agntrick-add-agent

name	agntrick-add-agent
description	Scaffold a new agntrick agent with agent.py, prompt.md, and test file
disable-model-invocation	true

Add Agent Skill

Scaffold a new agntrick agent with all required files.

Steps

1. Ask the user for:

   • Agent name (kebab-case, e.g. my-agent)
   • Description (one-line summary)
   • MCP servers (optional, e.g. ["toolbox"])
   • Tool categories (optional, e.g. ["web", "code"])

2. Create src/agntrick/agents/{name}.py using this template:

from typing import Any, Sequence

from agntrick.agent import AgentBase
from agntrick.prompts import load_prompt
from agntrick.registry import AgentRegistry


@AgentRegistry.register("{name}", mcp_servers=[{mcp_servers}], tool_categories=[{tool_categories}])
class {ClassName}Agent(AgentBase):
    """Agent description here."""

    @property
    def system_prompt(self) -> str:
        return load_prompt("{name}")

    def local_tools(self) -> Sequence[Any]:
        return []

Where {ClassName} is the name converted to PascalCase (e.g. my-agent -> MyAgent).

3. Create src/agntrick/prompts/{name}.md with a minimal system prompt:

You are a {description}.

Your role is to assist users with {description}.

4. Create tests/test_{name}_agent.py with basic tests:

"""Tests for the {name} agent."""
from agntrick.agents.{name} import {ClassName}Agent


def test_agent_class_exists():
    """Test that the agent class can be imported."""
    assert {ClassName}Agent is not None


def test_agent_has_correct_name():
    """Test that the agent is registered with the correct name."""
    # The agent is registered via @AgentRegistry.register decorator
    agent_cls = {ClassName}Agent
    assert agent_cls is not None

5. Export the agent from src/agntrick/agents/__init__.py if needed.

6. Run make check && make test to verify everything works.

Reference

Existing agents follow this pattern:

• assistant.py — generalist with MCP toolbox
• developer.py — code exploration with local tools
• committer.py — git commit automation
• learning.py — educational content
• news.py — news aggregation
• ollama.py — Ollama-backed
• youtube.py — YouTube transcript extraction
• github_pr_reviewer.py — PR review automation