Skill Creator

Skill Creator

This skill provides guidance for creating effective skills.

About Skills

Skills are modular, self-contained packages that extend Claude's capabilities by providing specialized knowledge, workflows, and tools. Think of them as "onboarding guides" for specific domains or tasks—they transform Claude from a general-purpose agent into a specialized agent equipped with procedural knowledge that no model can fully possess.

IMPORTANT:

• Skills are not documentation, they are practical instructions for Claude Code to use the tools, packages, plugins or APIs to achieve the tasks.
• Each skill teaches Claude how to perform a specific development task, not what a tool does.
• Claude Code can activate multiple skills automatically to achieve the user's request.

What Skills Provide

1. Specialized workflows - Multi-step procedures for specific domains 2. Tool integrations - Instructions for working with specific file formats or APIs 3. Domain expertise - Company-specific knowledge, schemas, business logic 4. Bundled resources - Scripts, references, and assets for complex and repetitive tasks

Anatomy of a Skill

Every skill consists of a required SKILL.md file and optional bundled resources:

.claude/skills/
└── skill-name/
    ├── SKILL.md (required)
    │   ├── YAML frontmatter metadata (required)
    │   │   ├── name: (required)
    │   │   ├── description: (required)
    │   │   ├── license: (optional)
    │   │   └── version: (optional)
    │   └── Markdown instructions (required)
    └── Bundled Resources (optional)
        ├── scripts/          - Executable code (Python/Bash/etc.)
        ├── references/       - Documentation intended to be loaded into context as needed
        └── assets/           - Files used in output (templates, icons, fonts, etc.)

Requirements (IMPORTANT)

• Skill should be combined into specific topics, for example: cloudflare, cloudflare-r2, cloudflare-workers, docker, gcloud should be combined into devops
• SKILL.md should be less than 150 lines and include the references of related markdown files and scripts.
• Each script or referenced markdown file should be also less than 150 lines, remember that you can always split them into multiple files (progressive disclosure principle).
• Descriptions in metadata of SKILL.md files should be both concise (less than 200 characters) and still contains enough usecases of the references and scripts, this will help skills can be activated automatically during the implementation process of Claude Code.
• Referenced markdowns:
• Sacrifice grammar for the sake of concision when writing these files.
• Can reference other markdown files or scripts as well.
• Referenced scripts:
• Prefer nodejs or python scripts instead of bash script, because bash scripts are not well-supported on Windows.
• If you're going to write python scripts, make sure you have requirements.txt
• Make sure scripts respect .env file follow this order: process.env > $HOME/.claude/skills/${SKILL}/.env (global) > $HOME/.claude/skills/.env (global) > $HOME/.claude/.env (global) > ./.claude/skills/${SKILL}/.env (cwd) > ./.claude/skills/.env (cwd) > ./.claude/.env (cwd)
• Create .env.example files to show the required environment variables.
• Always write tests for these scripts.

IMPORTANT:

• Always keep in mind that SKILL.md and reference files should be token consumption efficient, so that progressive disclosure can be leveraged at best.
• SKILL.md should be less than 150 lines
• Referenced markdown files should be also less than 150 lines, remember that you can always split them into multiple files (progressive disclosure principle).
• Referenced scripts: no limit on length, just make sure it works, no compile issues, no runtime issues, no dependencies issues, no environment issues, no platform issues.

Why? Better context engineering: leverage progressive disclosure technique of Agent Skills, when agent skills are activated, Claude Code will consider to load only relevant files into the context, instead of reading all long SKILL.md as before.

SKILL.md (required)

File name: SKILL.md (uppercase) File size: Under 150 lines, if you need more, split it to multiple files (<150 lines each) in references folder. SKILL.md is always short and concise, straight to the point, treat it as a quick reference guide.

Metadata Quality: The name and description (MUST be under 200 characters) in YAML frontmatter determine when Claude will use the skill. Be specific about what the skill does and when to use it, DO NOT sound generic, vague or educational. Use the third-person (e.g. "This skill should be used when..." instead of "Use this skill when...").

Bundled Resources (optional)

Scripts (scripts/)

Executable code (Python/Bash/etc.) for tasks that require deterministic reliability or are repeatedly rewritten.

