erpnext-frappe（erpnext 冰沙）

Business Claw Skills

High-level business workflows that combine multiple MCP tools into reusable, executable skills for ERPNext.

Overview

Skills are pre-defined workflows stored as JSON files in definitions/. Each skill defines:

• Triggers: Natural language patterns that activate the skill
• Tools: MCP tools to execute in sequence
• Input Schema: Required and optional parameters
• Workflow Steps: Ordered execution plan with variable substitution
• Guardrails: Validation rules for safe execution
• Output Template: Formatted response message

Available Skills

CRM Skills

Sales Skills

Purchase Skills

Inventory Skills

Project Skills

Financial Skills

Utility Skills

Usage

Loading Skills

from bc_skills import get_available_skills, load_skill

# List all available skills
skills = get_available_skills()
print(skills)  # ['create_customer', 'create_sales_order', ...]

# Load a specific skill
skill = load_skill("create_customer")

Executing Skills

from bc_skills.loader import execute_skill

result = execute_skill(
    name="create_customer",
    context={
        "customer_name": "ACME Corp",
        "customer_type": "Company",
        "customer_group": "Commercial",
        "email": "contact@acme.com"
    },
    user="Administrator"
)

print(result)

Trigger Examples

Skills respond to natural language triggers:

Skill Definition Schema

{
  "name": "skill_name",
  "version": "1.0.0",
  "description": "What the skill does",
  "author": "Business Claw Team",
  "category": "crm|sales|purchase|inventory|project|payments|utility",
  
  "triggers": [
    "trigger phrase 1",
    "trigger phrase 2"
  ],
  
  "tools": [
    {
      "name": "tool_name",
      "description": "What it does",
      "required": true
    }
  ],
  
  "input_schema": {
    "type": "object",
    "properties": {
      "param_name": {
        "type": "string",
        "description": "Parameter description",
        "enum": ["option1", "option2"]
      }
    },
    "required": ["required_param"]
  },
  
  "workflow": {
    "steps": [
      {
        "step": "step_name",
        "tool": "tool_to_call",
        "arguments": {
          "doctype": "DocType",
          "data": {
            "field": "${variable}"
          }
        }
      }
    ]
  },
  
  "guardrails": {
    "rule_name": true
  },
  
  "output_template": "Formatted output {{variable}}"
}

Variable Substitution

Workflow steps support ${variable} substitution from execution context:

{
  "step": "create_order",
  "tool": "create_document",
  "arguments": {
    "doctype": "Sales Order",
    "data": {
      "customer": "${customer_id}",
      "items": "${items}"
    }
  }
}

Creating Custom Skills

1. Create a JSON file in definitions/
2. Define triggers, tools, input schema, and workflow
3. Use the SkillLoader to load and execute

Example custom skill structure:

{
  "name": "my_custom_skill",
  "version": "1.0.0",
  "description": "My custom workflow",
  "category": "utility",
  "triggers": ["my trigger"],
  "tools": [
    {"name": "get_doctype_meta", "required": true},
    {"name": "create_document", "required": true}
  ],
  "input_schema": {
    "type": "object",
    "properties": {
      "param1": {"type": "string"}
    },
    "required": ["param1"]
  },
  "workflow": {
    "steps": [
      {
        "step": "step1",
        "tool": "get_doctype_meta",
        "arguments": {"doctype": "Item"}
      }
    ]
  },
  "output_template": "Result: {{result}}"
}

Architecture

• loader.py - SkillLoader class manages skill loading and execution
• definitions/ - JSON files containing skill definitions
• Skills use the ToolRouter to execute MCP tools in sequence
• Guardrails provide validation before skill execution

Requirements

• Frappe/ERPNext environment
• bc_mcp module for tool routing
• JSON or YAML skill definitions

License

MIT