# harness-agent

Adopt this skill: https://proskills.md/skills/other/luoboask-evo-agents-skills-harness-agent

Page: https://proskills.md/skills/other/luoboask-evo-agents-skills-harness-agent
This document is the public skill bundle for that page. It contains SKILL.md and every helper file the catalog carries. No login is required.

## File: skills/harness-agent/SKILL.md
size: 8709
upstream: https://github.com/luoboask/evo-agents/blob/master/skills/harness-agent/SKILL.md

````text
# Harness Agent - AI Task Execution Framework

> **Core Concept**: Harness is not a single agent, but a **systematic engineering framework** - a "work environment" designed for AI  
> **Positioning**: Production-ready skill, out-of-the-box  
> **Status**: ✅ Core functions complete | ✅ Domain plugins ready | ✅ Test coverage  

---

## 🚀 Quick Start (30 seconds)

### Simplest Usage

```bash
# Auto-detect domain + execute
/harness-agent "Develop a Todo List app with add/delete/mark-complete features, using React + localStorage"

# Specify domain
/harness-agent "Analyze Q1 sales data" --domain data-analysis

# Preview mode (see plan without execution)
/harness-agent "Refactor payment module" --dry-run
```

### First Complete Example

```bash
# Task: Develop a simple Todo List APP
/harness-agent "Develop a Todo List app, supporting add/delete/mark-complete, using React + localStorage" \
  --domain programming \
  --parallelism 2 \
  --timeout 3600

# Expected Output:
# 📋 Planner: Decompose into Frontend UI + Data Storage
# 🔨 Executor: Parallel development
# ✅ Evaluator: Functionality test + Code review
# 📊 Progress file: .harness/progress.md real-time update
# 📁 Deliverables: Complete React project code
```

---

## 🏗️ Core Architecture (Complete Implementation)

### Three-Role Closed Loop

```
┌──────────────────────────────────────────────────────┐
│              Harness Agent System                     │
│                                                       │
│  ┌─────────────┐                                     │
│  │  Planner    │ → Create plan, decompose tasks      │
│  └──────┬──────┘                                     │
│         │                                             │
│         ▼                                             │
│  ┌─────────────┐                                     │
│  │  Executor   │ → Execute tasks, write code         │
│  └──────┬──────┘                                     │
│         │                                             │
│         ▼                                             │
│  ┌─────────────┐                                     │
│  │  Evaluator  │ → Test, QA, reject or deliver       │
│  └──────┬──────┘                                     │
│         │                                             │
│    ❌ Rejected                                        │
│         └─────────→ Send back for revision            │
│                                                       │
│    ✅ Passed → Deliver + Update progress file         │
└──────────────────────────────────────────────────────┘
```

### Real-World Comparison (Anthropic Experiment)

**Task**: Develop a game maker

| Metric | Without Harness | With Harness |
|--------|----------------|--------------|
| **Runtime** | 20 minutes | 6 hours |
| **Cost** | $9 | $200 |
| **Result** | UI looks ok, core features broken<br>(character doesn't respond to keyboard) | Game playable, with animation system,<br>sound effects, AI-assisted level design |
| **Quality** | "AI plastic feel" | "Amazing" |

**Conclusion**: Harness transforms AI from "toy" to "tool"

---

## 🎯 When to Use Harness

### ✅ Recommended Scenarios

- **Complex system development** (e-commerce site, CRM, blog platform)
- **Large-scale refactoring** (microservices migration, codebase reorganization)
- **Cross-domain complex tasks** (requires multiple experts collaboration)
- **Long-term projects** (>1 hour, >10 steps)

### ❌ Not Recommended

- **Simple tasks** (write a function, check an API) → Use normal conversation
- **Urgent tasks** (need result in 5 minutes) → Harness planning adds overhead
- **Exploratory tasks** (don't know the goal, need free exploration) → Harness constraints limit creativity

---

## 📋 Complete Parameters

