Help us improve
Share bugs, ideas, or general feedback.
Share bugs, ideas, or general feedback.
Share bugs, ideas, or general feedback.
By zxdxjtu
Automates spec-driven development (SDD) workflows: from spec generation and test creation (unit/integration/e2e) to task decomposition, auto-loop sprint execution, adversarial code reviews, baseline capture with Playwright, and code cleanup — enabling AI teams to iterate from specification to verified delivery.
npx claudepluginhub zxdxjtu/harness --plugin harness对抗式代码审查 — 独立 Agent 挑战实现质量
Phase 3: merge spec/design deltas, code review, atomic commits, push, and trigger self-evolution
Capture baseline of a reference product using Playwright MCP for clone/replicate scenarios
Cancel an active sprint loop
Decompose tests into atomic task DAG with wave-based parallel execution and module stewardship
Executes bash commands
Hook triggers when Bash tool is used
Modifies files
Hook triggers on file write and edit operations
Share bugs, ideas, or general feedback.
Own this plugin?
Verify ownership to unlock analytics, metadata editing, and a verified badge.
Sign in to claimOwn this plugin?
Verify ownership to unlock analytics, metadata editing, and a verified badge.
Sign in to claimBased on adoption, maintenance, documentation, and repository signals. Not a security audit or endorsement.
Spec-Driven Development Workflow for Claude Code. A 6-step pipeline: Requirements → Code Analysis → Design → Implement → Verify → Self-Improve. Specifications are the source of truth, code is a verified artifact.
Spec-driven development methodology for full-stack teams
Spec-Driven Development (SDD) plugin for Claude Code providing a comprehensive workflow for building features with requirements, design, tasks, and implementation phases.
Specification-first AI harness: 11 structural gates, 11 Ouroboros commands, 11 agent personas, and 3-tier architecture enforcement. v2.1: Pair Mode (Navigator-Driver + independent test design + /review command).
Spec-driven development workflow - transforms ideas into structured specifications (requirements, design, tasks) before implementation.
HarnessFlow — From idea to shipped product: high-quality engineering workflows for AI agents. Spec-anchored SDD, gated TDD, evidence-based routing, independent reviews, and formal closeout.
一个面向编码 Agent 的 Spec-Driven Development 工作流包。它保留 Claude Code 的原生插件体验,同时补齐 Codex 原生插件清单,并通过 agent-compatible skills 目录兼容 OpenCode。
| Tool | Native packaging | What works |
|---|---|---|
| Claude Code | .claude-plugin/ | Native marketplace/plugin install, commands, hooks |
| Codex | .codex-plugin/ | Native plugin bundle with skills |
| OpenCode | .agents/skills/ | Repo-local or global skill discovery |
# 添加 marketplace 并安装
/plugin marketplace add zxdxjtu/harness
/plugin install harness@harness
git clone https://github.com/zxdxjtu/harness ~/.codex/plugins/harness
然后按你所用 Codex 版本的本地 plugin 流程启用它。Codex manifest 位于 .codex-plugin/plugin.json。
OpenCode 不使用 Claude/Codex 的 plugin manifest 体系。它主要通过 .agents/skills/、.claude/skills/ 等 agent-compatible 技能目录发现工作流。
因此可以直接把这个仓库作为项目级 skill pack 使用,或将 skill 目录软链/拷贝到你的全局 OpenCode skills 路径。
/harness 实现用户登录功能,支持邮箱密码和 GitHub OAuth
就这一行。Harness 自动完成:
了解你的项目 → 讨论需求 → 生成测试 → 分解任务 → 并行执行 → 验证交付 → 从失败中学习
↑ ↑
你确认 你确认 其余全自动
也可以显式使用核心命令:
/proposal "Add user authentication with JWT"
/harness-status
Codex 和 OpenCode 目前不完全复刻 Claude Code 的 slash command 面板。它们更适合直接消费同一套 Harness skills,例如:
proposaltdd-aligndecomposesprintevaluateverifyharness-status| 命令 | 作用 |
|---|---|
/harness 功能描述 | 一站式:从想法到交付 |
/harness | 自动判断当前状态,接着干 |
/harness status | 查看全景 |
/cancel-sprint | 紧急停止 |
四个命令覆盖日常使用。高级子命令见 /help。
读 lockfile、CI 配置、CONTRIBUTING.md 等自动发现项目规范和团队约束。你只需确认分析结果。
Generator 和 Evaluator 分离,防止自评偏差。评估维度在 Spec 中按功能性质定制。
TRIVIAL: proposal -> implement -> verify
SMALL: proposal -> tdd-align -> implement -> verify
MEDIUM: proposal -> tdd-align -> decompose -> sprint -> evaluate -> verify -> archive
LARGE: proposal -> spec-review -> tdd-align -> decompose -> sprint -> evaluate -> eval-fix -> verify -> archive
Clone 场景额外: baseline 采集参考产品基线
| 来源 | 借鉴 |
|---|---|
| Anthropic — Harness Design | Generator 与 Evaluator 对抗分离 |
| Anthropic — Effective Harnesses | Progress file + git 状态恢复 |
| OpenAI — Harness Engineering | 机械约束 > 指令遵从 |
| OpenAI — Eval Skills | JSONL 追踪 + 多维评估 |
| EvoMap/evolver | 后台自动进化 daemon loop |
| wow-harness | 失败模式不变量化 + 文件编辑时上下文路由 |
| get-shit-done | Wave 级任务并行 + 目标反向验证 |
| DSPy GEPA | 反思式 prompt 进化 |
/harness 编排一切,用户不需要记子命令.harness/
├── config.json
├── full-spec.md
├── full-design.md
├── specs/
├── designs/
├── reviews/
├── tasks.md
├── progress.md
├── module-graph.json
├── baseline/
├── evidence/FXXX/
└── evolution/
MIT