@FakeMaidenMaker: 用 AI agent 写代码最怕的就是失控：agent 自顾自跑、质量飘忽、你不知道它现在在哪个阶段、改到一半又乱了。 AWS 刚开源了一套专门给 AI coding agent 用的开发生命周期工作流规则——AI-DLC，让 agent…

X AI KOLs Timeline 工具
ai-coding-agent open-source aws development-workflow ai-dlc software-development agent-control

摘要

AWS 开源了 AI-DLC（AI-Driven Development Life Cycle），一套为 AI coding agent 设计的开发生命周期工作流规则，帮助开发者控制 agent 行为，确保质量。支持 Claude Code、Cursor、GitHub Copilot 等多种平台。

用 AI agent 写代码最怕的就是失控：agent 自顾自跑、质量飘忽、你不知道它现在在哪个阶段、改到一半又乱了。 AWS 刚开源了一套专门给 AI coding agent 用的开发生命周期工作流规则——AI-DLC，让 agent 按方法论一步步走——需求分析、设计、实现三阶段自动适配你的项目，维持质量标准，整个流程你始终掌控。 GitHub：https://github.com/awslabs/aidlc-workflows… 跨平台支持做得很全：Claude Code、Cursor、GitHub Copilot、Kiro（用 Steering Files）、Amazon Q Developer、Cline 都能用。部署极简——下载 release zip，解压出 aidlc-rules 目录，按你的 agent 平台放到对应位置，规则即刻生效。 AI agent 越强大，越需要一套靠谱的工作流把它框住。
查看原文
查看缓存全文

缓存时间: 2026/06/02 19:37

用 AI agent 写代码最怕的就是失控：agent 自顾自跑、质量飘忽、你不知道它现在在哪个阶段、改到一半又乱了。

AWS 刚开源了一套专门给 AI coding agent 用的开发生命周期工作流规则——AI-DLC，让 agent 按方法论一步步走——需求分析、设计、实现三阶段自动适配你的项目，维持质量标准，整个流程你始终掌控。

GitHub：https://github.com/awslabs/aidlc-workflows…

跨平台支持做得很全：Claude Code、Cursor、GitHub Copilot、Kiro（用 Steering Files）、Amazon Q Developer、Cline 都能用。部署极简——下载 release zip，解压出 aidlc-rules 目录，按你的 agent 平台放到对应位置，规则即刻生效。

AI agent 越强大，越需要一套靠谱的工作流把它框住。

awslabs/aidlc-workflows

Source: https://github.com/awslabs/aidlc-workflows

AI-DLC (AI-Driven Development Life Cycle)

Generative AI can make mistakes. You should consider reviewing all output and costs generated by your chosen AI model and agentic coding assistant. See AWS Responsible AI Policy.

AI-DLC is an intelligent software development workflow that adapts to your needs, maintains quality standards, and keeps you in control of the process. For learning more about AI-DLC Methodology, read this blog and the Method Definition Paper referred in it.

Table of Contents

Common

  1. Download the latest release zip file named ai-dlc-rules-v<release-number>.zip from the Releases page to a folder outside your project directory (e.g., ~/Downloads).
  2. Extract the zip. It contains an aidlc-rules/ folder with two subdirectories:
    • aws-aidlc-rules/ — the core AI-DLC workflow rules
    • aws-aidlc-rule-details/ — detailed rules conditionally referenced by the core rules
  3. Follow the setup instructions for your coding agent and platform below.

Platform-Specific Setup

Kiro

AI-DLC uses Kiro Steering Files within your project workspace.

The commands below assume you extracted the zip to your Downloads folder so that the resulting path is Downloads/aidlc-rules/. If you used a different location, replace Downloads with your actual folder path.

Windows users: if you used File Explorer’s Extract All… dialog, it defaults to creating a wrapper folder named after the zip (e.g., ai-dlc-rules-v0.1.8\aidlc-rules\...). Either uncheck/edit that destination so the contents land directly in Downloads\aidlc-rules\ (matching the commands below), or prepend ai-dlc-rules-v<version>\ to each Downloads\ path in the commands — substituting <version> with the release you downloaded.

