Skills add-backend-tool
🔧

add-backend-tool

Safe 🌐 Network access📁 Filesystem access⚙️ External commands

Add Backend Tools for AI Agents

Extending AI agent capabilities requires understanding backend tool patterns. This skill provides step-by-step instructions for adding new tools to the OpenAI function calling system, including helper functions, tool definitions, and security guardrails.

Supports: Claude Codex Code(CC)
📊 71 Adequate
1

Download the skill ZIP

2

Upload in Claude

Go to Settings → Capabilities → Skills → Upload skill

3

Toggle on and start using

Test it

Using "add-backend-tool". Add a tool to list directory contents

Expected outcome:

  • Step 1: Read backend/main.py to understand existing tool patterns
  • Step 2: Create helper function list_directory with path parameter
  • Step 3: Add tool definition to tools list with proper parameters schema
  • Step 4: Add handler in WebSocket message loop for list_directory
  • Step 5: Configure mode restrictions - full access in Agent mode, read-only safe in Chat mode

Using "add-backend-tool". Create a tool for git operations

Expected outcome:

  • Step 1: Review existing helper functions in backend/main.py
  • Step 2: Create git_run_command helper with command and args parameters
  • Step 3: Add tool definition for git_run_command with command string parameter
  • Step 4: Add WebSocket handler for git_run_command invocation
  • Step 5: Set Agent mode full access - git not safe for Chat mode

Using "add-backend-tool". Add a tool that connects to my PostgreSQL database

Expected outcome:

  • Step 1: Analyze existing database tool patterns in backend/main.py
  • Step 2: Create db_query helper function with connection params and query
  • Step 3: Add tool definition with host, database, query parameters
  • Step 4: Implement WebSocket handler for db_query function calls
  • Step 5: Configure Chat mode restrictions - read-only queries only

Security Audit

Safe
v5 • 1/16/2026

Pure documentation skill containing only instructions and guidance. No executable code, no file operations, no network calls, no command execution. All 23 static findings are false positives from pattern-matching on documentation code examples. The scanner misidentified YAML front matter, JSON metadata fields, markdown code blocks, and security documentation as security vulnerabilities.

2
Files scanned
239
Lines analyzed
3
findings
5
Total audits
Audited by: claude View Audit History →

Quality Score

38
Architecture
100
Maintainability
85
Content
30
Community
100
Security
91
Spec Compliance

What You Can Build

Extend agent capabilities

Add new backend tools to enable AI agents to perform custom operations like file processing or API calls.

Integrate external services

Create tools that connect AI agents to external services, databases, or APIs for agent interactions.

Design tool frameworks

Establish patterns and security guardrails for tool development across teams or organizations.

Try These Prompts

Basic tool addition
Add a tool to list directory contents
Git integration
Create a tool for git operations
Web scraping
Add web scraping capability
Custom backend tool
Add a tool that connects to my PostgreSQL database and runs read-only queries

Best Practices

  • Always include try-except error handling in helper functions
  • Add timeouts for long-running operations to prevent agent hanging
  • Document the tool purpose clearly in the function docstring

Avoid

  • Hardcoding API keys or secrets in tool implementations
  • Adding tools without reviewing security implications for Chat mode
  • Skipping error handling for external service calls

Frequently Asked Questions

Does this skill work with all Claude versions?
This skill works with Claude, Codex, and Claude Code. Tool patterns may vary slightly between versions.
What are the limits on tool complexity?
Tools should be focused and single-purpose. Complex operations should be broken into multiple tools.
How do I integrate with existing MCP tools?
Review backend/main.py for existing MCP integration patterns. Follow the same structure for consistency.
Is my data safe when using this skill?
This skill only provides instructions. It does not access files, execute code, or send data externally.
Why is my new tool not working?
Check that you added both the helper function and the tool definition. Verify the handler in WebSocket loop.
How is this different from MCP tools?
This skill adds backend tools to the function calling system. MCP tools are separate plugin-like integrations.

Developer Details

File structure

📄 SKILL.md