1. Define the necessary tools using the `@tool` decorator or `tool()` function, providing clear descriptions and schemas.
2. Configure the agent using `create_agent()`, specifying the model, tools, and system prompt.
3. Implement middleware for custom flows, such as human approval or error handling, using `HumanInTheLoopMiddleware` or `@wrap_tool_call`.
4. Add a checkpointer for state persistence, such as `MemorySaver()`, and configure a `thread_id` to maintain conversation history.
5. Set a `recursion_limit` in the invoke config to prevent infinite loops.
6. Invoke the agent with the user's message and any necessary configuration options.
7. Process the agent's response, extracting the relevant information or structured output.
SynapticSkills