On macOS/Linux:

mkdir -p .kiro/steering
cp -R ~/Downloads/aidlc-rules/aws-aidlc-rules .kiro/steering/
cp -R ~/Downloads/aidlc-rules/aws-aidlc-rule-details .kiro/

On Windows (PowerShell):

New-Item -ItemType Directory -Force -Path ".kiro\steering"
Copy-Item -Recurse "$env:USERPROFILE\Downloads\aidlc-rules\aws-aidlc-rules" ".kiro\steering\"
Copy-Item -Recurse "$env:USERPROFILE\Downloads\aidlc-rules\aws-aidlc-rule-details" ".kiro\"

On Windows (CMD):

mkdir .kiro\steering
xcopy %USERPROFILE%\Downloads\aidlc-rules\aws-aidlc-rules .kiro\steering\aws-aidlc-rules\ /E /I
xcopy %USERPROFILE%\Downloads\aidlc-rules\aws-aidlc-rule-details .kiro\aws-aidlc-rule-details\ /E /I

Your project should look like:

<project-root>/
    ├── .kiro/
    │     ├── steering/
    │     │      ├── aws-aidlc-rules/
    │     ├── aws-aidlc-rule-details/

To verify the rules are loaded:

Verify in Kiro IDE

Open the steering files panel and confirm you see an entry for core-workflow under Workspace as shown in the screenshot below.

AI-DLC Rules in Kiro IDE

We use Kiro IDE in Vibe mode to run the AI-DLC workflow. This ensures that AI-DLC workflow guides the development workflow in Kiro. At times, Kiro may nudge you to switch to spec mode. Select No to such prompts to stay in Vibe mode.

Staying in Kiro Vibe mode

Verify in Kiro CLI

Run kiro-cli, then /context show, and confirm entries for .kiro/steering/aws-aidlc-rules.

AI-DLC Rules in Kiro CLI

Amazon Q Developer IDE Plugin/Extension

AI-DLC uses Amazon Q Rules within your project workspace.

The commands below assume you extracted the zip to your Downloads folder so that the resulting path is Downloads/aidlc-rules/. If you used a different location, replace Downloads with your actual folder path.

Windows users: if you used File Explorer’s Extract All… dialog, it defaults to creating a wrapper folder named after the zip (e.g., ai-dlc-rules-v0.1.8\aidlc-rules\...). Either uncheck/edit that destination so the contents land directly in Downloads\aidlc-rules\ (matching the commands below), or prepend ai-dlc-rules-v<version>\ to each Downloads\ path in the commands — substituting <version> with the release you downloaded.

On macOS/Linux:

mkdir -p .amazonq/rules
cp -R ~/Downloads/aidlc-rules/aws-aidlc-rules .amazonq/rules/
cp -R ~/Downloads/aidlc-rules/aws-aidlc-rule-details .amazonq/

On Windows (PowerShell):

New-Item -ItemType Directory -Force -Path ".amazonq\rules"
Copy-Item -Recurse "$env:USERPROFILE\Downloads\aidlc-rules\aws-aidlc-rules" ".amazonq\rules\"
Copy-Item -Recurse "$env:USERPROFILE\Downloads\aidlc-rules\aws-aidlc-rule-details" ".amazonq\"

On Windows (CMD):

mkdir .amazonq\rules
xcopy %USERPROFILE%\Downloads\aidlc-rules\aws-aidlc-rules .amazonq\rules\aws-aidlc-rules\ /E /I
xcopy %USERPROFILE%\Downloads\aidlc-rules\aws-aidlc-rule-details .amazonq\aws-aidlc-rule-details\ /E /I

Your project should look like:

<project-root>/
    ├── .amazonq/
    │     ├── rules/
    │     │     ├── aws-aidlc-rules/
    │     ├── aws-aidlc-rule-details/

To verify the rules are loaded:

  1. In the Amazon Q Chat window, click the Rules button in the lower right corner.
  2. Confirm you see entries for .amazonq/rules/aws-aidlc-rules.
AI-DLC Rules in Q Developer IDE plugin

Cursor IDE

AI-DLC uses Cursor Rules to implement its intelligent workflow.

The commands below assume you extracted the zip to your Downloads folder so that the resulting path is Downloads/aidlc-rules/. If you used a different location, replace Downloads with your actual folder path.

Windows users: if you used File Explorer’s Extract All… dialog, it defaults to creating a wrapper folder named after the zip (e.g., ai-dlc-rules-v0.1.8\aidlc-rules\...). Either uncheck/edit that destination so the contents land directly in Downloads\aidlc-rules\ (matching the commands below), or prepend ai-dlc-rules-v<version>\ to each Downloads\ path in the commands — substituting <version> with the release you downloaded.

