Custom AI Agents with Kilo Code

Define a JSON schema for a custom trading function (e.g., `get_historical_data`) that an LLM can invoke during a conversation.

Implement a complete tool-calling loop in Python: send a request with a tool schema, detect the tool-use response, execute the local function, and return the result to the LLM.

Describe the core agent loop (Observe-Reason-Act) and the ReAct pattern, explaining how explicit reasoning steps improve task success and reliability.

Compare and contrast single-agent vs. multi-agent architectures, outlining the communication and coordination challenges in multi-agent systems for trading workflows.

Describe Kilo Code's built-in agent modes (e.g., Orchestrator, Code) and explain how it manages context through its internal tool set (file reading, terminal commands).

Create a custom Kilo Code mode for a specific trading task by defining a tailored system prompt, specifying allowed tools, and setting file access permissions.

Explain the purpose of the Model Context Protocol (MCP) in standardizing communication between LLM agents and external tool servers.

Implement a basic MCP server to expose a custom trading tool (e.g., a backtesting endpoint) and connect it to a Kilo Code custom mode.

Design a multi-agent workflow for a trading task, specifying the distinct role, persona, tools, and hand-offs for each agent (e.g., Scientist, Strategy Coder, Backtester).

Implement the multi-agent pipeline using custom Kilo Code modes and MCP tools, executing the workflow sequentially to generate and validate a trading strategy.

Establish evaluation criteria for the agent team's output and perform one iteration of refinement by adjusting prompts, tools, or workflow design to improve performance.