https://github.com/easyinplay/harnessed
AI coding harness composition orchestrator — manifest-described upstreams, composition skill workflows. Apache-2.0.
https://github.com/easyinplay/harnessed
ai claude claude-code claude-code-plugin claude-skills harness-engineering workflow
Last synced: 3 days ago
JSON representation
AI coding harness composition orchestrator — manifest-described upstreams, composition skill workflows. Apache-2.0.
- Host: GitHub
- URL: https://github.com/easyinplay/harnessed
- Owner: easyinplay
- License: apache-2.0
- Created: 2026-05-11T17:07:49.000Z (about 2 months ago)
- Default Branch: main
- Last Pushed: 2026-06-24T00:59:15.000Z (8 days ago)
- Last Synced: 2026-06-24T02:24:41.019Z (8 days ago)
- Topics: ai, claude, claude-code, claude-code-plugin, claude-skills, harness-engineering, workflow
- Language: TypeScript
- Homepage: https://harnessed.cc/
- Size: 7.06 MB
- Stars: 2
- Watchers: 0
- Forks: 1
- Open Issues: 0
-
Metadata Files:
- Readme: README-cn.md
- Changelog: CHANGELOG.md
- Contributing: CONTRIBUTING.md
- Funding: .github/FUNDING.yml
- License: LICENSE
- Security: SECURITY.md
- Notice: NOTICE
Awesome Lists containing this project
- awesome-claude-skills - easyinplay/harnessed
README
# harnessed
[English](./README.md) | **简体中文** | [繁體中文](./README-tw.md) | [日本語](./README-ja.md) | [한국어](./README-ko.md) | [Português (Brasil)](./README-pt-BR.md) | [Türkçe](./README-tr.md) | [Русский](./README-ru.md) | [Tiếng Việt](./README-vi.md) | [ไทย](./README-th.md)
> AI coding harness 包管理器 + composition orchestrator
> 把三层栈协作方法论 (gstack 决策 + GSD 项目经理 + superpowers 资深工程师 + karpathy 心法 + mattpocock 招式) 编排化为可执行 engine
[](https://npmjs.com/package/harnessed)
[](./LICENSE)
[](https://github.com/sponsors/easyinplay)
> Not affiliated with, endorsed by, or sponsored by Harness Inc. (见 [NOTICE](./NOTICE))
---
## ✨ 一句话定位
**把原始的 Claude Code 变成一支纪律严明的资深工程团队。** 一次安装,就把治理 (governance)、规划 (planning)、TDD、审查 (review) 织进一条 Discuss→Ship 工作流 —— 进度和证据落在磁盘上,不消散在对话里。
> _AI coding harness 包管理器 + composition orchestrator_ —— 装配市面上最优秀的开源 Claude Code 生态组件,用强意见 composition skill 织成统一工作流;不 vendor 上游代码,通过 manifest 描述 install/check + composition skill 编排多上游协作。
---
> 等等 — harnessed 真能跟 superpowers / gstack / GSD 这种上游 PK?
> 当然 —— 我们**站在巨人的肩膀上**,看得更远嘛。(牛顿同款 quote,没收版权费就先用着了 🧐)
> ……(小声) 不过仔细看更像肩上那只鹦鹉。
> 算了,鹦鹉学舌,我们至少**会编排**。🦜
---
## 🎯 关键差异化
- **三层栈机器化** — `gstack 决策` + `GSD 项目经理` + `superpowers 资深工程师` + `karpathy 4 心法` + `mattpocock 23 招式`,5 支柱 100% capture
- **不 vendor 上游** — manifest describe install/check;上游升级用户 re-install 即获最新版
- **Composition Skill** — 自家 workflow skill 当指挥棒,调度多个上游协同。**1 super-master `/auto` + 4 stage master + 18 sub-workflow + 2 standalone = 25 namespace-layered workflow**,完整 4-stage 机器化 (`/auto` 跨 stage 一键 / `/discuss /plan /task /verify` 单 stage / 三层栈 18 sub / `/research /retro` 2 standalone)
- **L0 Discipline Substrate** — 全局 cross-stage 行为基准 (karpathy 心法 + output-style + language + operational + priority + protocols),applied universally
- **包管理器思维** — install dependency graph 自动解析, doctor 健康检查, install-base 一键装齐
- **统一入口** — 用户面对 `/discuss /plan /task /verify` 等 master slash command,不需学每家上游术语;sub command 显式调用单 stage (例如 `/discuss-strategic` 只跑战略层澄清)