• When to include: When the same code is being rewritten repeatedly or deterministic reliability is needed
• Example: scripts/rotate_pdf.py for PDF rotation tasks
• Benefits: Token efficient, deterministic, may be executed without loading into context
• Note: Scripts may still need to be read by Claude for patching or environment-specific adjustments

IMPORTANT:

• Write tests for scripts.
• Run tests and make sure it works, if tests fail, fix them and run tests again, repeat until tests pass.
• Run scripts manually with some usecases to make sure it works.
• Make sure scripts respect .env file follow this order: process.env > $HOME/.claude/skills/${SKILL}/.env (global) > $HOME/.claude/skills/.env (global) > $HOME/.claude/.env (global) > ./.claude/skills/${SKILL}/.env (cwd) > ./.claude/skills/.env (cwd) > ./.claude/.env (cwd)

References (references/)

Documentation and reference material intended to be loaded as needed into context to inform Claude's process and thinking.

• When to include: For documentation that Claude should reference while working
• Examples: references/finance.md for financial schemas, references/mnda.md for company NDA template, references/policies.md for company policies, references/api_docs.md for API specifications
• Use cases: Database schemas, best practices, common workflows, cheatsheets, tool instructions, API documentation, domain knowledge, company policies, detailed workflow guides
• Benefits: Keeps SKILL.md lean, loaded only when Claude determines it's needed, makes information discoverable without hogging the context window.
• Best practice: If files are large (>150 lines), split them into multiple files (<150 lines each) in references folder, include grep search patterns in SKILL.md
• Avoid duplication: Information should live in either SKILL.md or references files, not both. Prefer references files for detailed information unless it's truly core to the skill—this keeps SKILL.md lean while making information discoverable without hogging the context window. Keep only essential procedural instructions and workflow guidance in SKILL.md; move detailed reference material, schemas, and examples to references files.

IMPORTANT:

• Referenced markdown files should be also less than 150 lines, remember that you can always split them into multiple files (progressive disclosure principle).
• Referenced markdown files are practical instructions for Claude Code to use the tools, packages, plugins or APIs to achieve the tasks.
• Each skill teaches Claude how to perform a specific development task, not what a tool does.

Assets (assets/)

Files not intended to be loaded into context, but rather used within the output Claude produces.

• When to include: When the skill needs files that will be used in the final output
• Examples: assets/logo.png for brand assets, assets/slides.pptx for PowerPoint templates, assets/frontend-template/ for HTML/React boilerplate, assets/font.ttf for typography
• Use cases: Templates, images, icons, boilerplate code, fonts, sample documents that get copied or modified
• Benefits: Separates output resources from documentation, enables Claude to use files without loading them into context

Progressive Disclosure Design Principle

Skills use a three-level loading system to manage context efficiently:

1. Metadata (name + description) - Always in context (less than 200 characters) 2. SKILL.md body - When skill triggers (<5k words) 3. Bundled resources - As needed by Claude (Unlimited*)

*Unlimited because scripts can be executed without reading into context window.

Skill Creation Process

To create a skill, follow the "Skill Creation Process" in order, skipping steps only if there is a clear reason why they are not applicable.

Step 1: Understanding the Skill with Concrete Examples

Skip this step only when the skill's usage patterns are already clearly understood. It remains valuable even when working with an existing skill.

To create an effective skill, clearly understand concrete examples of how the skill will be used. This understanding can come from either direct user examples or generated examples that are validated with user feedback.

Use AskUserQuestion tool to gather user feedback and validate understanding.

For example, when building an image-editor skill, relevant questions include:

• "What functionality should the image-editor skill support? Editing, rotating, anything else?"
• "Can you give some examples of how this skill would be used?"
• "I can imagine users asking for things like 'Remove the red-eye from this image' or 'Rotate this image'. Are there other ways you imagine this skill being used?"
• "What would a user say that should trigger this skill?"

To avoid overwhelming users, avoid asking too many questions in a single message. Start with the most important questions and follow up as needed for better effectiveness.

Conclude this step when there is a clear sense of the functionality the skill should support.

