Implement evaluation-driven development workflows for Claude Code sessions.
npx skills add https://github.com/shygoly/sapbase --skill eval-harness-shygoly
Please help me install this Agent Skill. Skill: eval-harness Source: https://github.com/shygoly/sapbase/tree/main/docs/zh-CN/skills/eval-harness Command: npx skills add https://github.com/shygoly/sapbase --skill eval-harness-shygoly
This Skill provides a formal evaluation framework for Claude Code sessions, implementing evaluation-driven development (EDD) principles to treat evaluation as the unit test for AI development, ensuring early success criteria, continual checks, and traceable changes.
Define your evaluation plan, implement the evaluators, run tests, and review results.
High-intent search queries and answers about installing and using this skill.
Evaluation-driven development (EDD) treats evaluation as the unit test for AI development, ensuring early success criteria, continual checks, and traceable changes within Claude Code sessions.
Define an evaluation plan to validate new capabilities or guard existing functionality, implement code-based, model-based, or human graders, and store them in a structured repo under .claude/evals.
Yes, the framework supports pass@k metrics alongside multiple grading methods including code-based, model-based, and human scoring to track evaluation success rates accurately.
Use a structured repo layout under .claude/evals with clear guidelines for planning, running, and reviewing assessments to maintain evaluation baselines and track capability regressions.
Yes, you can apply code-based, model-based, and human scoring graders simultaneously within the evaluation workflow to comprehensively assess regression and capability test results.
You need formal capability evaluation when defining and validating new AI capabilities before coding, ensuring early success criteria are met and existing functionality is guarded against regressions.
andrew-starosciakEvaluate Claude Code sessions with pass@k scoring and structured reports.
mbadozAutomate formal evaluation of Claude Code sessions with pass@k metrics.
rongaredeEvaluate Claude Code sessions with capability and regression evals.
Fabio29TImplement formal evaluation frameworks for AI agents with pass@k metrics.