---
## 📦 快速安装
```bash
npm install -g harnessed && harnessed setup
```
> Windows PowerShell 5.x 不支持 `&&` 链接,需改 `;` 或分两行 (`npm install -g harnessed; harnessed setup`)。bash / zsh / PowerShell 7+ / cmd.exe 都正常。
**卸载 harnessed：**
```bash
harnessed uninstall # 删除 harnessed 自身文件（上游组件不受影响）
```
> `harnessed uninstall` 会清理 commands、workflow skills、settings 环境变量和状态目录。上游组件（npm 包、MCP 服务器、CC 插件、git clone 仓库、npx skills）保持完整。运行 `harnessed uninstall ` 可单独移除某个上游。加 `--dry-run` 可预览。
🤖 **或让 AI 帮你装** — 把下面这句话发给 Claude Code (或任何 AI 助手):
> 按 `https://github.com/easyinplay/harnessed/blob/main/INSTALL-WITH-AI.md` 的指导帮我装 harnessed
AI 会自动 fetch 文档 + 跑安装,处理 OS / 权限 / PATH / corepack 等 edge case,无需复制大段文字。
> [!TIP]
> 🚀 **很多人关心的 Agent Teams 和 Subagent 功能,在 harnessed 中会根据任务自动启用!**
> 无需手动配置 `CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS`,`harnessed setup` v3.3.1+ 会自动写入 `~/.claude/settings.json`;Pattern A 全栈三路 / Pattern C 4-specialist 等 multi-agent workflow 即开即用。
---
## 🚀 快捷使用 — 3 种选择
按介入程度由低到高:
### 🎯 整体自动 (推荐新手 / 不想动脑子)
```
/auto "需求 X"
# 大需求可显式分阶段 (一般不需要如下指定 AI 自动判断进入;若自认大需求可强制):
/auto "需求 X" --staged
```
> 不想动脑子或者刚入门 — 一切交给 harnessed。自动跑完 6 stage (research conditional → discuss → plan → task → verify → retro mandatory),中间不停。AI 1-shot 自动判断需求复杂度,大需求建议切 `--staged` 模式 (每 stage 完停 review);开始前 prompt "对需求清晰认知吗?" n → 自动加跑 `/research` 多源调研;末尾 `/retro` 强制总结。失败 fail-fast,`harnessed resume` 续。
### 📂 分类自动 (推荐熟手 / 想 review 中间结果)
```
/discuss "需求 X" # 战略 + Phase + 子任务 3 层澄清
/plan "需求 X" # 架构 (conditional) + 计划持久化
/task "subtask-1" # 4 sub 串行 (clarify → code → test → deliver)
/verify "phase-1" # 7 sub conditional 验证
```
> 想自己决定从哪个 stage 开始 / review 中间产出 — 4 个 master 独立调用,每个 master 内部仍自动 fan-out 该 stage 所有 sub。
### 🔬 精确调用 (大神 mode / 知道自己要什么)
```
/discuss-phase "..." # 单跑 Phase 层澄清
/plan-architecture "..." # 单跑架构审查
/verify-paranoid "..." # 单跑 Paranoid Staff Engineer 审查
# ... 其他 18 个 sub-workflow 任选
```
> "我是大神,我自己决定" — 跳过 master,直接调某 sub-workflow。适合已知精确需要哪个 sub 的高级用户 / 复用某单一环节。
---
## 📐 4-stage 流程图
```mermaid
graph TD
RS([⓪ /research — 前置多源调研 可选]):::optional
subgraph Discuss[① Discuss 战略澄清]
DM[/discuss master/]
DS[discuss-strategic]
DP[discuss-phase]
DT[discuss-subtask]
DM --> DS & DP & DT
end
subgraph Plan[② Plan 任务规划]
PM[/plan master/]
PA[plan-architecture]
PP[plan-phase]
PM --> PA & PP
end
subgraph Task[③ Task 执行]
TM[/task master/]
TC[task-clarify]
TCo[task-code]
TT[task-test]
TD[task-deliver]
TM --> TC --> TCo --> TT --> TD
end
subgraph Verify[④ Verify 验证]
VMs[/verify master/]
VP[verify-progress]
VC[verify-code-review]
VPa[verify-paranoid]
VQ[verify-qa]
VS[verify-security]
VD[verify-design]
VSi[verify-simplify]
VM[verify-multispec]
VMs --> VP & VC & VPa & VQ & VS & VD & VSi & VM
end
RT([⑤ /retro — 里程碑总结 可选]):::optional
RS --> Discuss
Discuss --> Plan --> Task --> Verify
Verify --> RT
classDef optional stroke-dasharray:5 5,fill:#f5f5f5,color:#666
```
> 虚框 = 可选 standalone (`/research` 战略前调研 / `/retro` 里程碑后总结);实框 = 主流程 4-stage cadence。
### 25 workflow 总览表
| Slash cmd | Stage | Type | Capability / Upstream | Brief |
|-----------|-------|------|----------------------|-------|
