Skills-Based Plugins: 4 Real-World Examples from the Claude Plugins Community

Skills-based plugins are self-contained capabilities defined in SKILL.md files that Claude uses to execute domain-specific tasks through a declarative markdown specification and MCP (Model-Control-Protocol) endpoints.

The anthropics/claude-plugins-community repository hosts a curated collection of reusable skills that extend Claude's abilities for specialized domains. Each skill lives inside a plugin folder and follows a strict architecture separating manifest metadata from execution logic.

How Skills-Based Plugins Are Structured

Every skills-based plugin in the repository shares a common layout:

Component	Purpose	Location
Plugin manifest (plugin.json)	Declares plugin name, description, permissions, and skill list	/<plugin-name>/.claude-plugin/plugin.json
Marketplace manifest (marketplace.json)	Supplies Claude Marketplace metadata (icons, pricing, categories)	/<plugin-name>/.claude-plugin/marketplace.json
Skill definition (SKILL.md)	Human-readable spec with steps, validation, and GraphQL calls	/<plugin-name>/skills/<skill-name>/SKILL.md
Reference assets	Supporting docs, model cards, and pipelines	/<plugin-name>/skills/<skill-name>/references/*
Tests / examples	Optional end-to-end test scripts	/<plugin-name>/skills/<skill-name>/tests/*

When Claude receives a user request, it scans all SKILL.md files for matching description fields, loads the corresponding skill, executes its prescribed steps, and calls the plugin's MCP endpoints using embedded GraphQL snippets.

Example 1: Financial Data Onboarding with tres-wallets-upload

The tres-wallets-upload skill from the tres-finance-plugin demonstrates complex multi-step workflows for enterprise crypto finance.

Located at tres-finance-plugin/skills/tres-wallets-upload/SKILL.md, this skill handles uploading on-chain wallets or exchange accounts into the Tres Finance platform. It implements:

• Validation steps for wallet identifiers and parent platforms
• Preview mode showing what will change before execution
• Batch mutations via GraphQL calls to https://ai.tres.finance/mcp

{
  "skill": "tres-wallets-upload",
  "parameters": {
    "wallet_type": "on_chain",
    "input_method": "manual",
    "wallets": [
      {
        "name": "Treasury Hot",
        "identifier": "0xABCD1234EF567890...",
        "parentPlatform": "ETHEREUM",
        "tags": ["defi", "treasury"]
      },
      {
        "name": "Cold Storage",
        "identifier": "bc1qxyz9876...",
        "parentPlatform": "BITCOIN"
      }
    ],
    "confirm": true
  }
}

Claude reads the skill definition, runs validation steps, then executes the GraphQL batch mutation defined in the Execute Batch Upload section.

Example 2: AI Media Generation with quickdesign

The quickdesign skill showcases skills-based plugins for creative automation, generating UGC videos, image edits, and upscales through model selection rules.

Defined in quickdesign/skills/quickdesign/SKILL.md, this skill enforces cardinal generation rules:

1. Always show a plan summary before execution
2. Require explicit user confirmation when auto_mode is false
3. Select appropriate models (e.g., seedance-2.0-r2v for 30-second talking-avatar videos)

{
  "skill": "quickdesign",
  "parameters": {
    "model": "seedance-2.0-r2v",
    "references": {
      "product": "/mnt/user-data/uploads/product.jpg",
      "avatar": "/mnt/user-data/uploads/avatar.jpg"
    },
    "script": "Hey, meet our new product! It keeps your coffee hot for 12 hours.",
    "duration_seconds": 30,
    "resolution": "1080p",
    "auto_mode": false
  }
}

After confirmation, Claude runs the CLI command quickdesign video generate … and optionally applies the auto-subtitle pipeline. The skill references model cards stored at quickdesign/models/seedance-2.0-r2v.md.

Example 3: Simple Question Answering with eli5

The eli5 skill demonstrates minimal-overhead skills-based plugins that require no external MCP calls.

Located at eli5/skills/eli5/SKILL.md, this single-step skill forwards the user's question to Claude with an "Explain like I'm 5" instruction and returns a plain-language answer.

{
  "skill": "eli5",
  "parameters": {
    "question": "Why do we need sleep?"
  }
}

This example proves that skills-based plugins can be as simple as a prompt template—no GraphQL, no validation, no external APIs.

Example 4: Testing Infrastructure with testdino-sessions

The testdino-sessions skill shows how skills-based plugins support developer tooling and quality assurance.

Found at testdino/skills/testdino-sessions/SKILL.md, this skill manages test sessions for the TestDino platform, demonstrating that the SKILL.md architecture works for infrastructure and debugging workflows, not just end-user features.

How Skills Are Discovered and Executed

The runtime flow for all skills-based plugins follows four stages:

1. Intent matching — Claude scans description fields across all SKILL.md files for semantic matches to user queries.

2. Step execution — Claude follows ordered steps in the skill definition (collect data → validate → preview → confirm → execute).

3. MCP resolution — Steps requiring remote data use GraphQL snippets embedded in SKILL.md, resolved against the plugin's MCP endpoint.

4. Result handling — Claude surfaces validation errors, success counts, or warnings per the skill's Result handling section.

Repository Validation and CI

The anthropics/claude-plugins-community repository enforces quality through .github/workflows/validate-plugins.yml, which validates every SKILL.md against schema invariants. This ensures all skills-based plugins remain stable as the ecosystem grows.

Adding a new capability requires only dropping a new SKILL.md into an existing plugin or creating a new plugin folder with the proper manifest structure.

Summary

• Skills-based plugins are defined by SKILL.md files that specify execution flow, validation, and MCP calls in human-readable markdown.
• tres-wallets-upload demonstrates complex financial workflows with batch mutations and preview modes.
• quickdesign shows creative automation with model selection rules and cardinal safety checks.
• eli5 proves skills can be minimal prompt templates with zero external dependencies.
• testdino-sessions extends the pattern to developer tooling and testing infrastructure.
• All skills are validated by CI and discovered through semantic matching against their description fields.

Frequently Asked Questions

What makes a plugin "skills-based" versus other plugin architectures?

A skills-based plugin centers on declarative SKILL.md files that Claude interprets directly, rather than requiring custom code execution. Each skill specifies its own workflow steps, validation rules, and GraphQL calls in markdown, making the behavior transparent and editable without redeploying software.

How does Claude know which skill to invoke for a user request?

Claude scans the description field in every SKILL.md file for semantic similarity to the user's intent. The description is a free-text sentence like "Upload on-chain wallets or exchange accounts to Tres Finance."

Can I create a skills-based plugin without external APIs or MCP endpoints?

Yes. The eli5 skill requires no MCP calls—it simply reformulates the user's question and returns a response. Any skill that needs only Claude's built-in capabilities can omit GraphQL snippets and external endpoints entirely.

{
  "mcpServers": {
    "instagit": {
      "command": "npx",
      "args": ["-y", "instagit@latest"]
    }
  }
}