paladini/harness-score

Your AI coding agent is only as reliable as the harness around it. Measure that harness in seconds with harness-score.

What it solves

AI coding agents (like Cursor, Claude Code, or Devin) are often unreliable because they lack a structured "harness"—the set of rules, guides, and guardrails that prevent them from making mistakes or rediscovering project conventions every session. Harness Score provides a deterministic way to measure the maturity of this infrastructure, allowing teams to identify gaps in their agent's control system and track improvement over time.

How it works

The tool is a deterministic filesystem scanner that analyzes a repository for 36 concrete checks across six dimensions: Context & Guides, Skills & Commands, Hooks & Guardrails, Sensors & Feedback, CI Feedback, and Hygiene & Safety. It does not use LLMs or network requests, ensuring the same result every time. Based on these checks, it assigns a maturity level from L0 (Unharnessed) to L4 (Self-correcting) and provides a ranked list of specific files or patterns that are missing or need fixing.

Who it’s for

Developers and teams using AI coding agents who want to move beyond "vibe-coding" and build a reliable, production-ready environment for autonomous AI development.

Highlights

  • Deterministic Scoring: Zero LLM calls and zero network access for consistent, repeatable results.
  • Maturity Ladder: A 5-level system (L0–L4) that gates progress based on the shape and coverage of the harness.
  • CI Integration: Ability to gate pipelines using the --min-level flag to prevent maturity regression.
  • Actionable Diagnosis: Every failed check links to a specific recipe for remediation.
  • Multi-Tool Support: Works across various AI coding tools including Cursor, Claude Code, Windsurf, Cline, and Continue.

Related

  • Project
    scanaislop/aislop

    aislop is a deterministic CLI that scans code for AI‑generated “slop” (unused imports, narrative comments, dead code, etc.) across 10 languages, scores the repo 0‑100, can auto‑fix safe issues, and integrates with AI agents, git hooks, and CI.

  • Project
    awslabs/aidlc-workflows

    AI-DLC is a structured, harness-neutral framework that turns AI coding assistants into verifiable, audited software delivery workflows with human approval gates and multi-stage lifecycle management.

  • Project
    mgechev/skillgrade

    Skillgrade is a Node.js‑based evaluation harness that lets you define coding tasks, run AI coding agents (Gemini, Claude, Codex, OpenCode, or any ACP‑compatible CLI), and score the outcomes with deterministic scripts, LLM rubrics, or human reviewers. It supports multi‑turn tasks, dataset‑driven expansion, filtering, and CI‑friendly reporting in JSON, JUnit, CSV, or Markdown.

  • Project
    a5c-ai/babysitter

    A deterministic orchestration framework that enforces strict, code-defined workflows for AI agents to prevent hallucinations and ensure reliability.

  • Project
    QoderAI/better-harness

    Better Harness is an open‑source tool that analyses the whole workflow of AI coding agents (Claude Code, Codex, Qoder, Cursor, GitHub Copilot, etc.). It gathers project and session evidence, evaluates five dimensions of the *Agent Work Loop* (task understanding, controlled execution, change validation, reliable delivery, learning capture), and produces a visual HTML report with prioritized, evidence‑backed findings and repair suggestions. Missing evidence is shown explicitly, helping teams improve repeatability, validation, and learning while still benefiting from AI‑generated code.