Option 1: Project Rules (Recommended)

Unix/Linux/macOS:

mkdir -p .cursor/rules

cat > .cursor/rules/ai-dlc-workflow.mdc << 'EOF'
---
description: "AI-DLC (AI-Driven Development Life Cycle) adaptive workflow for software development"
alwaysApply: true
---

EOF
cat ~/Downloads/aidlc-rules/aws-aidlc-rules/core-workflow.md >> .cursor/rules/ai-dlc-workflow.mdc

mkdir -p .aidlc-rule-details
cp -R ~/Downloads/aidlc-rules/aws-aidlc-rule-details/* .aidlc-rule-details/

Windows PowerShell:

New-Item -ItemType Directory -Force -Path ".cursor\rules"

$frontmatter = @"
---
description: "AI-DLC (AI-Driven Development Life Cycle) adaptive workflow for software development"
alwaysApply: true
---

"@
$frontmatter | Out-File -FilePath ".cursor\rules\ai-dlc-workflow.mdc" -Encoding utf8

Get-Content "$env:USERPROFILE\Downloads\aidlc-rules\aws-aidlc-rules\core-workflow.md" | Add-Content ".cursor\rules\ai-dlc-workflow.mdc"

New-Item -ItemType Directory -Force -Path ".aidlc-rule-details"
Copy-Item "$env:USERPROFILE\Downloads\aidlc-rules\aws-aidlc-rule-details\*" ".aidlc-rule-details\" -Recurse

Windows CMD:

mkdir .cursor\rules

(
echo ---
echo description: "AI-DLC (AI-Driven Development Life Cycle) adaptive workflow for software development"
echo alwaysApply: true
echo ---
echo.
) > .cursor\rules\ai-dlc-workflow.mdc

type "%USERPROFILE%\Downloads\aidlc-rules\aws-aidlc-rules\core-workflow.md" >> .cursor\rules\ai-dlc-workflow.mdc

mkdir .aidlc-rule-details
xcopy "%USERPROFILE%\Downloads\aidlc-rules\aws-aidlc-rule-details" ".aidlc-rule-details\" /E /I

Option 2: AGENTS.md (Simple Alternative)

Unix/Linux/macOS:

cp ~/Downloads/aidlc-rules/aws-aidlc-rules/core-workflow.md ./AGENTS.md
mkdir -p .aidlc-rule-details
cp -R ~/Downloads/aidlc-rules/aws-aidlc-rule-details/* .aidlc-rule-details/

Windows PowerShell:

Copy-Item "$env:USERPROFILE\Downloads\aidlc-rules\aws-aidlc-rules\core-workflow.md" ".\AGENTS.md"
New-Item -ItemType Directory -Force -Path ".aidlc-rule-details"
Copy-Item "$env:USERPROFILE\Downloads\aidlc-rules\aws-aidlc-rule-details\*" ".aidlc-rule-details\" -Recurse

Windows CMD:

copy "%USERPROFILE%\Downloads\aidlc-rules\aws-aidlc-rules\core-workflow.md" ".\AGENTS.md"
mkdir .aidlc-rule-details
xcopy "%USERPROFILE%\Downloads\aidlc-rules\aws-aidlc-rule-details" ".aidlc-rule-details\" /E /I

Verify Setup:

  1. Open Cursor Settings → Rules, Commands
  2. Under Project Rules, you should see ai-dlc-workflow listed
  3. For AGENTS.md, it will be automatically detected and applied

AI-DLC Rules in Cursor

Directory Structure (Option 1):

<my-project>/
├── .cursor/
│   └── rules/
│       └── ai-dlc-workflow.mdc
└── .aidlc-rule-details/
    ├── common/
    ├── inception/
    ├── construction/
    ├── extensions/
    └── operations/

Cline

AI-DLC uses Cline Rules to implement its intelligent workflow.

The commands below assume you extracted the zip to your Downloads folder so that the resulting path is Downloads/aidlc-rules/. If you used a different location, replace Downloads with your actual folder path.

Windows users: if you used File Explorer’s Extract All… dialog, it defaults to creating a wrapper folder named after the zip (e.g., ai-dlc-rules-v0.1.8\aidlc-rules\...). Either uncheck/edit that destination so the contents land directly in Downloads\aidlc-rules\ (matching the commands below), or prepend ai-dlc-rules-v<version>\ to each Downloads\ path in the commands — substituting <version> with the release you downloaded.

Option 1: .clinerules Directory (Recommended)

Unix/Linux/macOS:

mkdir -p .clinerules
cp ~/Downloads/aidlc-rules/aws-aidlc-rules/core-workflow.md .clinerules/
mkdir -p .aidlc-rule-details
cp -R ~/Downloads/aidlc-rules/aws-aidlc-rule-details/* .aidlc-rule-details/

Windows PowerShell:

New-Item -ItemType Directory -Force -Path ".clinerules"
Copy-Item "$env:USERPROFILE\Downloads\aidlc-rules\aws-aidlc-rules\core-workflow.md" ".clinerules\"
New-Item -ItemType Directory -Force -Path ".aidlc-rule-details"
Copy-Item "$env:USERPROFILE\Downloads\aidlc-rules\aws-aidlc-rule-details\*" ".aidlc-rule-details\" -Recurse