Step 2: Research on The Internet

An effective skill is a subset of real-life workflows from professional workflows and case studies, so it's important to research on the internet to understand the current state of the art and best practices.

Activate /docs-seeker skill to search for documentation if needed.

If you receive a lot of URLs or files, use multiple WebFetch tools and Explore subagents to explore them in parallel, then report back to main agent.

Activate /research skill to research for:

• Best practices & industry standards
• Existing CLI tools (executable via npx, bunx or pipx) and their usage patterns
• Workflows & success case studies
• Common patterns, use cases & examples
• Edge cases, potential pitfalls & avoidance strategies

Write down the reports from the research to be used in the next step.

Step 3: Planning the Reusable Skill Contents

To turn concrete examples into an effective skill, analyze each example by:

1. Considering how to execute on the example from scratch 2. Prefer existing CLI tools' execution (via npx, bunx or pipx) over writing custom code 3. Identifying what scripts, references, and assets would be helpful when executing these workflows repeatedly 4. Analyze the current skills catalog to avoid duplicating functionality, re-using existing skills where possible.

Example: When building a pdf-editor skill to handle queries like "Help me rotate this PDF," the analysis shows:

1. Rotating a PDF requires re-writing the same code each time 2. A scripts/rotate_pdf.py script would be helpful to store in the skill

Example: When designing a frontend-webapp-builder skill for queries like "Build me a todo app" or "Build me a dashboard to track my steps," the analysis shows:

1. Writing a frontend webapp requires the same boilerplate HTML/React each time 2. An assets/hello-world/ template containing the boilerplate HTML/React project files would be helpful to store in the skill

Example: When building a big-query skill to handle queries like "How many users have logged in today?" the analysis shows:

1. Querying BigQuery requires re-discovering the table schemas and relationships each time 2. A references/schema.md file documenting the table schemas would be helpful to store in the skill

To establish the skill's contents, analyze each concrete example to create a list of the reusable resources to include: scripts, references, and assets.

• Make sure scripts respect .env file follow this order: process.env > $HOME/.claude/skills/${SKILL}/.env (global) > $HOME/.claude/skills/.env (global) > $HOME/.claude/.env (global) > ./.claude/skills/${SKILL}/.env (cwd) > ./.claude/skills/.env (cwd) > ./.claude/.env (cwd)
• Make sure scripts have tests, run all tests and ensure they pass. DO NOT SKIP THIS STEP.

Step 4: Initializing the Skill

At this point, it is time to actually create the skill.

Skip this step only if the skill being developed already exists, and iteration or packaging is needed. In this case, continue to the next step.

When creating a new skill from scratch, always run the init_skill.py script. The script conveniently generates a new template skill directory that automatically includes everything a skill requires, making the skill creation process much more efficient and reliable.

Usage:

scripts/init_skill.py <skill-name> --path <output-directory>

The script:

• Creates the skill directory at the specified path
• Generates a SKILL.md template with proper frontmatter and TODO placeholders
• Creates example resource directories: scripts/, references/, and assets/
• Adds example files in each directory that can be customized or deleted

After initialization, customize or remove the generated SKILL.md and example files as needed.

Step 5: Edit the Skill

When editing the (newly-generated or existing) skill, remember that the skill is being created for another instance of Claude to use. Focus on including information that would be beneficial and non-obvious to Claude. Consider what procedural knowledge, domain-specific details, or reusable assets would help another Claude instance execute these tasks more effectively.

Start with Reusable Skill Contents

To begin implementation, start with the reusable resources identified above: scripts/, references/, and assets/ files. Note that this step may require user input. For example, when implementing a brand-guidelines skill, the user may need to provide brand assets or templates to store in assets/, or documentation to store in references/.

Also, delete any example files and directories not needed for the skill. The initialization script creates example files in scripts/, references/, and assets/ to demonstrate structure, but most skills won't need all of them.

Update SKILL.md

Writing Style: Write the entire skill using imperative/infinitive form (verb-first instructions), not second person. Use objective, instructional language (e.g., "To accomplish X, do Y" rather than "You should do X" or "If you need to do X"). This maintains consistency and clarity for AI consumption.