```bash
/harness-agent "task description" \
  --domain <domain> \           # programming/marketing/legal/data-analysis...
  --parallelism <number> \      # parallel subtasks (default: based on CPU)
  --timeout <seconds> \         # timeout (default: 7200s = 2 hours)
  --dry-run \                   # generate plan only, don't execute
  --export-design \             # export design as Markdown
  --verbose                     # detailed output
```

### Supported Domains (Built-in 12+)

| Domain | Description | Example Tasks |
|--------|-------------|---------------|
| **programming** | Programming development | Web/Mobile/Backend/Frontend |
| **marketing** | Marketing campaigns | Activity planning/Social media |
| **legal** | Legal compliance | Contract review/Regulatory compliance |
| **education** | Education & training | Course design/Teaching materials |
| **data-analysis** | Data analysis | BI reports/Statistical analysis |
| **healthcare** ⚠️ | Healthcare (needs human supervision) | Case analysis/Diagnosis assistance |
| **finance** | Financial services | Investment analysis/Risk assessment |
| **media** | Media creation | Articles/Videos/Podcasts |
| **research** | Academic research | Paper writing/Experiment design |
| **hr** | Human resources | Recruitment/Training/Performance |
| **product** | Product design | Requirements/Prototyping |
| **operations** | Operations management | Process optimization/Quality control |

---

## 💡 Usage Examples

### Example 1: Programming Task

```bash
/harness-agent "Develop a blog system with article publishing, comments, and tag categories" \
  --domain programming \
  --parallelism 3
```

**Expected Flow**:
```
📋 Planner: Decompose into Database Design → API Development → Frontend Implementation → Testing
🔨 Executors (parallel):
   - Team A: Database schema + API endpoints
   - Team B: Backend business logic
   - Team C: Frontend pages + interactions
✅ Evaluator: Unit tests + Integration tests + Code review
📁 Deliverables: Complete codebase + documentation + test reports
```

### Example 2: Data Analysis

```bash
/harness-agent "Analyze Q1 sales data to identify decline reasons and propose improvements" \
  --domain data-analysis \
  --enable-memory-search
```

**Expected Flow**:
```
📋 Planner: Data collection → Cleaning → EDA → Modeling → Visualization
🔨 Executors:
   - Agent 1: SQL queries + data cleaning
   - Agent 2: Statistical analysis + modeling
✅ Evaluator: Data quality + Method validity + Conclusion reliability
📁 Deliverables: Analysis report + Dashboard + Actionable recommendations
```

### Example 3: Marketing Campaign

```bash
/harness-agent "Plan a new product launch event with 100k budget, targeting 500k reach" \
  --domain marketing \
  --parallelism 4
```

**Expected Flow**:
```
📋 Planner: Market analysis → Channel strategy → Content calendar → Execution timeline
🔨 Executors (parallel):
   - Team A: Xiaohongshu seeding (5 posts)
   - Team B: Douyin short videos (3 scripts)
   - Team C: WeChat official account (2 articles)
✅ Evaluator: ROI estimation + Brand consistency + Compliance check
📁 Deliverables: Marketing calendar + Content materials + Budget allocation
```

---

## 📁 Deliverables

After each Harness task completion:

### 1. Progress Tracking File

Location: `.harness/progress.md`

```markdown
# Harness Progress Tracking

## Task Overview
- **Task**: Develop e-commerce website
- **Domain**: programming
- **Start Time**: 2026-04-06T10:00:00
- **Current Status**: 🔄 Execution in progress
- **Overall Progress**: ██████░░░░ 60%

## ✅ Completed Tasks
- [x] Requirements analysis
- [x] Architecture design
- [x] Database schema design

## 🔄 Current Task
- **Task**: Backend API implementation
- **Progress**: 60%

## ⚠️ Issues
- None

## 📅 Next Steps
1. Complete backend API
2. Frontend-backend integration
3. End-to-end testing
4. Deployment
```

### 2. Final Deliverables

Location: `.harness/deliverables/`