Windows CMD:

mkdir .clinerules
copy "%USERPROFILE%\Downloads\aidlc-rules\aws-aidlc-rules\core-workflow.md" ".clinerules\"
mkdir .aidlc-rule-details
xcopy "%USERPROFILE%\Downloads\aidlc-rules\aws-aidlc-rule-details" ".aidlc-rule-details\" /E /I

Option 2: AGENTS.md (Alternative)

Unix/Linux/macOS:

cp ~/Downloads/aidlc-rules/aws-aidlc-rules/core-workflow.md ./AGENTS.md
mkdir -p .aidlc-rule-details
cp -R ~/Downloads/aidlc-rules/aws-aidlc-rule-details/* .aidlc-rule-details/

Windows PowerShell:

Copy-Item "$env:USERPROFILE\Downloads\aidlc-rules\aws-aidlc-rules\core-workflow.md" ".\AGENTS.md"
New-Item -ItemType Directory -Force -Path ".aidlc-rule-details"
Copy-Item "$env:USERPROFILE\Downloads\aidlc-rules\aws-aidlc-rule-details\*" ".aidlc-rule-details\" -Recurse

Windows CMD:

copy "%USERPROFILE%\Downloads\aidlc-rules\aws-aidlc-rules\core-workflow.md" ".\AGENTS.md"
mkdir .aidlc-rule-details
xcopy "%USERPROFILE%\Downloads\aidlc-rules\aws-aidlc-rule-details" ".aidlc-rule-details\" /E /I

Verify Setup:

  1. In Cline’s chat interface, look for the Rules popover under the chat input field
  2. Verify that core-workflow.md is listed and active
  3. You can toggle the rule file on/off as needed

AI-DLC Rules in Cline

Directory Structure (Option 1):

<my-project>/
├── .clinerules/
│   └── core-workflow.md
└── .aidlc-rule-details/
    ├── common/
    ├── inception/
    ├── construction/
    ├── extensions/
    └── operations/

Claude Code

AI-DLC uses Claude Code’s project memory file (CLAUDE.md) to implement its intelligent workflow.

The commands below assume you extracted the zip to your Downloads folder so that the resulting path is Downloads/aidlc-rules/. If you used a different location, replace Downloads with your actual folder path.

Windows users: if you used File Explorer’s Extract All… dialog, it defaults to creating a wrapper folder named after the zip (e.g., ai-dlc-rules-v0.1.8\aidlc-rules\...). Either uncheck/edit that destination so the contents land directly in Downloads\aidlc-rules\ (matching the commands below), or prepend ai-dlc-rules-v<version>\ to each Downloads\ path in the commands — substituting <version> with the release you downloaded.

Option 1: Project Root (Recommended)

Unix/Linux/macOS:

cp ~/Downloads/aidlc-rules/aws-aidlc-rules/core-workflow.md ./CLAUDE.md
mkdir -p .aidlc-rule-details
cp -R ~/Downloads/aidlc-rules/aws-aidlc-rule-details/* .aidlc-rule-details/

Windows PowerShell:

Copy-Item "$env:USERPROFILE\Downloads\aidlc-rules\aws-aidlc-rules\core-workflow.md" ".\CLAUDE.md"
New-Item -ItemType Directory -Force -Path ".aidlc-rule-details"
Copy-Item "$env:USERPROFILE\Downloads\aidlc-rules\aws-aidlc-rule-details\*" ".aidlc-rule-details\" -Recurse

Windows CMD:

copy "%USERPROFILE%\Downloads\aidlc-rules\aws-aidlc-rules\core-workflow.md" ".\CLAUDE.md"
mkdir .aidlc-rule-details
xcopy "%USERPROFILE%\Downloads\aidlc-rules\aws-aidlc-rule-details" ".aidlc-rule-details\" /E /I

Option 2: .claude Directory

Unix/Linux/macOS:

mkdir -p .claude
cp ~/Downloads/aidlc-rules/aws-aidlc-rules/core-workflow.md .claude/CLAUDE.md
mkdir -p .aidlc-rule-details
cp -R ~/Downloads/aidlc-rules/aws-aidlc-rule-details/* .aidlc-rule-details/

Windows PowerShell:

New-Item -ItemType Directory -Force -Path ".claude"
Copy-Item "$env:USERPROFILE\Downloads\aidlc-rules\aws-aidlc-rules\core-workflow.md" ".claude\CLAUDE.md"
New-Item -ItemType Directory -Force -Path ".aidlc-rule-details"
Copy-Item "$env:USERPROFILE\Downloads\aidlc-rules\aws-aidlc-rule-details\*" ".aidlc-rule-details\" -Recurse

Windows CMD:

mkdir .claude
copy "%USERPROFILE%\Downloads\aidlc-rules\aws-aidlc-rules\core-workflow.md" ".claude\CLAUDE.md"
mkdir .aidlc-rule-details
xcopy "%USERPROFILE%\Downloads\aidlc-rules\aws-aidlc-rule-details" ".aidlc-rule-details\" /E /I

Verify Setup:

  1. Start Claude Code in your project directory (CLI: claude or VS Code extension)
  2. Use the /config command to view current configuration
  3. Ask Claude: “What instructions are currently active in this project?”

