Skills

Skills are structured capability definitions that describe what an AI agent can do. They follow the Anthropic skill structure format, making them compatible with Claude and other AI systems.

What is a Skill?

A skill is a reusable capability definition that includes:

Name - Unique identifier for the skill
Description - What the skill does (used for AI reasoning)
Definition - Structured parameters and schema (JSON Schema)
Examples - Sample inputs/outputs for better AI understanding

Why Use Skills?

Portability - Same skill works across ChatGPT, Claude, custom agents
Consistency - Standardized format ensures reliable behavior
Reusability - Create once, use everywhere
Community - Browse and add skills from the public repository

Skill Structure

{
  "name": "search_codebase",
  "description": "Search for code patterns, function definitions, or file contents in the project codebase",
  "definition": {
    "type": "object",
    "properties": {
      "query": {
        "type": "string",
        "description": "The search query or pattern to look for"
      },
      "file_type": {
        "type": "string",
        "description": "Filter by file extension (e.g., 'ts', 'py', 'md')",
        "optional": true
      },
      "max_results": {
        "type": "integer",
        "description": "Maximum number of results to return",
        "default": 10
      }
    },
    "required": ["query"]
  },
  "examples": [
    {
      "input": { "query": "handleSubmit", "file_type": "tsx" },
      "output": "Found 3 matches in src/components/..."
    }
  ]
}

Demonstrations

A skill can point at recordings of itself being performed, declared in its SKILL.md frontmatter under demonstrations:. For a robot foundation model that learns a task from one or a handful of demonstrations, this is the executable core of the skill rather than an illustration beside it — and the order of the list is the order of execution.

demonstrations:
  - provider: youtube
    ref: dQw4w9WgXcQ
    segments:
      - { start: 134, end: 158, label: grip the handle, comment: from below }

A text-only reader gets the same recordings as a timestamped list, so nothing is lost either way. See Demonstrations.

Creating Skills

Via Dashboard

Navigate to your playbook
Click "Add Skill"
Fill in the skill details
Test with example inputs
Save and optionally publish to community

Via API

POST /api/playbooks/:guid/skills
Authorization: Bearer apb_live_xxx
Content-Type: application/json

{
  "name": "my_custom_skill",
  "description": "Does something useful",
  "definition": { ... }
}

Anthropic Compatibility

AgentPlaybooks skills are designed to be directly compatible with Anthropic's tool use format:

# Python example with Claude
import anthropic

# Fetch skills from AgentPlaybooks
skills_response = requests.get("https://apbks.com/api/playbooks/YOUR_GUID?format=anthropic")
tools = skills_response.json()["tools"]

# Use with Claude
client = anthropic.Anthropic()
response = client.messages.create(
    model="claude-sonnet-4-20250514",
    max_tokens=1024,
    tools=tools,  # Skills from AgentPlaybooks
    messages=[{"role": "user", "content": "Search for handleSubmit"}]
)

OpenAI GPT Compatibility

Skills are also exportable in OpenAPI format for ChatGPT Custom GPTs:

GET /api/playbooks/:guid?format=openapi

This returns an OpenAPI 3.0 specification that can be used as a GPT Action.

Public Skill Repository

Browse community-contributed skills at apbks.com/explore:

Coding - Code search, refactoring, documentation
Writing - Grammar check, summarization, translation
Data - Data analysis, visualization, conversion
Automation - Task scheduling, notifications, integrations
Research - Web search, fact-checking, citations

Best Practices

Clear descriptions - AI uses this to decide when to use the skill
Specific parameters - Well-defined schemas reduce errors
Examples - Include diverse examples for better AI understanding
Atomic skills - Each skill should do one thing well
Version carefully - Changing schemas may break existing integrations

Claude Coworker Compatible Skills

Claude Coworker is Anthropic's desktop AI agent that can organize files, convert documents, and automate workflows. Skills from AgentPlaybooks work directly with Coworker via MCP.

File Organization Skill

{
  "name": "organize_files",
  "description": "Organize files in a folder according to rules based on file type, date, and naming patterns",
  "input_schema": {
    "type": "object",
    "properties": {
      "source_folder": {
        "type": "string",
        "description": "Path to folder containing files to organize"
      },
      "rules": {
        "type": "string",
        "description": "Organization rules in natural language (e.g., 'PDFs to Documents/PDFs, images to Photos by month')"
      },
      "dry_run": {
        "type": "boolean",
        "description": "Preview changes without moving files",
        "default": true
      }
    },
    "required": ["source_folder"]
  }
}

Document Formatting Skill

{
  "name": "format_document",
  "description": "Convert raw notes or data into formatted documents following company templates",
  "input_schema": {
    "type": "object",
    "properties": {
      "input_path": {
        "type": "string",
        "description": "Path to source document"
      },
      "template": {
        "type": "string",
        "enum": ["meeting_notes", "weekly_report", "project_brief", "custom"],
        "description": "Template to apply"
      },
      "output_format": {
        "type": "string",
        "enum": ["markdown", "html", "pdf"],
        "default": "markdown"
      }
    },
    "required": ["input_path", "template"]
  }
}

Report Generation Skill

{
  "name": "generate_report",
  "description": "Generate structured reports from data files with optional charts and executive summary",
  "input_schema": {
    "type": "object",
    "properties": {
      "data_sources": {
        "type": "array",
        "items": {"type": "string"},
        "description": "Paths to data files (CSV, Excel, JSON)"
      },
      "report_type": {
        "type": "string",
        "enum": ["weekly", "monthly", "quarterly", "custom"]
      },
      "include_charts": {
        "type": "boolean",
        "description": "Include data visualizations",
        "default": true
      },
      "executive_summary": {
        "type": "boolean",
        "description": "Generate executive summary section",
        "default": true
      }
    },
    "required": ["data_sources", "report_type"]
  }
}

Using Skills with Coworker

Export your playbook skills for Claude Coworker:

# Export skills in Anthropic format
curl -s "https://apbks.com/api/playbooks/YOUR_GUID?format=anthropic" \
  | jq '.tools' > ~/Documents/CoworkerSkills/my_skills.json

Connecting over MCP is the better route, because skills and memory stay live instead of being copied once. Cowork reads plugins and connectors from your claude.ai account: install the AgentPlaybooks plugin from Customize → Plugins, or add https://agentplaybooks.ai/api/mcp/YOUR_GUID as a custom connector — see AgentPlaybooks in Claude. claude_desktop_config.json does not work for this: it holds local stdio servers only, and a url there is ignored without an error.