To complete SKILL.md, answer the following questions:

1. What is the purpose of the skill, in a few sentences? 2. When should the skill be used? 3. In practice, how should Claude use the skill? All reusable skill contents developed above should be referenced so that Claude knows how to use them.

Step 5: Packaging a Skill

Once the skill is ready, it should be packaged into a distributable zip file that gets shared with the user. The packaging process automatically validates the skill first to ensure it meets all requirements:

scripts/package_skill.py <path/to/skill-folder>

Optional output directory specification:

scripts/package_skill.py <path/to/skill-folder> ./dist

The packaging script will:

1. Validate the skill automatically, checking:

• YAML frontmatter format and required fields
• Skill naming conventions and directory structure
• Description completeness and quality (MUST be under 200 characters)
• File organization and resource references

2. Package the skill if validation passes, creating a zip file named after the skill (e.g., my-skill.zip) that includes all files and maintains the proper directory structure for distribution.

If validation fails, the script will report the errors and exit without creating a package. Fix any validation errors and run the packaging command again.

Step 6: Iterate

After testing the skill, users may request improvements. Often this happens right after using the skill, with fresh context of how the skill performed.

Iteration workflow: 1. Use the skill on real tasks 2. Notice struggles or inefficiencies 3. Notice token usage and performance 4. Identify how SKILL.md or bundled resources should be updated 5. Implement changes and test again

Validation Criteria

Detailed validation criteria for evaluating skills:

• Quick checklist: references/validation-checklist.md
• Metadata quality: references/metadata-quality-criteria.md
• Token efficiency: references/token-efficiency-criteria.md
• Script quality: references/script-quality-criteria.md
• Structure & organization: references/structure-organization-criteria.md

Plugin Marketplaces

For distributing skills as plugins via marketplaces, see:

• Overview: references/plugin-marketplace-overview.md
• Schema: references/plugin-marketplace-schema.md
• Sources: references/plugin-marketplace-sources.md
• Hosting: references/plugin-marketplace-hosting.md
• Troubleshooting: references/plugin-marketplace-troubleshooting.md

References

• Agent Skills
• Agent Skills Spec
• Agent Skills Overview
• Best Practices
• Plugin Marketplaces

Metadata Quality Criteria

Metadata determines when Claude activates the skill. Poor metadata = wrong activation or missed activation.

Name Field

Format: kebab-case, lowercase

Good Examples:

• pdf-editor - clear domain
• bigquery-analyst - tool + role
• frontend-webapp-builder - specific function

Bad Examples:

• helper - too generic
• mySkill - wrong case
• pdf - too short, unclear purpose

Description Field

Constraint: Under 200 characters

Purpose: Trigger automatic activation during implementation

Good Descriptions

Specific, action-oriented, includes use cases:

description: Build React/TypeScript frontends with modern patterns. Use for components, Suspense, lazy loading, performance optimization.

description: Process PDFs with rotation, splitting, merging. Use for document manipulation, page extraction, PDF conversion.

Bad Descriptions

Too generic or educational:

description: A skill for working with databases.  # Too vague

description: This skill helps you understand how React works.  # Educational, not actionable

Trigger Precision

Description should answer: "What phrases would a user say that should trigger this skill?"

Example for `image-editor` skill:

• "Remove red-eye from this image"
• "Rotate this photo 90 degrees"
• "Crop the background out"

Include these trigger phrases/actions in description.

Third-Person Style

Correct: "This skill should be used when..." Wrong: "Use this skill when..." or "You should use this..."

Validation

Check with packaging script:

scripts/package_skill.py <skill-path>

Fails if:

• Missing name or description
• Description exceeds 200 characters
• Invalid YAML syntax

Plugin Marketplace Hosting & Distribution

GitHub (Recommended)

1. Create repository for marketplace 2. Add .claude-plugin/marketplace.json with plugin definitions 3. Share: users add with /plugin marketplace add owner/repo

Benefits: Built-in version control, issue tracking, team collaboration.

Other Git Services

GitLab, Bitbucket, self-hosted servers all work:

/plugin marketplace add https://gitlab.com/company/plugins.git

Private Repositories

Manual Installation/Updates

Claude Code uses existing git credential helpers. If git clone works in terminal, it works in Claude Code.