| `/auto` | All | **Super-master** | masterOrchestrator (跨 6 stage) | 一键自动跑 6 stage (research conditional → discuss → plan → task → verify → retro mandatory); AI 复杂度 1-shot judge + 理解度 check + retro mandatory; `--staged` opt-in stage gate |
| `/discuss` | ① Discuss | Master | masterOrchestrator | 3 sub 并行 gate-eval (chain-isolation 铁律) |
| `/discuss-strategic` | ① Discuss | Sub | gstack `/office-hours` + `/plan-ceo-review` + planning-with-files | 战略层 — 新功能 / 新 milestone / 产品方向强制治理 (findings.md 持久化) |
| `/discuss-phase` | ① Discuss | Sub | GSD `/gsd-discuss-phase` + planning-with-files | Phase 层 — ≥2 open decisions / 灰色地带澄清 (findings.md + knowledge.md 持久化) |
| `/discuss-subtask` | ① Discuss | Sub | superpowers brainstorming + `/grill-with-docs` | 子任务层 — ≥2 approach / 核心算法 / API contract (ephemeral 短讨论, 不持久化) |
| `/plan` | ② Plan | Master | masterOrchestrator | 串行 invoke 2 sub (architecture conditional → phase always) |
| `/plan-architecture` | ② Plan | Sub | gstack `/plan-eng-review` | 架构层 — 复杂架构强制治理关卡 |
| `/plan-phase` | ② Plan | Sub | GSD `/gsd-plan-phase` + planning-with-files `/plan` | 计划层 — 持久化 `task_plan.md` + `progress.md` |
| `/task` | ③ Task | Master | masterOrchestrator | 串行 invoke 4 sub per subtask (clarify → code → test → deliver) |
| `/task-clarify` | ③ Task | Sub | superpowers brainstorming + `/grill-with-docs` conditional | 子任务起步澄清 gate |
| `/task-code` | ③ Task | Sub | karpathy 4 心法 + `/zoom-out` / `/improve-codebase-architecture` / `/diagnose` conditional | 子任务编码 + 跨 session progress.md 同步 |
| `/task-test` | ③ Task | Sub | superpowers TDD red-green-refactor + `/diagnose` conditional | 核心逻辑 TDD 强制 (alias mattpocock `/tdd`) |
| `/task-deliver` | ③ Task | Sub | `ralph-loop` SDK wrapper + Agent Teams conditional | 至 verbatim `COMPLETE` + R20.10 max_iter fallback |
| `/verify` | ④ Verify | Master | masterOrchestrator | 7 sub 按场景 conditional dispatch |
| `/verify-progress` | ④ Verify | Sub | GSD `/gsd-verify-work` + `/gsd-progress` | 必跑串行起点 — UAT 验收 + 状态同步 |
| `/verify-code-review` | ④ Verify | Sub | `code-review` 多 subagent fan-out | 高置信度 finding 并行 |
| `/verify-paranoid` | ④ Verify | Sub | gstack `/review` (Paranoid Staff Engineer) | 关键模块 PR 前强制 |
| `/verify-qa` | ④ Verify | Sub | gstack `/qa` + playwright-cli / `@playwright/test` / webapp-testing | 端到端 QA (has_ui_changes conditional) |
| `/verify-security` | ④ Verify | Sub | gstack `/cso` | OWASP / auth / secrets (has_auth_or_secrets conditional) |
| `/verify-design` | ④ Verify | Sub | gstack `/design-review` + ui-ux-pro-max + frontend-design | 设计系统一致性 (has_design_changes conditional) |
| `/verify-simplify` | ④ Verify | Sub | `code-simplifier` | 末尾串行简化 |
| `/verify-multispec` | ④ Verify | Sub | 4-specialist Agent Team Pattern C | 关键发布 / 大重构 PR 升级 (互相 SendMessage 质询) |
| `/research` | Standalone | Standalone | Tavily / Exa MCP + ctx7 + GSD `/gsd-discuss-phase` | 多源调研 (Stage ① alternate) |
| `/retro` | Standalone | Standalone | gstack `/retro` + planning-with-files RETROSPECTIVE.md | 项目 / 里程碑结束总结 |
> Master orchestrator 自动 gate-route 到合适的 sub (chain-isolation 铁律 — 不 fire 的 sub 透明声明跳过)。
> 直接调用 sub 也可绕过 master 单跑某 stage,例如 `/discuss-strategic "新功能 X"`。
---
## ⚡ 使用流程
4-stage 三层栈方法论 — 推荐 4 个 master orchestrator 串行驱动:
```
/discuss → /plan → /task → /verify
