eval-harness

Name: eval-harness
Availability: InStock
Author: bereniketech

Define and run eval-driven development criteria before coding.

Updated Aug 27, 2026

Search Tags:#testing #metrics #evaluation #quality-assurance #development-process #eval

Ready to run

npx skills add https://github.com/bereniketech/claude_kit --skill eval-harness-bereniketech

💬 Prompt for ChatGPT / Claude Chat▼

Please help me install this Agent Skill.
Skill: eval-harness
Source: https://github.com/bereniketech/claude_kit/tree/main/skills/core/eval-harness
Command: npx skills add https://github.com/bereniketech/claude_kit --skill eval-harness-bereniketech

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

Defines a disciplined approach to development by requiring explicit eval definitions before implementation, turning tests into the evaluation framework that guides AI-assisted coding and guarantees measurable reliability.

Core Features & Use Cases

Eval-first development: write capability and regression eval definitions prior to coding to establish clear success criteria.
Structured evaluation suite: support capability evals, regression evals, and pass@k metrics to quantify progress and stability.
Versioned artifacts and governance: organize eval definitions, baselines, and review workflow under a repeatable process for auditable software delivery.

Quick Start

Create a new eval definition at .claude/evals/<feature>.md and implement the feature to satisfy the defined criteria.

Frequently Asked Questions about eval-harness

High-intent search queries and answers about installing and using this skill.

FAQPage Schema

How do I implement eval-driven development for AI-assisted coding?▼

Implement eval-driven development by defining explicit capability and regression eval criteria before coding. This approach establishes measurable success metrics, ensuring AI-assisted software projects ship with quantifiable reliability and stability.

What is the best way to measure AI agent capability and regression metrics?▼

Measuring AI agent capability and regression metrics requires a structured evaluation suite capturing pass@k metrics. Defining these eval baselines prior to feature implementation quantifies progress and captures stability across agent changes.

How do I set up versioned evaluation artifacts for software traceability?▼

Set up versioned evaluation artifacts by organizing eval definitions and baselines under the .claude/evals directory. This creates an auditable, repeatable governance process for tracking software delivery changes over time.

Does eval-first development work without external testing dependencies?▼

Yes, eval-first development works without external dependencies by using a frontmatter-based skill entrypoint. It transforms tests into a structured evaluation framework to guide AI-assisted coding directly within your project.

When do I need to define capability evals before writing code?▼

Define capability evals before writing code when you need clear, measurable success criteria for AI-assisted projects. This front-loaded evaluation approach guarantees reliability and prevents regressions in agent capabilities.