Common credential helpers:

• gh auth login for GitHub
• macOS Keychain
• git-credential-store

Background Auto-Updates

Set authentication token in environment:

export GITHUB_TOKEN=ghp_xxxxxxxxxxxxxxxxxxxx

Require Marketplaces for Team

Add to .claude/settings.json:

{
  "extraKnownMarketplaces": {
    "company-tools": {
      "source": {
        "source": "github",
        "repo": "your-org/claude-plugins"
      }
    }
  },
  "enabledPlugins": {
    "code-formatter@company-tools": true,
    "deployment-tools@company-tools": true
  }
}

Managed Marketplace Restrictions

Admins use strictKnownMarketplaces in managed settings:

Allow Specific Marketplaces Only

{
  "strictKnownMarketplaces": [
    { "source": "github", "repo": "acme-corp/approved-plugins" },
    { "source": "url", "url": "https://plugins.example.com/marketplace.json" }
  ]
}

Allow All from Internal Server (Regex)

{
  "strictKnownMarketplaces": [
    { "source": "hostPattern", "hostPattern": "^github\\.example\\.com$" }
  ]
}

Test Locally Before Distribution

/plugin marketplace add ./my-local-marketplace
/plugin install test-plugin@my-local-marketplace

Plugin Marketplace Overview

Build and host plugin marketplaces to distribute Claude Code extensions across teams and communities.

What is a Plugin Marketplace

A plugin marketplace is a catalog that lets you distribute plugins to others. Marketplaces provide:

• Centralized discovery
• Version tracking
• Automatic updates
• Support for multiple source types (git repositories, local paths, etc.)

Creating & Distributing a Marketplace

1. Create plugins: Build plugins with commands, agents, hooks, MCP servers, or LSP servers 2. Create marketplace file: Define marketplace.json listing plugins and sources 3. Host the marketplace: Push to GitHub, GitLab, or another git host 4. Share with users: Users add with /plugin marketplace add and install individual plugins

Directory Structure

my-marketplace/
├── .claude-plugin/
│   └── marketplace.json       # Required: marketplace catalog
└── plugins/
    └── my-plugin/
        ├── .claude-plugin/
        │   └── plugin.json    # Plugin manifest
        └── skills/
            └── my-skill/
                └── SKILL.md

Marketplace File Location

Create .claude-plugin/marketplace.json in repository root.

User Commands

• Add marketplace: /plugin marketplace add owner/repo or /plugin marketplace add ./local-path
• Install plugin: /plugin install plugin-name@marketplace-name
• Update marketplace: /plugin marketplace update
• Validate: /plugin validate . or claude plugin validate .

Related References

• Schema details: references/plugin-marketplace-schema.md
• Plugin sources: references/plugin-marketplace-sources.md
• Hosting & distribution: references/plugin-marketplace-hosting.md
• Troubleshooting: references/plugin-marketplace-troubleshooting.md

Official Documentation

https://code.claude.com/docs/en/plugin-marketplaces.md

Plugin Marketplace Schema

Marketplace JSON Structure

{
  "name": "company-tools",
  "owner": {
    "name": "DevTools Team",
    "email": "devtools@example.com"
  },
  "metadata": {
    "description": "Brief marketplace description",
    "version": "1.0.0",
    "pluginRoot": "./plugins"
  },
  "plugins": [...]
}

Required Fields

Reserved Names

Cannot use: claude-code-marketplace, claude-code-plugins, claude-plugins-official, anthropic-marketplace, anthropic-plugins, agent-skills, life-sciences. Names impersonating official marketplaces also blocked.

Optional Metadata

Plugin Entry Schema

Required Plugin Fields

Optional Plugin Fields

Component Configuration Fields

Example Plugin Entry

{
  "name": "code-formatter",
  "source": "./plugins/formatter",
  "description": "Automatic code formatting on save",
  "version": "2.1.0",
  "author": { "name": "DevTools Team" },
  "license": "MIT",
  "keywords": ["formatting", "linting"],
  "strict": false
}

Plugin Marketplace Sources

Source Types

Relative Paths (Local)

For plugins in same repository:

{
  "name": "my-plugin",
  "source": "./plugins/my-plugin"
}

Note: Relative paths only work when users add marketplace via Git. For URL-based distribution, use GitHub, npm, or git URL sources.