① ② ③ ④
```
| Stage | Master | 主要 sub-workflow | 上游协同 |
| ---- | ---- | ---- | ---- |
| ① **Discuss** | `/discuss` | strategic / phase / subtask (3 并行) | gstack `/office-hours` + GSD `/gsd-discuss-phase` + superpowers brainstorming |
| ② **Plan** | `/plan` | architecture (conditional) → phase | gstack `/plan-eng-review` + GSD `/gsd-plan-phase` + planning-with-files |
| ③ **Task** | `/task` | clarify → code → test → deliver (4 串行 per subtask) | karpathy 心法 + mattpocock 招式 + superpowers TDD + `ralph-loop` |
| ④ **Verify** | `/verify` | progress → 5 parallel conditional → simplify (+ multispec critical) | GSD `/gsd-verify-work` + code-review + gstack `/review` / `/qa` / `/cso` / `/design-review` + code-simplifier |
实操示例:
```bash
# 1. 装 workflow 上游 (一行装齐 gstack + GSD + superpowers + planning-with-files)
harnessed setup
# 2. 在 Claude Code 内跑 4-stage cadence
/discuss "新功能 X" # 战略 + Phase + 子任务 3 层澄清
/plan "新功能 X" # 架构 (conditional) + 计划 (任务图持久化)
/task "subtask-1: API contract" # 4 sub 串行 per subtask
/verify "phase-1" # 7 sub conditional
# 3. 中断后恢复 (任何时候)
harnessed resume
```
> 也可直接调 sub 绕过 master 单跑某一层,例如 `/verify-paranoid` 只跑 Paranoid Staff Engineer 审查。
📊 详细 mermaid + 各 stage 完整说明:[docs/WORKFLOW.md](./docs/WORKFLOW.md)
---
## 🗂️ 架构 (4-stage namespace-layered)
### 1. 目录结构
```
harnessed/
├── manifests/ # L1: 上游描述层 (NOT vendored)
├── workflows/ # L6: composition skill (4-stage 指挥棒)
│ ├── discuss/ # Stage ① 3 layer (strategic + phase + subtask)
│ │ ├── auto/ # /discuss master gate-route
│ │ ├── strategic/ # /discuss-strategic (gstack /office-hours + /plan-ceo-review)
│ │ ├── phase/ # /discuss-phase (GSD /gsd-discuss-phase)
│ │ └── subtask/ # /discuss-subtask (superpowers brainstorming)
│ ├── plan/ # Stage ② (architecture + phase 任务图)
│ ├── task/ # Stage ③ (clarify + code + test + deliver)
│ ├── verify/ # Stage ④ (progress + code-review + paranoid + qa + cso + design + simplify + multispec)
│ ├── research/ # standalone Stage ① alternate
│ ├── retro/ # standalone post-④ milestone close
│ ├── capabilities.yaml # L5a: ~100 entry, 7 category SoT
│ ├── defaults.yaml # ralph_max_iterations per workflow phase
│ ├── judgments/ # L5a: 三层栈判据 + parallelism + tdd + fallback + rules-routing
│ │ ├── strategic-gate.yaml
│ │ ├── phase-gate.yaml
│ │ ├── subtask-gate.yaml
│ │ ├── parallelism-gate.yaml # L5b execution mechanism routing
│ │ ├── tdd-gate.yaml
│ │ ├── fallback.yaml # 3 铁律: skip_with_transparency + override + chain_isolation
│ │ ├── web-design-routing.yaml # UI 设计工具路由
│ │ ├── web-testing-routing.yaml # E2E / 浏览器测试工具路由
│ │ ├── web-search-routing.yaml # 网页搜索 / 文档抓取路由
│ │ └── stage-routing.yaml # master orchestrator sub-stage 路由
│ └── disciplines/ # L0: 全局 cross-stage 行为基准
│ ├── karpathy.yaml # 4 心法 + ≤200L
│ ├── output-style.yaml # BLUF + no-emoji + no-em-dash
│ ├── language.yaml # zh-Hans default + English preserve
│ ├── operational.yaml # biome preempt + A7 + commit safety
│ ├── priority.yaml # skill conflict 仲裁
│ └── protocols.yaml # cc-handoff design doc 自包含
├── routing/ # L4: routing engine SSOT (decision_rules.yaml)
├── schemas/ # L3: JSON Schema (IDE / CI consume)
├── src/ # L4: TS engine (workflow + routing + cli + installers + checkpoint + audit + state)