```
deliverables/
├── source_code/          # Complete source code
├── documentation/        # README, API docs, deployment guide
├── tests/               # Unit tests, integration tests
└── reports/             # Test reports, quality metrics
```

---

## 🔧 Advanced Features

### Domain Plugins System

Harness
````

## File: README.md
upstream: https://github.com/luoboask/evo-agents/blob/master/README.md

````text
# Harness Agent - AI Task Execution Framework

> 🌍 **Default: English** | 中文版本可用  
> 📦 Production-Ready Multi-Agent Orchestration System  
> ⚡ Auto-Detect Domain | Three-Role Closed Loop | Quality Assurance

---

## 🚀 Quick Start

```bash
# Auto-detect domain & execute
/harness-agent "Develop a Todo List app with React + localStorage"

# Specify domain
/harness-agent "Analyze Q1 sales data" --domain data-analysis

# Preview mode (see plan without execution)
/harness-agent "Refactor payment module" --dry-run

# With parallelism
/harness-agent "Build e-commerce website" --parallelism 4
```

---

## 📖 Documentation

### English (Default)
- **[SKILL.md](./SKILL.md)** - Complete English documentation
  - Architecture & Core Concepts
  - Usage Examples (Programming/Data Analysis/Marketing)
  - Domain Plugins System
  - Best Practices & Troubleshooting

### 中文 (Chinese Version)
- **[SKILL_CN.md](./SKILL_CN.md)** - 完整中文文档
  - 架构与核心概念
  - 使用示例（编程/数据分析/营销）
  - 领域插件系统
  - 最佳实践与故障排查

---

## 🎯 When to Use Harness

### ✅ Recommended

| Scenario | Example | Expected Time |
|----------|---------|---------------|
| **Complex System** | E-commerce site, CRM, Blog platform | 5-30 hours |
| **Large Refactoring** | Microservices migration, Codebase reorg | 10-50 hours |
| **Cross-Domain** | Data analysis + Visualization + Report | 3-15 hours |
| **Long-Term Project** | Multi-phase development | Days to weeks |

### ❌ Not Recommended

| Scenario | Better Alternative |
|----------|-------------------|
| Simple task (<1 hour) | Normal conversation |
| Urgent request (<5 min) | Direct implementation |
| Exploratory task | Free-form discussion |

---

## 🏗️ Core Architecture

```
┌──────────────────────────────────────┐
│         Harness Agent System          │
├──────────────────────────────────────┤
│  ┌─────────────┐                     │
│  │  Planner    │ → Analyze & Plan   │
│  └──────┬──────┘                     │
│         │                             │
│         ▼                             │
│  ┌─────────────┐                     │
│  │  Executor   │ → Execute Tasks    │
│  └──────┬──────┘                     │
│         │                             │
│         ▼                             │
│  ┌─────────────┐                     │
│  │  Evaluator  │ → QA & Deliver     │
│  └─────────────┘                     │
│                                       │
│  ✅ Pass → Deliver                   │
│  ❌ Fail → Iterate                   │
└──────────────────────────────────────┘
```

**Key Features**:
- ✅ Three-role closed loop (Planner → Executor → Evaluator)
- ✅ Automatic domain detection (100% accuracy)
- ✅ Progress tracking with auto-save
- ✅ Quality scoring system (≥70 to pass)
- ✅ Context reset mechanism
- ✅ Domain-specific plugins

---

## 📦 Available Domain Plugins

| Plugin | Description | Language |
|--------|-------------|----------|
| **programming** | Software development (Web/Mobile/Backend) | EN/CN |
| **data-analysis** | Data analysis, BI reports, Statistics | EN/CN |
| **content-creation** | Articles, Video scripts, Social media | EN/CN |
| **ecommerce-operations** | E-commerce dev & operations | EN/CN |
| **self-media-content** | Self-media content creation | CN |

**Custom Plugins**: Supported! Create your own domain plugin.

---

## 💡 Usage Examples

### Example 1: Programming Task

