Agent skill

custom-tool-builder

Create custom tools with sandboxed Python code. Use when you need to define new tools dynamically without modifying the codebase.

View SKILL.md on GitHub Repository
Stars 1
Forks 0

Install this agent skill to your Project

npx add-skill https://github.com/binome-dev/humcp/tree/main/src/tools/builder

SKILL.md

Tool Builder

Create and manage custom tools with sandboxed Python execution.

Overview

The tool builder allows you to create new tools by providing Python code that runs in a secure sandbox. Custom tools can be enabled to become first-class MCP/REST tools.

Creating a Custom Tool

python
result = await tool_builder_create(
    name="text_processor",
    description="Process text with various transformations",
    parameters={
        "text": {"type": "string", "description": "Input text"},
        "uppercase": {"type": "boolean", "default": False}
    },
    code='''
def execute(text: str, uppercase: bool = False) -> dict:
    result = text.upper() if uppercase else text.lower()
    return {"success": True, "data": {"result": result}}
''',
    category="custom"
)

Code Requirements

  1. Must define an execute function
  2. Function must return a dict with success and data/error keys
  3. Code runs in a sandboxed environment with limited capabilities

Testing a Custom Tool

Before enabling, test your tool:

python
result = await tool_builder_test(
    name="text_processor",
    params={"text": "Hello World", "uppercase": True}
)
# Returns: {"success": True, "data": {"tool_name": "text_processor", "result": {...}}}

Enabling/Disabling Tools

Tools are created disabled. Enable to make them available via MCP/REST:

python
# Enable - registers with MCP
await tool_builder_enable(name="text_processor")

# Disable - unregisters from MCP
await tool_builder_disable(name="text_processor")

Managing Custom Tools

python
# List all custom tools
result = await tool_builder_list()

# Get tool details
result = await tool_builder_get(name="text_processor")

# Update tool code or description
result = await tool_builder_update(
    name="text_processor",
    description="Updated description",
    code="def execute(): return {'success': True, 'data': {}}"
)

# Delete a tool
result = await tool_builder_delete(name="text_processor")

Sandbox Environment

Allowed

  • Basic Python: variables, functions, loops, conditionals
  • Type constructors: list, dict, set, str, int, float, etc.
  • Utility functions: len, range, sorted, min, max, sum, etc.
  • Allowed imports: json, re, math, datetime

Blocked

  • File system access: open, os, pathlib
  • Network access: socket, requests, httpx
  • System calls: subprocess, os.system
  • Dangerous builtins: exec, eval, compile, __import__
  • Module imports (except allowlist)

Execution Limits

  • Timeout: 60 seconds maximum execution time
  • Memory: No hard limit (Python process limit applies)

Example: JSON Transformer

python
await tool_builder_create(
    name="json_transform",
    description="Transform JSON data by extracting specific fields",
    parameters={
        "data": {"type": "object", "description": "Input JSON object"},
        "fields": {"type": "array", "description": "Fields to extract"}
    },
    code='''
def execute(data: dict, fields: list) -> dict:
    result = {field: data.get(field) for field in fields if field in data}
    return {"success": True, "data": result}
'''
)

Example: Text Statistics

python
await tool_builder_create(
    name="text_stats",
    description="Calculate statistics about text",
    parameters={
        "text": {"type": "string", "description": "Input text"}
    },
    code='''
def execute(text: str) -> dict:
    words = text.split()
    return {
        "success": True,
        "data": {
            "char_count": len(text),
            "word_count": len(words),
            "line_count": text.count("\\n") + 1,
            "avg_word_length": sum(len(w) for w in words) / len(words) if words else 0
        }
    }
'''
)

Response Format

All tools return:

json
{
  "success": true,
  "data": {
    "name": "my_tool",
    "description": "...",
    "code": "...",
    "parameters": {...},
    "category": "custom",
    "enabled": false,
    "created_at": "2026-01-03T...",
    "updated_at": "2026-01-03T..."
  }
}

On error:

json
{
  "success": false,
  "error": "Error description"
}

Recommended Agent Skills

Expand your agent's capabilities with these related and highly-rated skills.

binome-dev/humcp

llm

Generate text, call tools, and get structured JSON output using LLM providers (Claude, OpenAI, Gemini, Ollama). Use when the user needs to call an LLM API for text generation, function calling, or structured data extraction.

1 0
Explore
binome-dev/humcp

calendar-tools

Manages calendar bookings and video conferencing. Use when the user needs to check availability, create bookings on Cal.com, or schedule and manage Zoom meetings.

1 0
Explore
binome-dev/humcp

http-api-client

Make HTTP requests to any URL or API endpoint. Use when the user needs to call a REST API, fetch data from a URL, send webhooks, or test HTTP endpoints. Supports GET, POST, PUT, PATCH, DELETE methods with custom headers and JSON body.

1 0
Explore
binome-dev/humcp

processing-data

Processes CSV files and pandas DataFrames. Use when working with CSV files, tabular data, spreadsheets, or when the user asks to query, analyze, or manipulate structured data.

1 0
Explore
binome-dev/humcp

persistent-memory

Store, search, and retrieve persistent memories and conversation history using Mem0 or Zep. Use when the user needs to remember facts, maintain context across sessions, or manage conversational memory.

1 0
Explore
binome-dev/humcp

storage

Use these tools for S3-compatible object storage operations with MinIO

1 0
Explore

Didn't find tool you were looking for?

Be as detailed as possible for better results