framework-distillation

English | 简体中文

Turn open-source frameworks into architecture playbooks for humans and AI.

An AgentSkills-compatible skill for reverse engineering framework architecture, runtime, extension points, and tradeoffs.

Extract architecture, not code. This project helps another AI or team reproduce similar architecture without copying implementation.

Why use it

Use framework-distillation when you want to:

• Understand how a framework is really structured beyond README and public API
• Reverse engineer runtime model, lifecycle, state machine, and extension points
• Reproduce similar architecture in another language or domain
• Compare frameworks by architecture and tradeoffs instead of feature checklists
• Create AI-readable knowledge packages for future implementation work

Core strengths:

• Architecture-first — focuses on system shape, module boundaries, runtime, and tradeoffs
• Evidence-aware — separates observed facts from inference and handles weak evidence explicitly
• Reproduction-oriented — produces outputs that help another AI or team build something similar

Who this is for

• AI coding agent users working with AgentSkills-style skill systems
• Platform engineers designing internal runtimes or orchestration systems
• Architects studying mature open-source frameworks
• Teams building agent runtimes, workflow engines, plugin systems, or similar infrastructure

Quick example

Example prompt:

Analyze LangGraph using framework-distillation.
Goal: extract architecture knowledge for building an enterprise agent runtime in Java.
Depth: deep.
Focus: runtime, graph execution, checkpoint, state model, extension mechanism, production readiness.

Expected outcome:

• Architecture overview
• Runtime and lifecycle model
• State and data model analysis
• Extension mechanism analysis
• Design decisions and tradeoffs
• Reproduction guide
• AI simulation prompt

Installation

This repository follows the AgentSkills / SKILL.md format.

Claude Code

Place this project under your Claude Code skills directory using the folder name framework-distillation.

Codex / local agents using ~/.agents/skills

mkdir -p ~/.agents/skills
cd ~/.agents/skills
git clone https://github.com/yezhwi/framework-distillation.git framework-distillation

Generic AgentSkills-compatible setup

Any skill runner that supports SKILL.md-based skills can load this repository as a skill directory:

framework-distillation/
├── SKILL.md
└── references/

If your platform requires a different install location, copy this directory into that platform's configured skills path.

How to use

Minimum input:

• Repository URL or local source path

Recommended input:

• Framework name
• Analysis goal
• Target implementation language or domain
• Depth: mini, standard, or deep
• Focus areas such as runtime, state machine, plugins, storage, scheduler, or production readiness

Output modes

• mini — fast first-pass architecture summary
• standard — full architecture extraction for most studies
• deep — exhaustive reverse engineering for framework reproduction

See references/output-modes.md for full output shapes.

How to use the output

After generating a distillation:

• Start with the repository summary and architecture sections for a human overview
• Use the knowledge package, reproduction guide, and AI simulation prompt as the main inputs for AI-assisted implementation
• Use the design decisions, runtime, state, and extension sections before refactoring or comparing frameworks

See references/deliverables.md for a detailed usage guide.

Examples

See the example prompts in examples/ for framework-specific starting points.

• examples/langgraph.md — agent runtime, graph execution, checkpoint, state model
• examples/temporal.md — workflow engine, durable execution, recovery, scheduling
• examples/vscode-extensions.md — plugin system, extension lifecycle, compatibility

For a quick overview, start with examples/README.md.

What the skill covers

The skill helps analyze:

• Repository purpose and target users
• Architecture and module boundaries
• Runtime and execution flow
• Lifecycle and state machine
• Core data model
• Extension mechanism
• Engineering decisions and tradeoffs
• Performance and production readiness
• Reusable architecture principles
• Reproduction roadmap

See references/analysis-checklist.md for full checklist.

When to use it

Good fit:

• Reverse engineering framework architecture
• Studying source code for reusable architectural patterns
• Building a similar framework in another language
• Architecture-level framework comparisons

Not good fit:

• README summary only
• API tutorial or usage guide
• File-by-file code walkthrough
• Copy-paste implementation guide

Repository structure

framework-distillation/
├── SKILL.md
├── README.md
├── README.zh-CN.md
├── CONTRIBUTING.md
├── LICENSE
├── social-preview.svg
├── examples/
│   ├── README.md
│   ├── README.zh-CN.md
│   ├── langgraph.md
│   ├── temporal.md
│   └── vscode-extensions.md
└── references/
    ├── analysis-checklist.md
    ├── deliverables.md
    ├── output-modes.md
    ├── self-test.md
    └── simulated-dialogue-tests.md

Important files

• SKILL.md — main skill entrypoint
• README.md — project overview, installation, examples, and brand copy
• CONTRIBUTING.md — contribution workflow and review expectations
• LICENSE — Apache License 2.0
• examples/README.md — example prompt overview
• references/analysis-checklist.md — detailed analysis phases and quality gate
• references/output-modes.md — mini / standard / deep output structures
• references/deliverables.md — deliverable layout, usage guide, and AI simulation prompt template
• references/self-test.md — documented loopholes and fixes
• references/simulated-dialogue-tests.md — reusable pressure-test prompts
• social-preview.svg — source asset for GitHub social preview images

Contributing

Contributions welcome, especially in these areas:

• Better prompts and examples
• Stronger output templates
• More framework-specific test scenarios
• Improved installation guidance for different agent platforms
• Sharper wording for discovery and usability

Before opening a PR:

• Keep SKILL.md concise and discovery-friendly
• Put heavy reference material in references/
• Preserve architecture-first, evidence-aware behavior
• Update self-test or simulated test docs when changing behavior

Project status

Early open-source package. Structure and guidance are stable enough for use, but wording, examples, and platform instructions will continue to improve.

Brand assets

Available assets

• social-preview.svg — default GitHub social preview source

Recommended copy

Tagline

Turn open-source frameworks into architecture playbooks for humans and AI.

Short description

An AgentSkills-compatible skill for reverse engineering framework architecture, runtime, extension points, and tradeoffs.

GitHub About description

Turn open-source frameworks into architecture playbooks for humans and AI.

Social preview title

Framework Distillation

Social preview subtitle

Reverse-engineer framework architecture, runtime, extension points, and tradeoffs.

Suggested usage

• Use the tagline in README hero sections and project summaries
• Use the GitHub About description for repository metadata
• Export social-preview.svg to PNG for GitHub social preview images
• Reuse the social preview title and subtitle in release posts or documentation cards

Consistency notes

• Keep messaging architecture-first, not feature-first
• Prefer reverse engineering, runtime, state, extension points, and tradeoffs over generic AI wording
• Reuse Extract architecture, not code. where a short principle line helps
• Avoid framing the project as an API tutorial or code-copying guide

Sponsor

If this project helps you, you can support it here.

License

Licensed under the Apache License 2.0.