GitHub Repositories

{
  "name": "github-plugin",
  "source": {
    "source": "github",
    "repo": "owner/plugin-repo"
  }
}

Pin to specific branch, tag, or commit:

{
  "name": "github-plugin",
  "source": {
    "source": "github",
    "repo": "owner/plugin-repo",
    "ref": "v2.0.0",
    "sha": "a1b2c3d4e5f6a7b8c9d0e1f2a3b4c5d6e7f8a9b0"
  }
}

Git Repositories (GitLab, Bitbucket, etc.)

{
  "name": "git-plugin",
  "source": {
    "source": "url",
    "url": "https://gitlab.com/team/plugin.git"
  }
}

Advanced Plugin Entry Example

{
  "name": "enterprise-tools",
  "source": {
    "source": "github",
    "repo": "company/enterprise-plugin"
  },
  "description": "Enterprise workflow automation tools",
  "version": "2.1.0",
  "commands": [
    "./commands/core/",
    "./commands/enterprise/"
  ],
  "agents": ["./agents/security-reviewer.md"],
  "hooks": {
    "PostToolUse": [{
      "matcher": "Write|Edit",
      "hooks": [{
        "type": "command",
        "command": "${CLAUDE_PLUGIN_ROOT}/scripts/validate.sh"
      }]
    }]
  },
  "mcpServers": {
    "enterprise-db": {
      "command": "${CLAUDE_PLUGIN_ROOT}/servers/db-server",
      "args": ["--config", "${CLAUDE_PLUGIN_ROOT}/config.json"]
    }
  },
  "strict": false
}

Key variable: ${CLAUDE_PLUGIN_ROOT} - Use in hooks and MCP configs to reference files within plugin's installation directory (plugins are copied to cache when installed).

Plugin Marketplace Troubleshooting

Marketplace Not Loading

Symptoms: Can't add marketplace or see plugins

Solutions:

• Verify marketplace URL is accessible
• Check .claude-plugin/marketplace.json exists at specified path
• Validate JSON syntax: claude plugin validate . or /plugin validate .
• For private repos, confirm access permissions

Marketplace Validation Errors

Run claude plugin validate . or /plugin validate . from marketplace directory.

Common Errors

Warnings (Non-blocking)

• Marketplace has no plugins defined: Add at least one plugin to plugins array
• No marketplace description provided: Add metadata.description
• Plugin "x" uses npm source...: Use github or local path sources instead

Plugin Installation Failures

Symptoms: Marketplace appears but installation fails

Solutions:

• Verify plugin source URLs are accessible
• Check plugin directories contain required files
• For GitHub sources, ensure repos are public or you have access
• Test plugin sources manually by cloning/downloading

Private Repository Authentication Fails

Manual Installation

• Verify authentication: gh auth status for GitHub
• Check credential helper: git config --global credential.helper
• Try cloning repository manually

Background Auto-Updates

• Verify token set: echo $GITHUB_TOKEN
• Check token permissions (read access to repository)
• GitHub: ensure repo scope for private repos
• GitLab: ensure at least read_repository scope
• Verify token not expired

Relative Paths Fail in URL-based Marketplaces

Symptoms: Added via URL (like https://example.com/marketplace.json), plugins with "./plugins/my-plugin" fail with "path not found"

Cause: URL-based marketplaces only download marketplace.json file, not plugin files.

Solutions:

• Use external sources (GitHub, npm, git URL) instead of relative paths:

  { "name": "my-plugin", "source": { "source": "github", "repo": "owner/repo" } }

• Use Git-based marketplace (clones entire repo, relative paths work)

Files Not Found After Installation

Symptoms: Plugin installs but references to files fail, especially files outside plugin directory

Cause: Plugins copied to cache directory. Paths like ../shared-utils won't work.

