SKILLशिल्प (SKILLSHILP)

Build Better Agent Skills.

Skillshilp is an open-source collection of meta-skills for creating, reviewing, and maintaining production-quality Agent Skills that follow the official Agent Skills specification.

Rather than simply generating SKILL.md files, skillshilp applies proven software engineering principles—such as modularity, progressive disclosure, single responsibility, maintainability, and token efficiency—to help produce skills that remain easy to understand, evolve, and reuse.

Features

• 🏗️ Design production-quality Agent Skills from ideas, prompts, and specifications
• 🛠️ Review and refactor existing Agent Skills while preserving their intent
• 📦 Encourage modular architecture and progressive disclosure
• 🔍 Detect common design issues and architectural anti-patterns
• 📚 Include concise reference material for validation and best practices
• 🧪 Provide lightweight portable Bash validation for basic skill structure
• ⚡ Optimized for low token usage without sacrificing capability

Included Skills

1. skillshilp-create

Designs new production-quality Agent Skills from ideas, requirements, reusable prompts, or specifications. It's architectural and generative: it designs new skills and collections from first principles.

2. skillshilp-edit

Reviews, refactors, and improves existing Agent Skills while preserving their intended behaviour and architecture. It's incremental and conservative: it evolves existing skills while preserving their contracts.

Installation

npx skills

npx skills@latest add ViGi-P/skillshilp

GitHub Skills

gh skill install ViGi-P/skillshilp

When to Use

Use skillshilp-create when you want to:

• create a new Agent Skill
• convert a reusable prompt into a skill
• design a modular skill collection
• architect a new skill from requirements

Use skillshilp-edit when you want to:

• improve an existing skill
• refactor skill architecture
• modernize documentation
• simplify prompts
• review a skill against current best practices

Repository Structure

skillshilp/
└── skills/
    ├── skillshilp-create/
    │   ├── SKILL.md
    │   ├── README.md
    │   ├── scripts/
    │   │   └── validate-skill.sh
    │   └── references/
    │       ├── constraints.md
    │       ├── patterns.md
    │       └── skill-smells.md
    │
    └── skillshilp-edit/
        ├── SKILL.md
        ├── README.md
        ├── scripts/
        │   └── validate-skill.sh
        └── references/
            ├── constraints.md
            ├── patterns.md
            └── skill-smells.md

Source Layout vs. Skill Discovery

The skills/ directory is the canonical source and distribution directory for the skills in this repository.

The Agent Skills specification defines the structure of an individual skill, not the parent directory in which skills must be stored. A skill is a directory containing SKILL.md and optional supporting resources.

When installed into an agent project, a skill may be placed in the host's supported discovery directory. .agents/skills/ is a widely adopted cross-client convention, so an installed skill may look like:

.agents/
└── skills/
    └── skillshilp-create/
        ├── SKILL.md
        ├── references/
        └── scripts/

This does not mean .agents/skills/ is required for every Agent Skill. Individual clients may use different discovery locations.

In other words:

• skills/ in this repository = source/distribution layout
• <skill-name>/ = the Agent Skill itself
• .agents/skills/ = one commonly adopted installation/discovery location

Philosophy

Every Agent Skill should strive to be:

• Reusable
• Discoverable
• Modular
• Composable
• Maintainable
• Deterministic
• Token-efficient

skillshilp is built around these principles and treats Agent Skills as software components rather than one-off prompts.

Contributing

Discussions, issues, suggestions, and pull requests are welcome.

If you discover opportunities to improve skill architecture, progressive disclosure, maintainability, or token efficiency, please open an issue or submit a pull request.