluate and rank multiple agent outputs, you can execute custom performance metrics or use LLM-based diff analysis. This skill supports automated session management and comparative benchmarking to validate agent performance objectively.

LLM judge mode is a qualitative analysis approach that compares git diffs and result documentation to rank agents. It evaluates correctness and code quality to provide comprehensive performance insights for software engineering workflows.

Yes, you can use hybrid ranking to combine quantitative metrics with qualitative LLM judgment. This approach breaks ties and provides comprehensive performance insights when benchmarking multiple agent worktrees.

Yes, agent evaluation requires integration with session state tracking and git-based version control. These dependencies are necessary to validate agent performance across multiple worktrees and manage comparative benchmarking sessions.

To measure performance metrics like latency across agent worktrees, you can execute custom commands within the metric-based evaluation mode. This allows you to quantitatively assess and rank agent results by running specified performance benchmarks.

The best way to automate comparative benchmarking is to run a skill that evaluates all agents in the current session using default or custom criteria. It executes performance metrics and qualitative diff analysis to rank results automatically.