better-harness
Help your coding agents (Claude Code, Codex, Qoder, Cursor, and other coding agents) get better at getting better.
What is it?
What it is
A tool that reviews AI coding workflows by analyzing five dimensions of the Agent Work Loop (Task Understanding, Controlled Execution, Change Validation, Reliable Delivery, Learning Capture) to identify gaps and provide actionable findings based on evidence.
Why it exists
To address weak points in AI coding workflows by making findings explicit and tied to evidence, helping teams improve their processes through actionable insights.
Who should use it
Software engineers using AI coding agentsDevOps engineers looking to improve AI delivery safeguardsEngineering managers monitoring AI coding workflow qualityTeams using Claude Code, Cursor, GitHub Copilot, or QoderAI agent developers seeking workflow evaluation metrics
Who should avoid it
Users not using supported AI coding agentsDevelopers looking for a simple code completion tool rather than a workflow reviewerUsers without Node.js environment for local development
How it works
A quick walkthrough in plain English
How better-harness works
Step 1 of 3
You interact with it
Open better-harness, send a request, or connect it to your stack.
Features
Advantages
- Transparency in workflow analysis through visible evidence
- Structured evaluation of critical agent workflow dimensions
- Multi-agent support with standardized reporting
- Open-source flexibility for customization and extension
- Community-driven improvements via contribution model
- Focus on systemic improvements over superficial fixes
Disadvantages
- Requires specific coding agent integrations for full functionality
- Setup complexity due to multiple host dependencies
- Learning curve for understanding Agent Work Loop framework
- Evidence coverage may vary between agents
- May require additional tooling for optimal use
Installation
FAQ
What does Better Harness actually do?
Better Harness reviews your AI coding workflow by evaluating how agents understand tasks, execute changes, validate results, deliver code, and learn. It identifies gaps in these processes and turns them into prioritized findings with evidence-backed repair plans.
Which AI coding agents are supported?
It supports a wide range of hosts including Claude Code, Codex (Desktop and CLI), Qoder (Desktop and CLI), Cursor, Qwen Code, GitHub Copilot CLI, and Pi.
How do I run a review in my coding agent?
Once the plugin is installed, you can trigger a review by using the slash command: `/better-harness review this project's AI coding workflow and generate a report`.
What kind of output does the tool provide?
Depending on the host, it produces a self-contained HTML report, a Markdown report, or a Canvas report (for Qoder). These reports include prioritized findings, impact assessments, and scoped repair actions.
Does the tool infer behavior if evidence is missing?
No. Better Harness is designed to be deliberately honest; if behavior cannot be observed through session transcripts or project mechanisms, the report explicitly marks it as missing or partial rather than making unsupported claims.
Featured in Videos
YouTube tutorials and walkthroughs for better-harness
Alternatives
Similar projects ranked by category, topics, and text overlap.