Solutions:

• Use symlinks (followed during copying)
• Restructure so shared directory is inside plugin source path
• Reference: Plugin caching and file resolution docs

Script Quality Criteria

Scripts provide deterministic reliability and token efficiency.

When to Include Scripts

• Same code rewritten repeatedly
• Deterministic operations needed
• Complex transformations
• External tool integrations

Cross-Platform Requirements

Prefer: Node.js or Python Avoid: Bash scripts (not well-supported on Windows)

If bash required, provide Node.js/Python alternative.

Testing Requirements

Mandatory: All scripts must have tests

# Run tests before packaging
python -m pytest scripts/tests/
# or
npm test

Tests must pass. No skipping failed tests.

Environment Variables

Respect hierarchy (first found wins):

1. process.env (runtime) 2. $HOME/.claude/skills/<skill-name>/.env (skill-specific) 3. $HOME/.claude/skills/.env (shared skills) 4. $HOME/.claude/.env (global) 5. ./.claude/skills/${SKILL}/.env (cwd) 6. ./.claude/skills/.env (cwd) 7. ./.claude/.env (cwd)

Implementation pattern (Python):

from dotenv import load_dotenv
import os

# Load in reverse order (last loaded wins if not set)
load_dotenv('$HOME/.claude/.env')
load_dotenv('$HOME/.claude/skills/.env')
load_dotenv('$HOME/.claude/skills/my-skill/.env')
load_dotenv('./.claude/skills/my-skill/.env')
load_dotenv('./.claude/skills/.env')
load_dotenv('./.claude/.env')
# process.env already takes precedence

Documentation Requirements

.env.example

Show required variables without values:

API_KEY=
DATABASE_URL=
DEBUG=false

requirements.txt (Python)

Pin major versions:

requests>=2.28.0
python-dotenv>=1.0.0

package.json (Node.js)

Include scripts:

{
  "scripts": {
    "test": "jest"
  }
}

Manual Testing

Before packaging, test with real use cases:

# Example: PDF rotation script
python scripts/rotate_pdf.py input.pdf 90 output.pdf

Verify output matches expectations.

Error Handling

• Clear error messages
• Graceful failures
• No silent errors
• Exit codes: 0 success, non-zero failure

Structure & Organization Criteria

Proper structure enables discovery and maintainability.

Required Directory Layout

.claude/skills/
└── skill-name/
    ├── SKILL.md          # Required, uppercase
    ├── scripts/          # Optional: executable code
    ├── references/       # Optional: documentation
    └── assets/           # Optional: output resources

SKILL.md Requirements

File name: Exactly SKILL.md (uppercase)

YAML Frontmatter: Required at top

---
name: skill-name
description: Under 200 chars, specific triggers
license: Optional
version: Optional
---

Resource Directories

scripts/

Executable code for deterministic tasks.

scripts/
├── main_operation.py
├── helper_utils.py
├── requirements.txt
├── .env.example
└── tests/
    └── test_main_operation.py

references/

Documentation loaded into context as needed.

references/
├── api-documentation.md
├── schema-definitions.md
└── workflow-guides.md

assets/

Files used in output, not loaded into context.

assets/
├── templates/
├── images/
└── boilerplate/

File Naming

Format: kebab-case, descriptive

Good:

• api-endpoints-authentication.md
• database-schema-users.md
• rotate-pdf-script.py

Bad:

• docs.md - not descriptive
• apiEndpoints.md - wrong case
• 1.md - meaningless

Cleanup

After initialization, delete unused example files:

# Remove if not needed
rm -rf scripts/example_script.py
rm -rf references/example_reference.md
rm -rf assets/example_asset.txt

Scope Consolidation

Related topics should be combined into single skill:

Consolidate:

• cloudflare + cloudflare-r2 + cloudflare-workers → devops
• mongodb + postgresql → databases

Keep separate:

• Unrelated domains
• Different tech stacks with no overlap

Validation

Run packaging script to check structure:

scripts/package_skill.py <skill-path>

Checks:

• SKILL.md exists
• Valid frontmatter
• Proper directory structure

Token Efficiency Criteria

Skills use progressive disclosure to minimize context window usage.