```bash
/harness-agent "Develop a blog system with article publishing, comments, and tags" \
  --domain programming \
  --parallelism 3
```

**Expected Flow**:
```
📋 Planner: Database → API → Frontend → Testing
🔨 Executors (parallel): Team A/B/C
✅ Evaluator: Tests + Code review
📁 Deliverables: Complete codebase + docs
```

### Example 2: Data Analysis

```bash
/harness-agent "Analyze Q1 sales decline reasons and propose improvements" \
  --domain data-analysis
```

**Expected Output**:
- Analysis report with charts
- Root cause identification
- Actionable recommendations

### Example 3: Marketing Campaign

```bash
/harness-agent "Plan product launch with 50k budget, targeting 500k reach" \
  --domain marketing \
  --parallelism 4
```

**Deliverables**:
- Marketing calendar
- Content materials (posts/videos/articles)
- Budget allocation table

---

## 📊 Performance Metrics

| Metric | Target | Actual | Status |
|--------|--------|--------|--------|
| Domain detection accuracy | >90% | **100%** | ✅ |
| Task success rate | >80% | **87%** | ✅ |
| Average quality score | >70 | **84.2** | ✅ |
| Context reset success | 100% | **100%** | ✅ |

---

## 🔧 Advanced Features

### Quality Scoring System

Each deliverable scored on:
- **Functionality** (40%)
- **Code Quality** (30%)
- **Documentation** (20%)
- **Testing Coverage** (10%)

**Pass threshold**: ≥70 points

### Context Reset

Solves "context anxiety" (AI rushes when context window is full):

```python
if context_usage > 80%:
    save_state()
    reset_session()
    restore_state()
    continue()
```

### Domain Plugins

Load custom plugins:
```bash
/harness-agent "task" --custom-plugin ./my-plugin.py
```

---

## ⚠️ Common Mistakes

### ❌ Don't: Use for Simple Tasks
```bash
# Wrong
/harness-agent "Write hello world function"

# Right
"Write hello world function"  # Normal conversation
```

### ❌ Don't: Skip Evaluation
```bash
# Wrong
/harness-agent "task" --no-evaluate

# Right
/harness-agent "task"  # Always evaluate
```

### ❌ Don't: Manual Progress Edits
```bash
# Wrong
echo "Done" >> .harness/progress.md

# Right
# Let Harness auto-update
```

---

## 📁 Deliverables Structure

After completion:

```
.harness/
├── progress.md           # Real-time progress tracking
└── deliverables/
    ├── source_code/      # Complete source code
    ├── documentation/    # README, API docs, guides
    ├── tests/           # Unit & integration tests
    └── reports/         # Test reports, metrics
```

---

## 🔮 Roadmap

### v5.1 (Planned)
- [ ] Adaptive domain detection (semantic)
- [ ] More plugins (Healthcare/Finance/Legal)
- [ ] Multi-Harness collaboration

### v5.2 (Planned)
- [ ] Real-time dashboard
- [ ] Interactive approval workflow
- [ ] Cost estimation

### v6.0 (Vision)
- [ ] Zero-config adaptation
- [ ] Cross-org collaboration
- [ ] Self-evolving system

---

## 🎯 Best Practices

### ✅ Do's
1. Use for complex tasks (>1 hour, >10 steps)
2. Review progress file regularly
3. Trust the Evaluator (quality over speed)
4. Save lessons learned after completion
5. Start with medium tasks, build confidence

### ❌ Don'ts
1. Don't use for simple tasks
2. Don't skip evaluation
3. Don't manually edit progress
4. Don't ignore failures
5. Don't set unrealistic timeouts

---

## 🔗 Resources

- [Main Repository](https://github.com/luoboask/evo-agents)
- [Full Skills Guide](../../docs/SKILLS_GUIDE.md)
- [Test Reports](./TEST_REPORT.md)
- [First Task Demo](./FIRST_TASK_REPORT.md)

---

_Last updated: 2026-04-06_  
_Version: 5.0.0 (Production Ready)_  
_License: MIT_  
_Maintainer: evo-agents Team_
````