Directory Structure (Option 1):

<my-project>/
├── CLAUDE.md
└── .aidlc-rule-details/
    ├── common/
    ├── inception/
    ├── construction/
    ├── extensions/
    └── operations/

GitHub Copilot

AI-DLC uses GitHub Copilot custom instructions to implement its intelligent workflow. The .github/copilot-instructions.md file is automatically detected and applied to all chat requests in the workspace.

The commands below assume you extracted the zip to your Downloads folder so that the resulting path is Downloads/aidlc-rules/. If you used a different location, replace Downloads with your actual folder path.

Windows users: if you used File Explorer’s Extract All… dialog, it defaults to creating a wrapper folder named after the zip (e.g., ai-dlc-rules-v0.1.8\aidlc-rules\...). Either uncheck/edit that destination so the contents land directly in Downloads\aidlc-rules\ (matching the commands below), or prepend ai-dlc-rules-v<version>\ to each Downloads\ path in the commands — substituting <version> with the release you downloaded.

Unix/Linux/macOS:

mkdir -p .github
cp ~/Downloads/aidlc-rules/aws-aidlc-rules/core-workflow.md .github/copilot-instructions.md
mkdir -p .aidlc-rule-details
cp -R ~/Downloads/aidlc-rules/aws-aidlc-rule-details/* .aidlc-rule-details/

Windows PowerShell:

New-Item -ItemType Directory -Force -Path ".github"
Copy-Item "$env:USERPROFILE\Downloads\aidlc-rules\aws-aidlc-rules\core-workflow.md" ".github\copilot-instructions.md"
New-Item -ItemType Directory -Force -Path ".aidlc-rule-details"
Copy-Item "$env:USERPROFILE\Downloads\aidlc-rules\aws-aidlc-rule-details\*" ".aidlc-rule-details\" -Recurse

Windows CMD:

mkdir .github
copy "%USERPROFILE%\Downloads\aidlc-rules\aws-aidlc-rules\core-workflow.md" ".github\copilot-instructions.md"
mkdir .aidlc-rule-details
xcopy "%USERPROFILE%\Downloads\aidlc-rules\aws-aidlc-rule-details" ".aidlc-rule-details\" /E /I

Verify Setup:

  1. Open VS Code with your project folder
  2. Open the Copilot Chat panel (Cmd/Ctrl+Shift+I)
  3. Select Configure Chat (gear icon) > Chat Instructions and verify that copilot-instructions is listed
  4. Alternatively, type /instructions in the chat input to view active instructions

Directory Structure:

<my-project>/
├── .github/
│   └── copilot-instructions.md
└── .aidlc-rule-details/
    ├── common/
    ├── inception/
    ├── construction/
    ├── extensions/
    └── operations/

OpenAI Codex

AI-DLC supports OpenAI Codex as a supported coding agent, using the Codex AGENTS.md convention to deliver its intelligent workflow. Codex automatically discovers and loads AGENTS.md from your project root when you start a session.

The commands below assume you extracted the zip to your Downloads folder so that the resulting path is Downloads/aidlc-rules/. If you used a different location, replace Downloads with your actual folder path.

Windows users: if you used File Explorer’s Extract All… dialog, it defaults to creating a wrapper folder named after the zip (e.g., ai-dlc-rules-v0.1.8\aidlc-rules\...). Either uncheck/edit that destination so the contents land directly in Downloads\aidlc-rules\ (matching the commands below), or prepend ai-dlc-rules-v<version>\ to each Downloads\ path in the commands — substituting <version> with the release you downloaded.

Unix/Linux/macOS:

cp ~/Downloads/aidlc-rules/aws-aidlc-rules/core-workflow.md ./AGENTS.md
mkdir -p .aidlc-rule-details
cp -R ~/Downloads/aidlc-rules/aws-aidlc-rule-details/* .aidlc-rule-details/

Windows PowerShell:

Copy-Item "$env:USERPROFILE\Downloads\aidlc-rules\aws-aidlc-rules\core-workflow.md" ".\AGENTS.md"
New-Item -ItemType Directory -Force -Path ".aidlc-rule-details"
Copy-Item "$env:USERPROFILE\Downloads\aidlc-rules\aws-aidlc-rule-details\*" ".aidlc-rule-details\" -Recurse

Windows CMD:

copy "%USERPROFILE%\Downloads\aidlc-rules\aws-aidlc-rules\core-workflow.md" ".\AGENTS.md"
mkdir .aidlc-rule-details
xcopy "%USERPROFILE%\Downloads\aidlc-rules\aws-aidlc-rule-details" ".aidlc-rule-details\" /E /I

