一个自我改进的产品系统，每天读取报告，识别#1可执行优先事项，并自主实施。

概念： 你的产品生成关于其性能的报告。人工智能代理分析这些报告，挑选最有影响力的解决方案，制定计划并实施——这一切都发生在你睡觉时。你醒来时，PR已经准备好接受审查。

基于Kieran Klaassen的复合工程方法论、Geoffrey Huntley的Ralph模式和Ryan Carson的实现。

工作原理

快速启动

先决条件

• Amp CLI或Claude代码
• 用于基于浏览器的代理浏览器测试（npm install -g agent-browser）
• jq 安装（macOS 上的 brew install jq ）
• [gh CLI已安装并认证（brew install gh）
• 其中一个LLM提供者配置为：
• 拟人： export ANTHROPIC_API_KEY=sk-ant-...
• OpenRouter： export OPENROUTER_API_KEY=sk-or-...
• AI网关： export AI_GATEWAY_URL=https://... AI_GATEWAY_API_KEY=...

安装

Compound Product 会安装到你现有的项目仓库中。它会添加脚本和配置，这些配置可以与你的代码库协同工作。

或者告诉你的AI代理（在项目内部）：

这会产生：

Compound Product

A self-improving product system that reads daily reports, identifies the #1 actionable priority, and autonomously implements it.

The concept: Your product generates reports about its performance. An AI agent analyzes those reports, picks the most impactful fix, creates a plan, and implements it—all while you sleep. You wake up to a PR ready for review.

Built on Kieran Klaassen’s Compound Engineering methodology, Geoffrey Huntley’s Ralph pattern, and Ryan Carson’s implementation.

How It Works

flowchart TD
    subgraph "Phase 0: Your System"
        GEN[Generate Daily Reportyour code, cron, etc.]
        GEN --> R[📊 reports/*.mdmetrics, errors, feedback]
    end
    
    subgraph "Phase 1: Analysis"
        A[analyze-report.shLLM API]
        R --> A
        A --> J[analysis.jsonpriority + criteria]
    end
    
    subgraph "Phase 2: Planning"
        B[Create Branchcompound/feature-name]
        J --> B
        B --> PRD[AI AgentLoad prd skill]
        PRD --> MD[tasks/prd-feature.md]
        MD --> TASKS[AI AgentLoad tasks skill]
        TASKS --> JSON[prd.jsonexecutable tasks]
    end
    
    subgraph "Phase 3: Execution Loop"
        LOOP[loop.shmax N iterations]
        JSON --> LOOP
        LOOP --> PICK[Pick next taskwhere passes: false]
        PICK --> IMPL[AI AgentImplement task]
        IMPL --> CHECK{QualityChecks?}
        CHECK -->|Pass| COMMIT[Git Commit]
        CHECK -->|Fail| FIX[Fix & Retry]
        FIX --> CHECK
        COMMIT --> UPDATE[Update prd.jsonpasses: true]
        UPDATE --> DONE{All tasksdone?}
        DONE -->|No| PICK
        DONE -->|Yes| EXIT[Exit loop]
    end
    
    subgraph Output
        EXIT --> PUSH[Git Push]
        PUSH --> PR[🎉 Pull RequestReady for review]
    end
    
    click A "https://github.com/snarktank/compound-product/blob/main/scripts/analyze-report.sh" "View analyze-report.sh"
    click LOOP "https://github.com/snarktank/compound-product/blob/main/scripts/loop.sh" "View loop.sh"
    click PRD "https://github.com/snarktank/compound-product/blob/main/skills/prd/SKILL.md" "View PRD skill"
    click TASKS "https://github.com/snarktank/compound-product/blob/main/skills/tasks/SKILL.md" "View Tasks skill"

Quick Start

Prerequisites

• Amp CLI or Claude Code
• agent-browser for browser-based testing (npm install -g agent-browser)
• jq installed (brew install jq on macOS)
• gh CLI installed and authenticated (brew install gh)
• One of these LLM providers configured:
  • Anthropic: export ANTHROPIC_API_KEY=sk-ant-...