eval-harness

Create a new evaluation for the eval-harness framework. Use when asked to build, generate or scaffold an eval that tests a CLI coding agent (Claude Code, OpenCode, Copilot, Codex, Pi).

Run evaluations with the eval-harness framework. Use when asked to run, execute, benchmark or compare CLI coding agents (Claude Code, OpenCode, Copilot, Codex, Pi) on existing evals.

Monitor and interpret evaluation runs of the eval-harness framework. Use when asked to watch, interpret, debug or diagnose an eval run, its logs, its scores or its results.