Verify Setup:

  1. Start a Codex session in your project directory
  2. Ask Codex: For existing project - “Using AIDLC analyze the project?” or For new project “Using Aidlc what workflow do you see” .
  3. Codex should describe the AI-DLC three-phase workflow (Inception → Construction → Operations)

The AGENTS.md file is designed to fit within Codex’s instruction budget under default settings. If you add substantial project-specific content and Codex reports that the project documentation exceeds its instruction limit, you can increase the limit in your Codex configuration (for example, by adjusting project_doc_max_bytes in your config.toml file):

project_doc_max_bytes = 65536  # Example value; choose a limit appropriate for your project

Directory Structure:

<my-project>/
├── AGENTS.md
└── .aidlc-rule-details/
    ├── common/
    ├── inception/
    ├── construction/
    ├── extensions/
    └── operations/

Other Agents

AI-DLC works with any coding agent that supports project-level rules or steering files. The general approach:

  1. Place aws-aidlc-rules/ wherever your agent reads project rules from (consult your agent’s documentation).
  2. Place aws-aidlc-rule-details/ at a sibling level so the rules can reference it.

If your agent has no convention for rules files, place both folders at your project root and point the agent to aws-aidlc-rules/ as its rules directory.

Usage

  1. Start any software development project by stating your intent starting with the phrase “Using AI-DLC, …” in the chat
  2. AI-DLC workflow automatically activates and guides you from there
  3. Answer structured questions that AI-DLC asks you
  4. Carefully review every plan that AI generates. Provide your oversight and validation
  5. Review the execution plan to see which stages will run
  6. Carefully review the artifacts and approve each stage to maintain control
  7. All the artifacts will be generated in the aidlc-docs/ directory

Three-Phase Adaptive Workflow

AI-DLC follows a structured three-phase approach that adapts to your project’s complexity:

🔵 INCEPTION PHASE

Determines WHAT to build and WHY

  • Requirements analysis and validation
  • User story creation (when applicable)
  • Application Design and creating units of work for parallel development
  • Risk assessment and complexity evaluation

🟢 CONSTRUCTION PHASE

Determines HOW to build it

  • Detailed component design
  • Code generation and implementation
  • Build configuration and testing strategies
  • Quality assurance and validation

🟡 OPERATIONS PHASE

Deployment and monitoring (future)

  • Deployment automation and infrastructure
  • Monitoring and observability setup
  • Production readiness validation

Key Features

FeatureDescription
Adaptive IntelligenceOnly executes stages that add value to your specific request
Context-AwareAnalyzes existing codebase and complexity requirements
Risk-BasedComplex changes get comprehensive treatment, simple changes stay efficient
Question-DrivenStructured multiple-choice questions in files, not chat
Always in ControlReview execution plans and approve each phase
ExtensibleLayer custom rules e.g. security, compliance, and organization-specific rules on top of the core workflow

Extensions

AI-DLC supports an extension system that lets you layer additional rules on top of the core workflow. Extensions are markdown files organized under aws-aidlc-rule-details/extensions/ and grouped by category (e.g., security/, testing/).

How Extensions Work

Each extension consists of two files placed in the same directory:

  • A rules file (e.g., security-baseline.md) containing the extension’s rules.
  • An opt-in file (e.g., security-baseline.opt-in.md) containing a structured multiple-choice question presented to the user during Requirements Analysis.

At workflow start, AI-DLC scans the extensions/ directory and loads only *.opt-in.md files. During Requirements Analysis, it presents each opt-in prompt to the user. When the user opts in, the corresponding rules file is loaded (derived by naming convention: strip .opt-in.md, append .md). When the user opts out, the rules file is never loaded. Extensions without a matching *.opt-in.md file are always enforced.

Once enabled, extension rules are blocking constraints — at each stage, the model verifies compliance before allowing the stage to proceed.

Built-in Extensions

The extensions/ directory ships with the following (new extensions may be added over time):

aws-aidlc-rule-details/
└── extensions/
    ├── security/                      # Extension category
    │   └── baseline/
    │       ├── security-baseline.md          # Baseline security rules
    │       └── security-baseline.opt-in.md   # Opt-in prompt
    └── testing/                       # Extension category
        └── property-based/
            ├── property-based-testing.md          # Property-based testing rules
            └── property-based-testing.opt-in.md   # Opt-in prompt

The security extension rules are provided as a directional reference for building effective security rules within AI-DLC workflows. Each organization should build, customize, and thoroughly test their own security rules before deploying in production workflows.

Adding Your Own Extensions

You can extend an existing category or create an entirely new one.

