eval-harness

Automate Claude Code session evaluations with capability and regression tests.

5|2|Updated Feb 18, 2026

Search Tags:#automation #testing #evaluation #governance #code-quality #claude-code #eval-driven

Ready to run

npx skills add https://github.com/T-rav/hydraflow --skill eval-harness-t-rav

💬 Prompt for ChatGPT / Claude Chat▼

Please help me install this Agent Skill.
Skill: eval-harness
Source: https://github.com/T-rav/hydraflow/tree/main/.codex/skills/eval-harness
Command: npx skills add https://github.com/T-rav/hydraflow --skill eval-harness-t-rav

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

Formal evaluation of Claude Code sessions using eval-driven development (EDD) principles, providing clear criteria, repeatable tests, and auditable reports to improve model reliability.

Core Features & Use Cases

Define capability evaluations to describe expected behavior and success criteria.
Define regression evaluations to ensure existing functionality remains intact across changes.
Generate standardized evaluation reports that enable governance and human review when needed.

Quick Start

Define evaluation criteria, run the evaluation workflow, and review the resulting report.

Frequently Asked Questions about eval-harness

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

FAQPage Schema

How do I automate evaluation of Claude Code sessions?▼

Automating Claude Code session evaluation involves defining capability and regression tests during design, implementation, and QA phases. The workflow runs deterministic code-based graders, model graders, and human-review processes, storing definitions and per-run results under .claude/evals.

What is eval-driven development for AI coding assistants?▼

Eval-driven development establishes formal criteria, repeatable tests, and auditable reports to improve model reliability in AI coding sessions. It defines expected behaviors and regression checks to ensure existing functionality remains intact across changes.

Can I use model graders and human-review workflows together for code quality evaluation?▼

Yes, code quality evaluation supports deterministic code-based graders, model graders, and human-review workflows simultaneously. You can automate baseline regression checks while routing complex capability evaluations to human reviewers for governance.

How do I set up regression tests for Claude Code workflows?▼

Define regression evaluations during the design and QA phases to ensure existing functionality remains intact across changes. These evaluation definitions are stored under .claude/evals and executed with deterministic or model graders to produce per-run results.

Does Claude Code evaluation support governance and human review reporting?▼

Claude Code evaluation generates standardized reports that enable governance and human review workflows when needed. These reports document capability and regression test results, providing auditable records for compliance and oversight.

What is the best way to define capability evaluations for AI code generation?▼

Define capability evaluations by describing expected behavior and success criteria during the design phase. This establishes clear testing parameters that can be automatically graded using deterministic or model graders and reviewed in standardized reports.