agent-skills-standard A collection of Agent @codeKK typescriptOpen Source Website

agent-skills-standard

Project Url: HoangNguyen0403/agent-skills-standard

Introduction: A collection of Agent Skills Standard and Best Practice for Programming Languages, Frameworks that help our AI Agent follow best practies on frameworks and programming laguages

More: Author ReportBugs OfficialWebsite

Tags:

The open standard for High-Density AI coding instructions. Make your AI smarter, faster, and more consistent across your entire team.

Agent Skills Standard is the open-source framework to distribute, sync, and version-control engineering standards (often called "Cursor Rules" or "Agent Skills") across all major AI agents (Cursor, Claude Code, GitHub Copilot, Windsurf, Trae, Gemini, and more).

📌 Table of Contents

💡 Why use this? (For Managers & Teams)

Think of Agent Skills Standard as a universal instruction manual for your AI assistant.

When using AI for work, you constantly have to remind it of your "house rules" (e.g., "Make sure to handle errors this way" or "Use this specific layout"). If you feed it too many rules at once, it forgets them or wastes expensive API tokens. If you feed it too few, it writes bad code.

This project solves "The Context Wall" by packaging rules into high-density "Skills" that you can plug into any AI tool instantly.

Role	Benefit
🚀 Builders	Stop copy-pasting prompts. Get instant, modular standards directly in your IDE.
🛡️ Architects	Scale engineering excellence. Define "house rules" once; sync them to every agent in the team.
📈 Organizations	Lower the technical barrier. Enable non-experts to produce senior-level output via standard "playbooks."
💰 Finance/Ops	Sending a 3,600+ token architect prompt vs. a 413-token skill reduces API costs by ~9x.

🔄 The Workflow

graph LR
    A[Hardcoded Rules] --> B[Wordy Prompts]
    B --> C[Loss of Logic & High Cost]

    D[Agent Skills] --> E[Progressive Disclosure]
    E --> F[High-Density AI Logic]

    style F fill:#dcfce7,stroke:#166534
    style C fill:#fee2e2,stroke:#991b1b

🛡️ Security & Trust (For IT)

We understand that "injecting" instructions into your AI can sound risky to security teams. Here is how we keep you safe:

No Code Execution: Skills are pure Markdown/JSON files. They contain text instructions for the AI, not executable code. They cannot run commands on your machine.
Continuous Benchmarking: Most skills now include an evals.json dataset, used to verify AI adherence to constraints via automated regression testing.
Open Source: The entire registry is open source. You can audit every skill file on GitHub before using it.
Sandboxed: The CLI tool (agent-skills-standard) runs in user space to download text files. The "skills" themselves run inside the AI's isolated context window, not as OS processes.
Privacy: We do not collect any code or project data. Feedback is only sent if you manually trigger the feedback command or strongly opt-in.

🚀 Quick Start (For Developers)

Consume engineering standards in your project instantly.

1. Initialize Your Project

Run this once to detect your stack and setup your .skillsrc.

npx agent-skills-standard@latest init

2. Sync Standards

Fetch the latest high-density instructions and install them into your hidden agent folders.

npx agent-skills-standard@latest sync

3. Start Coding

The CLI automatically updates AGENTS.md in your project root. This serves as a universal entry point that helps all AI agents (Cursor, Copilot, etc.) proactively understand when to trigger specific skills based on your project context.

[!TIP] View all CLI Commands: Run npx agent-skills-standard@latest --help for commands like validate, feedback, and upgrade.

🌍 Supported Frameworks

The Agent Skills Standard is a modular library. We follow a strict Token Economy—every skill is audited for its footprint in the AI's context window (averaging ~400 tokens) to save you money.

Click to view all 20+ supported frameworks and stacks

Stack / Category	Key Modules	Version	Typical Saving	Skills
Common Patterns	Accessibility, Best Practices, Security	`v1.10.0`	83%	29
Flutter	BLoC, Riverpod, Clean Architecture	`v1.6.1`	85%	21
Dart	Language, Tooling	`v1.3.1`	85%	3
TypeScript	Type Safety, Tooling	`v1.3.1`	82%	4
JavaScript	Functional Programming, Patterns	`v1.3.1`	89%	3
React	React 18+, Hooks, Performance	`v1.3.1`	86%	8
React Native	Architecture, Performance	`v1.4.1`	88%	13
NestJS	Architecture, Security, BullMQ	`v1.4.1`	82%	21
Next.js	App Router, SEO, Performance	`v1.4.1`	82%	18
Angular	Architecture, Signals, RxJS	`v1.3.1`	86%	16
Android	Compose, Architecture, Serialization	`v1.3.1`	90%	22
iOS	Architecture, SwiftUI, Concurrency	`v1.4.1`	90%	15
Swift	Concurrency, Architecture	`v1.3.1`	87%	8
Kotlin	Language, Concurrency	`v1.3.1`	89%	4
Java	Language, Concurrency	`v1.3.1`	86%	5
Spring Boot	Architecture, Security	`v1.3.1`	87%	10
Go (Golang)	Clean Architecture, Security	`v1.3.1`	88%	11
PHP	Error Handling, PHP 8+	`v1.3.1`	86%	7
Laravel	Solid Patterns, Clean Architecture	`v1.3.1`	82%	10
Database	PostgreSQL, MongoDB, Redis	`v1.2.2`	85%	3
Quality Engineering	BA, TDD, Zephyr, Automation	`v1.4.0`	81%	4

[!TIP] View the Complete Registry: For a full list of all 160+ individual skills and token metrics, visit the Skills Directory and Benchmark Report.

⚙️ Configuration (`.skillsrc`)

The .skillsrc file allows you to customize how skills are synced to your project.

registry: https://github.com/HoangNguyen0403/agent-skills-standard
agents: [cursor, copilot, kiro]
skills:
  flutter:
    ref: flutter-v1.1.0
    # 🚫 Exclude specific sub-skills from being synced
    exclude: ['getx-navigation']
    # ➕ Include specific skills (supports cross-category 'category/skill' syntax)
    include:
      - 'bloc-state-management'
      - 'react/hooks'
    # 🔒 Protect local modifications from being overwritten
    custom_overrides: ['bloc-state-management']

❓ FAQ & Contributing

Does this work with Cursor or Gemini?

Yes. Agent Skills Standard is designed to be "Agent Agnostic." It generates the specific configuration files (like `.cursorrules` or `.agent/skills`) that these tools read natively.

How is this different from a system prompt?

System prompts are static and often too long. Agent Skills Standard uses a Discovery Bridge. The AI only "looks up" the specific engineering standards it needs for the file you are currently editing, keeping its memory (context window) free for your actual code.

Can I use this for non-coding tasks (e.g., HR, Finance)?

Absolutely. While many skills here are technical, the framework works for any domain where you want an AI agent to follow consistent, high-density instructions.