  1. Create a directory under extensions/ (e.g., security/compliance/ or performance/baseline/).
  2. Add a rules file (e.g., compliance.md). Follow the same structure as security-baseline.md:
    • Define each rule as a heading in the format ## Rule <PREFIX-NN>: <Title> where the prefix is a short category identifier and NN is a sequential number (e.g., COMPLIANCE-01, COMPLIANCE-02). These IDs are referenced in audit logs and compliance summaries, so they must be unique across all loaded extensions.
    • Include a Rule section describing the requirement.
    • Include a Verification section with concrete checks the model should evaluate.
  3. Add a matching opt-in file using the naming convention <name>.opt-in.md (e.g., compliance.opt-in.md). See security-baseline.opt-in.md for the expected format. Omitting this file means the extension is always enforced with no user opt-out.
  4. Rules are blocking by default — if verification criteria are not met, the stage cannot proceed until the finding is resolved.

Supporting Tools

The scripts/ directory contains supporting tools that enhance the AI-DLC workflow:

AIDLC Evaluator

Location: scripts/aidlc-evaluator/

Automated testing and reporting framework for validating changes to AI-DLC workflows. The evaluator provides:

  • Golden Test Cases — Curated baseline test cases for validation
  • Execution Framework — Orchestration for running test cases through evaluation pipelines
  • Semantic Evaluation — AI-based assessment of output correctness and completeness
  • Code Evaluation — Static analysis (linting, security scanning, duplication detection)
  • NFR Evaluation — Non-functional requirements testing (token usage, execution time, cross-model consistency)
  • CI/CD Integration — Automated pipelines for PR validation

Quick Start:

cd scripts/aidlc-evaluator
uv sync
uv run python run.py test

Documentation: See scripts/aidlc-evaluator/README.md

AIDLC Design Reviewer

Location: scripts/aidlc-designreview/

⚠️ EXPERIMENTAL FEATURE — AI-powered design review tool that analyzes AIDLC design artifacts using Claude models via AWS Bedrock.

Features:

  • Multi-Agent Review — Three specialized AI agents (Critique, Alternatives, Gap Analysis)
  • Quality Scoring — Weighted severity analysis with actionable recommendations
  • Two Deployment Modes:
    • CLI Tool — On-demand reviews for CI/CD pipelines
    • Claude Code Hook — Real-time review during development (experimental)

Installation (CLI Tool):

cd scripts/aidlc-designreview
uv sync --extra test
source .venv/bin/activate  # Linux/Mac
design-reviewer --aidlc-docs /path/to/aidlc-docs

Installation (Claude Code Hook):

# From workspace root
./scripts/aidlc-designreview/tool-install/install-linux.sh      # Linux
./scripts/aidlc-designreview/tool-install/install-mac.sh        # macOS
.\scripts\aidlc-designreview\tool-install\install-windows.ps1   # Windows PowerShell

The installer automatically detects your workspace root and installs the hook to .claude/.

Documentation:

Tenets

These are our core principles to guide our decision making.

  • No duplication. The source of truth lives in one place. If we add support for new tools or formats that require specific files, we generate them from the source rather than maintaining separate copies.

  • Methodology first. AI-DLC is fundamentally a methodology, not a tool. Users shouldn’t need to install anything to get started. That said, we’re open to convenience tooling (scripts, CLIs) down the road if it helps users adopt or extend the methodology.

  • Reproducible. Rules should be clear enough that different models produce similar outcomes. We know models behave differently, but the methodology should minimize variance through explicit guidance.

  • Agnostic. The methodology works with any IDE, agent, or model. We don’t tie ourselves to specific tools or vendors.

  • Human in the loop. Critical decisions require explicit user confirmation. The agent proposes, the human approves.

Prerequisites

Have one of our supported platforms/tools for Assisted AI Coding installed:

PlatformInstallation Link
KiroInstall
Kiro CLIInstall
Amazon Q Developer IDE PluginInstall
Cursor IDEInstall
Cline VS Code ExtensionInstall
Claude Code CLIInstall
GitHub CopilotInstall + Chat

Troubleshooting

General Issues

ProblemSolution
Rules not loadingCheck file exists in the correct location for your platform
File encoding issuesEnsure files are UTF-8 encoded
Rules not applied in sessionStart a new chat session after file changes
Rule details not loadingVerify .aidlc-rule-details/ exists with subdirectories

Platform-Specific Issues

Amazon Q Developer / Kiro

  • Use /context show to verify rules are loaded
  • Check .amazonq/rules/ or .kiro/steering/ directory structure

Cursor

  • For “Apply Intelligently”, ensure a description is defined in frontmatter
  • Check Cursor Settings → Rules to ensure the rule is enabled
  • If rule is too large (>500 lines), split into multiple focused rules

Cline

  • Check the Rules popover under the chat input field
  • Toggle rule files on/off as needed using the popover UI

Claude Code

  • Use /config command to view current configuration
  • Ask “What instructions are currently active in this project?”

