Alibaba Cloud's AI coding tool, Qoder, just open-sourced Better Harness under the MIT license. It's a tool that inspects your codebase to see if it's actually ready for a coding agent to work on — checking for rules, tools, and validation processes.
The audit covers the repository itself, the agent's configuration, and real task logs. It looks at whether the project can start up cleanly, whether test commands are clear, whether the agent has permission limits, and whether those tests actually run after changes are made.
When it's done, Better Harness spits out a list of the places where an agent is most likely to screw up. Each issue comes with evidence, impact, fix scope, and a way to re-verify. You can just have the agent fix the problem and re-run the check.
Better Harness already works with Qoder, Claude Code, Codex, Cursor, and Qwen Code — though not all platforms have full feature parity yet. Qoder itself currently has the most complete support.
Qoder also ran the tool on Better Harness's own codebase. It scored a 58 out of 100. That should give you an idea how early this thing is — the team is still patching issues fast since open-sourcing.