├── tests/ # vitest unit + integration + dogfood (R8.1 dogfood-first)
├── scripts/ # CI gate (check-workflow-schema, transparency-verdict, state-archive)
├── .planning/ # project memory (STATE + ROADMAP + REQUIREMENTS + per-phase + milestones)
└── docs/adr/ # 架构决策记录
```
### 2. 逻辑分层 (8 层)
```
┌────────────────────────────────────────────────────────────┐
│ L7 User-facing slash cmd + harnessed CLI │
│ /discuss /plan /task /verify (master) + 18 sub + /research /retro + /auto super-master
│ + direct gstack invoke (30+ optional): /office-hours /review /qa /...
├────────────────────────────────────────────────────────────┤
│ L6 Workflow orchestration (workflows///) │
├────────────────────────────────────────────────────────────┤
│ L5b Execution Mechanism (orthogonal): subagent / Agent Teams │
│ / 主 session + ralph-loop wrapper │
│ parallelism-gate.yaml: 默认 subagent → escalate 5 触发 │
│ Pattern A 全栈三路 / B 对立假设 / C 多维度审查 │
├────────────────────────────────────────────────────────────┤
│ L5a Capability + Judgment + Defaults SoT │
│ capabilities.yaml (7 category) + judgments/ (10 file) + │
│ defaults.yaml │
├────────────────────────────────────────────────────────────┤
│ L4 Runtime engine (workflow / routing / handlers) │
├────────────────────────────────────────────────────────────┤
│ L3 TypeBox schema + CI gate │
├────────────────────────────────────────────────────────────┤
│ L2 Installer + Manifest engine │
├────────────────────────────────────────────────────────────┤
│ L1 Upstream components (NOT vendored) │
├────────────────────────────────────────────────────────────┤
│ L0 Discipline Substrate (全局生效) │
│ karpathy 心法 + output-style + language + operational + │
│ priority + protocols (applied universally to L1-L7) │
└────────────────────────────────────────────────────────────┘
```
### 3. Cross-cutting Capabilities (capabilities.yaml 7 category, ~100 entry)
```
behavioral (6): karpathy-guidelines + output-style + language + operational + priority + protocols
tool-slash-cmd (~60): gstack 30+ optional + gsd 10+ + mattpocock 12 高频 + 等
tool-mcp (3): chrome-devtools-mcp / tavily-mcp / exa-mcp
tool-cli (2): ctx7 / gws
tool-plugin (2): planning-with-files / @playwright/test
tool-bundled (3): ralph-loop / webapp-testing / playwright-cli
agent-platform (3): agent-teams-create / send-message / shutdown
```
### 4. 数据流示例 (用户调用 `/discuss "新功能 X"`)
```
[L7] User invokes /discuss "新功能 X"
↓
[L6] workflows/discuss/auto/workflow.yaml master orchestrator
↓
[L5a] judgments.strategic-gate.fires + phase-gate.fires + subtask-gate.fires (3-way 并行 eval)
↓
[L4] judgmentResolver.ts (4-level ref split) + exprBuilder.ts (expr-eval evaluate)
↓
[L0] discipline.priority-hierarchy 仲裁工具冲突 / output-style 格式化输出
↓
[fires=true sub] → invoke sub-workflow (/discuss-strategic / /discuss-phase / /discuss-subtask)
↓ for each sub:
├─ behavioral_layer: karpathy-guidelines (always-on)
├─ tools_available: planning-with-files / ctx7 / mattpocock by-condition
├─ parallelism: judgments.parallelism-gate..fires (L5b mechanism)
└─ phase invocations execute via capability template interpolation
↓
[fallback.yaml chain-isolation] 三层独立判断, 不串行依赖
[Skip 透明声明] 不 fire 的 sub → "⚠️ 跳过 , 因为 "