GitHub Copilot

  • Select Configure Chat (gear icon) > Chat Instructions to verify instructions are loaded
  • Type /instructions in the chat input to view active instruction files
  • Check that .github/copilot-instructions.md exists in your workspace root

File Path Issues on Windows

  • Use forward slashes / in file paths within markdown files
  • Windows paths with backslashes may not work correctly

Version Control Recommendations

Commit to repository:

# These should be version controlled
CLAUDE.md
AGENTS.md
.amazonq/rules/
.amazonq/aws-aidlc-rule-details/
.kiro/steering/
.kiro/aws-aidlc-rule-details/
.cursor/rules/
.clinerules/
.github/copilot-instructions.md
.aidlc-rule-details/

Optional - Add to .gitignore (if needed):

# Local-only settings
.claude/settings.local.json

Generated aidlc-docs/ Reference

For the complete reference of all documentation artifacts generated by the AI-DLC workflow, see docs/GENERATED_DOCS_REFERENCE.md.

Experimental: AI-Assisted Setup (Release Download)

ResourceLink
AI-DLC Method Definition PaperPaper
AI-DLC Methodology BlogAWS Blog
AI-DLC Open-source Launch BlogAWS Blog
AI-DLC Example Walkthrough BlogAWS Blog
Amazon Q Developer DocumentationDocs
Kiro CLI DocumentationDocs
Cursor Rules DocumentationDocs
Claude Code DocumentationGitHub
GitHub Copilot DocumentationDocs
Working with AI-DLC (interaction patterns and tips)docs/WORKING-WITH-AIDLC.md
Contributing GuidelinesCONTRIBUTING.md
Code of ConductCODE_OF_CONDUCT.md

Additional Resources

ResourceLink
AI-DLC Method Definition PaperPaper
AI-DLC Methodology BlogAWS Blog
AI-DLC Open-source Launch BlogAWS Blog
AI-DLC Example Walkthrough BlogAWS Blog
Amazon Q Developer DocumentationDocs
Kiro CLI DocumentationDocs
Cursor Rules DocumentationDocs
Claude Code DocumentationGitHub
GitHub Copilot DocumentationDocs
Working with AI-DLC (interaction patterns and tips)docs/WORKING-WITH-AIDLC.md
Contributing GuidelinesCONTRIBUTING.md
Code of ConductCODE_OF_CONDUCT.md

Contributing

See CONTRIBUTING for more information.

License

This library is licensed under the MIT-0 License. See the LICENSE file.

相似文章

@Xudong07452910: 这篇论文很适合所有重度使用 Claude Code、Codex 或者其他AI Agent 的人看。 它研究的不是 Agent 在 benchmark 上怎么失败，而是一个更真实的问题： 在真实开发里，AI coding agent 到底是…

X AI KOLs Timeline

This paper analyzes 20,574 real-world coding-agent sessions to identify how AI agents misalign with developer intent, finding that constraint violations and inaccurate self-reporting are the most common failure modes, imposing trust and effort costs rather than irreversible damage.

@teach_fireworks: AI Coding 现在开始进入一个很有意思的阶段。 过去大家讨论最多的是模型能力、上下文长度、Agent Loop、Tool Use、自动化编程，但真正把 Agent 长时间放进真实开发环境之后，很多团队发现问题已经不只是“能不能生成代…

X AI KOLs Timeline

介绍开源工具 re_gent，它为 AI 编程 Agent 提供运行时级别的版本控制和可观测性基础设施，解决 Agent 长时间运行后的代码溯源与审计问题。

@GitHub_Daily: 用 AI 智能体生产级事情，写代码、跑流程、调接口，一开始还行，但规模一大就容易失控，权限太宽、上下文丢失、调试无从下手。 于是找到了 agents-best-practices 这套完整的智能体运行框架设计指南，不限于编码场景，运营、销…

X AI KOLs Timeline

介绍了 agents-best-practices 仓库，这是一份生产级 AI 智能体运行框架设计指南，涵盖工具权限分级、上下文压缩等，支持 Codex 和 Claude Code 安装。

@yibie: 每个程序员都应该自己手写一个 agent。 只要 50 行代码，很有趣，而且会让你惊讶。但为了获得最大收获，请做这两件事： 第一，从零手写。打开空白文本文件，自己敲完每一行代码。不要用任何 AI，连自动补全都不要。 第二，只依赖标准库文档…

X AI KOLs Timeline

本文鼓励程序员亲手手写一个简单的AI agent（约50行代码），以加深对agent工作机制的理解，并建议从零手工编写、只依赖标准库和API文档。