AgentPlays
Qoder open-sources Better Harness for evaluating coding-agent workflows
Qoder said Oct 15 that it is officially open-sourcing Better Harness under the MIT license. The toolkit checks how coding agents actually handled real tasks. You run /better-harness through a plugin or CLI in Claude Code, Codex, Cursor or Qoder (the README lists more hosts, including GitHub Copilot CLI). It gathers session, project and agent-setup evidence and returns findings, each with its evidence, the impact, the smallest fix and a way to verify it. Qoder says the hosts share one evaluation model but don't yet have full parity: Qoder's integration is the most complete, and Cursor loads from source. (Alibaba Cloud)
How you’d use it: In Claude Code, Codex, Qoder, or Cursor, install Better Harness from QoderAI/better-harness and run /better-harness on a project where an agent recently finished a task. Read Findings that cite session evidence first; treat missing evidence as a gap, not a failure. Start with one verifiable fix.