↓
planning-with-files /plan (cross-cutting tool) → write artifacts to .planning//
↓
[L4] state.ts writeCurrentWorkflow (proper-lockfile) + audit.append (12-field JSONL)
```
### 5. 抉择路由矩阵 (rules-based, codified in judgments + capabilities)
| 场景 | Default → Escalate |
|------|---------------------|
| 并行机制 | subagent → Agent Teams Pattern A/B/C (5 触发) |
| UI 设计主方案 | ui-ux-pro-max → frontend-design (用户明示风格) |
| E2E 浏览器探查 | playwright-cli (Bash 一行 token 省) |
| E2E commit-able TS | @playwright/test 默认 |
| E2E Python 后端联动 | webapp-testing |
| 性能 / a11y / 内存诊断 | chrome-devtools-mcp |
| Web 搜索 (关键词) | Tavily MCP 默认 |
| Web 搜索 (描述式 / 学术) | Exa MCP |
| 库 API 文档 | ctx7 CLI |
| GitHub URL | gh CLI |
| 单 URL 抓取 | WebFetch 内置 |
| Gmail / Drive / Calendar | gws CLI |
| 架构审查 (复杂) | gstack /plan-eng-review |
| TDD 强制 (核心算法) | superpowers TDD OR mattpocock /tdd |
| 关键模块 PR | gstack /review |
| 大重构 PR 多维度审查 | 4-specialist Agent Team Pattern C |
| 跨 session hand-off | discipline.protocols self-contained design doc |
| `/auto` 复杂度大需求 | AI 1-shot judge → 自动建议 `--staged` (n abort 建议手动 `/discuss`) |
| `/auto` 需求理解度 | 开始前 prompt → n 自动加 `/research` 多源调研 |
---
## 🛠️ 维护命令 (Operational)
> 这些是 harnessed 自身维护命令(setup / 健康检查 / 备份回滚 / 状态恢复等),日常 feature 开发用上面的 slash command 即可,这块通常不需要。
**v4.0 — 编排大脑。** slash command 在 Claude Code 主 session 内跑澄清（让问题能触达你），再 spawn CC-native subagent（启用 Agent Teams + 澄清 round-trip）。harnessed 负责 gate 评估（`harnessed gates`）和 spawn-ready prompt（`harnessed prompt`），由主 session 完成 spawn。`harnessed run` 保留供 CI/headless 使用。
### CLI 命令
| 命令 | 说明 |
| ---- | ---- |
| `harnessed setup` | 一次性 setup,装 workflow skills 到 `~/.claude/skills/` + MCP 到 `~/.claude.json` |
| `harnessed gates ` | 评估某 master stage 下哪些 sub-workflow 会 fire（JSON: fire/skip/parallelism）。供 slash command 编排 native spawn |
| `harnessed prompt ` | 输出某 sub-workflow 的 spawn-ready prompt（role + checklist + disciplines + completion/clarification 协议） |
| `harnessed checkpoint ` | 记录 sub-workflow 的 start/complete/fail 到 `~/.claude/harnessed/checkpoints/` |
| `harnessed run ` | 通过 in-process SDK spawn 运行 workflow（CI/headless 模式）。slash command 改用 CC-native spawn |
| `harnessed resume` | session 中断后恢复至最近 checkpoint |
| `harnessed status` | 当前 phase + lock holder |
| `harnessed doctor` | 8-check 健康检查 (Node / MCP / jq / Win bash / 路由 / token budget 等) |
| `harnessed install ` | 装上游 manifest |
| `harnessed uninstall [name]` | 统一卸载 — 不指定名称：删除 harnessed 自身文件（上游完整保留）；指定名称：移除单个上游组件 |
| `harnessed backup` | snapshot 备份管理 |
| `harnessed rollback ` | 一行回滚 (EOL preserve + sha1 verify) |
| `harnessed gc` | 清理过期 backups |
| `harnessed audit-log` | 路由透明日志 query (支持 `--filter` jq 表达式) |
### 参数 (Flags)
> 所有命令默认 **apply (immediate write)**,无需加 flag。高级用户可加 `--dry-run` 预览。
| Flag | 说明 |
| ---- | ---- |
| `--dry-run` | 预览不写盘 (高级用户 opt-in) |
| `--non-interactive` | CI / 脚本场景 |
| `--system` | L4 全局装允许 (否则降级 L1 npx ephemeral) |
| `--full-diff` | 展开 > 200 行的 diff 折叠 |
| `--no-color` | 强制 nocolor (即使 TTY) |
| `--task ` | `run` 子命令 — 任务描述 (传入 workflow `gateContext.task`) |
| `--task-stdin` | `run` 子命令 — 从 stdin 读任务描述直到 EOF (避免 shell 转义引号/$/`) |
---
## ❓ FAQ
Q1. 装了 harnessed 还需要装 superpowers / gstack / GSD 上游吗?
