Build Great Loop — Testing skill for Claude Code
A composable Loop Engineering skill group for Codex and Claude Code - interview your task and emit a top-tier, paste-ready agentic loop (harness).
How to install Build Great Loop
This entry records only its repository, not the path inside it, so there is no
exact command to give. Open VioletScar-Hui/Build_Great_Loop and copy the folder into
~/.claude/skills/, or the file into ~/.claude/agents/.
What Build Great Loop does
A composable Loop Engineering skill group for Codex and Claude Code - interview your task and emit a top-tier, paste-ready agentic loop (harness). 4 skills: spec, engineering, eval, review.
Alternatives in Testing
- Grilling.V1 — Interview the user to clarify or stress-test the request before acting 272 ★
- Superpowers Evals — Behavioral eval lab (Quorum) for the superpowers project that drives real coding-agent CLIs (Claude, Codex, Ge 107 ★
- CSV Mock Generator — Generate mock CSV financial data for testing the agentic finance review workflow 90 ★
README
Build Great Loop · Loop Engineering Skills
[中文](README.md) | [English](README.en.md)
         
一个**可组合的「循环工程(Loop Engineering)」技能组**,同时支持 Codex 和 Claude Code。 当你需要让 AI **自己跑、跑到完成、中途被打断也能续上、还不假装做完**时,触发它——它会 访谈你的任务,然后产出一份**可直接粘贴的顶级循环提示词(harness)**。
模型是发动机,循环是发动机外面的整个底盘:它要朝目标走的方向、它记得住的状态、让它 诚实的验证、以及告诉它何时该停的规矩。这套技能把「老手才会下意识做的事」变成提示词里 的默认动作。

这是什么
`Build Great Loop` 不是一个单体技能,而是 **6 个各司其职、可独立触发、也能串起来用的正式 loop skill**,外加一个负责把经验沉淀为可复用资产的 `skill-craft` 伴侣:
| 技能 | 当前版本 | 作用 | 什么时候触发 |
|---|---|---|---|
| loop-spec | 4.1.0 | 访谈式把模糊任务变成一份可执行的 SPEC.md |
任务还没想清、需要先定目标/成功标准/停止条件 |
| loop-engineering(核心) | 7.1.0 | 访谈 → 产出可粘贴的循环提示词本体 | 「帮我写个能一直跑到完成的 agent / 循环 / harness」 |
| loop-eval | 3.1.0 | 设计成功标准 + 小评测集 + 评分器(pass@k vs pass^k) | 「怎么知道这循环靠不靠谱 / 给 agent 写评测」 |
| loop-review | 4.1.0 | 体检并加固一份已有的循环提示词 | 「我的 agent 老是停不下来 / 说完成了其实没有 / 重启丢进度」 |
| loop-ops | 3.1.0 | 把循环跑成周期性/无人值守的自动化(调度·成本·安全门·分级放权) | 「每天/每个 PR/每晚跑这个 agent」「无人值守怎么跑不出事」「babysit 我的 PR」「自动 triage issue」 |
| loop-retro | 4.1.0 | 跑完复盘:证据引证诊断 → harness 修订提案 + 真实失败 gotcha + 标准提案 + 沉淀判定 | 「loop 跑完了帮我复盘」「看看 run log 哪里出问题」「为什么超预算/thrash」 |
Related Skills
Reviewer Tester
You are reviewing one focused group of changed test files from a pull request for an automated pipeline. Your
Harness Agentic AI Agent Best Practices And Use Case
Production-ready AI agent for UI/web testing on Amazon Bedrock AgentCore Harness — Memory, Skills, observabili
Research OS
Research-OS is an orchestration system built on top of Claude Code that combines rigorous research planning wi
Smart PR
Generate a structured PR description from the current branch diff. Run after /self-review passes. Produces tit
Hookwright Form Builder
Claude skill that turns a plain-language form description into a Deftmind Hookwright field spec (JSON) for bui
Nereus
An opinionated development harness for Claude Code — interview → spec → TDD build → parallel review → finish,
Related Agents
Bookie
Sole authority for arc-agents ledger WRITES (create, update, decompose, event, hitl emit, hygiene-emit). Worke
Cr Verifier
Adversarial verifier of the agentic code review — receives candidate findings and tries to refute them, by exe
PR Screenshotter
Produces real before/after PNG screenshots for a YAS branch's rendering changes and publishes them to the yas-