Three-Level Loading

1. Metadata - Always loaded (~200 chars) 2. SKILL.md body - Loaded when skill triggers (<150 lines) 3. Bundled resources - Loaded as needed (unlimited for scripts)

Size Limits

SKILL.md Content Strategy

Include in SKILL.md:

• Purpose (2-3 sentences)
• When to use (trigger conditions)
• Quick reference for common workflows
• Pointers to resources (scripts, references, assets)

Move to references/:

• Detailed documentation
• Database schemas
• API specs
• Step-by-step guides
• Examples and templates
• Best practices

No Duplication Rule

Information lives in ONE place:

• Either in SKILL.md
• Or in references/

Bad: Schema overview in SKILL.md + detailed schema in references/schema.md Good: Brief mention in SKILL.md + full schema only in references/schema.md

Splitting Large Files

If reference exceeds 150 lines, split by logical boundaries:

references/
├── api-endpoints-auth.md      # Auth endpoints
├── api-endpoints-users.md     # User endpoints
├── api-endpoints-payments.md  # Payment endpoints

Include grep patterns in SKILL.md for discoverability:

## API Documentation
- Auth: `references/api-endpoints-auth.md`
- Users: `references/api-endpoints-users.md`
- Payments: `references/api-endpoints-payments.md`

Scripts: Best Token Efficiency

Scripts execute without loading into context.

When to use scripts:

• Repetitive code patterns
• Deterministic operations
• Complex transformations

Example: PDF rotation via scripts/rotate_pdf.py vs rewriting rotation code each time.

Skill Validation Checklist

Quick validation before packaging. Run scripts/package_skill.py for automated checks.

Critical (Must Pass)

Metadata

• [ ] name: kebab-case, descriptive
• [ ] description: under 200 characters, specific triggers, not generic

Size Limits

• [ ] SKILL.md: under 150 lines
• [ ] Each reference file: under 150 lines
• [ ] No info duplication between SKILL.md and references

Structure

• [ ] SKILL.md exists with valid YAML frontmatter
• [ ] Unused example files deleted
• [ ] File names: kebab-case, self-documenting

Scripts (If Applicable)

• [ ] Tests exist and pass
• [ ] Cross-platform (Node.js/Python preferred)
• [ ] Env vars: respects hierarchy process.env > $HOME/.claude/skills/${SKILL}/.env (global) > $HOME/.claude/skills/.env (global) > $HOME/.claude/.env (global) > ./.claude/skills/${SKILL}/.env (cwd) > ./.claude/skills/.env (cwd) > ./.claude/.env (cwd)
• [ ] Dependencies documented (requirements.txt, .env.example)
• [ ] Manually tested with real use cases

Quality

Writing Style

• [ ] Imperative form: "To accomplish X, do Y"
• [ ] Third-person metadata: "This skill should be used when..."
• [ ] Concise, no fluff

Practical Utility

• [ ] Teaches how to do tasks, not what tools are
• [ ] Based on real workflows
• [ ] Includes concrete trigger phrases/examples

Integration

• [ ] No duplication with existing skills
• [ ] Related topics consolidated (e.g., cloudflare + docker → devops)
• [ ] Composable with other skills

Automated Validation

Run packaging script to validate:

scripts/package_skill.py <path/to/skill-folder>

Checks performed:

• YAML frontmatter format
• Required fields present
• Description length (<200 chars)
• Directory structure
• File organization

Fix all errors before distributing.

Subagent Delegation Enforcement

When a skill requires subagent delegation (via Task tool):

1. Use MUST language - "Use subagent" is weak; "MUST spawn subagent" is enforceable 2. Include Task pattern - Show exact syntax: Task(subagent_type="X", prompt="Y", description="Z") 3. Add validation rule - "If Task tool calls = 0 at end, workflow is INCOMPLETE" 4. Mark requirements clearly - Use table with "MUST spawn" column 5. Forbid direct implementation - "DO NOT implement X yourself - DELEGATE to subagent"

Anti-pattern (weak):

- Use `tester` agent for testing

Correct pattern (enforceable):

- **MUST** spawn `tester` subagent: `Task(subagent_type="tester", prompt="Run tests", description="Test")`
- DO NOT run tests yourself - DELEGATE