需要,但**用户感知 = 一行命令**:
```bash
harnessed setup # 自动装齐 gstack + GSD + superpowers + planning-with-files,25 workflow skill 一并落到 ~/.claude/skills/ + Agent Teams env var 自动写 ~/.claude.json
```
类比 `brew install ` 装全套依赖 — 你不需要单独 `brew install` 每个依赖项。
Q2. 为什么不直接 vendor superpowers / gstack 进 harnessed 仓库?
4 条理由:
1. **差异化哲学** — harnessed 是「装配主义包管理器」对位「all-in-one 自建派」。vendor = 失去 wedge → 沦为又一个 plugin pack
2. **License + attribution 噩梦** — vendor 4-5 个主动维护的上游 = 复杂 license 拼盘
3. **上游升级反向** — 当前 manifest 描述,上游升级用户 re-install 即得新版;vendor 后手动 sync code 永远落后
4. **Bus factor 1** — 单 maintainer 维护 vendor 4-5 上游 = 加速 burnout
Q3. gstack / GSD / superpowers 看起来都是 plan/discuss 类,是不是重叠?
**不是**。它们是三层栈的不同阶段:
| 阶段 | 上游 | 职责 |
| ---- | ---- | ---- |
| Governance | gstack | 多角色决策关卡 (CEO / EM / Designer / Paranoid Engineer) |
| Brainstorming | superpowers | 子任务设计澄清、方案对比 |
| Orchestration | GSD | 高层 phase 任务图 + 依赖分析 |
| Persistence | planning-with-files | 持久化 `task_plan.md` / `progress.md` / `findings.md` |
`/discuss /plan /task /verify` 4 个 master 把 4 阶段串起来,每个 master 内部再 delegate 到对应 sub。每个阶段做不同事,输出喂给下一阶段。**没有合并**。
Q4. workflow phase 之间是自动跑还是等用户?
看 `workflows//SKILL.md` frontmatter 的 `pause` 字段:
- `pause: human_review` → 阻塞等用户 approve (governance gate / final lock,如 `/discuss-strategic` gstack `/office-hours` + `/plan-architecture` `/plan-eng-review` 锁定关卡)
- 无 `pause` → 自动 chain 到下一 phase
每个 phase 输出写到 `.harnessed/checkpoints/`,session 中断后 `harnessed resume` 从最近 checkpoint 继续。
Q5. harnessed 自己是 CC plugin 吗?
混合体:
- `npx harnessed@latest setup` 跑的是 **Node.js CLI** (`bin/harnessed`)
- setup 装的 **workflow skills** (markdown) 进 `~/.claude/skills/`,由 Claude Code 运行时加载
- `/discuss` / `/plan` / `/task` / `/verify` 等是 CC 内的 slash command,触发 skill 执行
- CLI 和 CC skill 共享 `.harnessed/checkpoints/` 状态目录
---
## License
[Apache-2.0](./LICENSE) — 见 [NOTICE](./NOTICE) (含 Harness Inc. 商标 disclaimer)
支持开发: [](https://github.com/sponsors/easyinplay)