The Agent Loop

[ s01 ] s02 > s03 > s04 > s05 > s06 | s07 > s08 > s09 > s10 > s11 > s12

"One loop & Bash is all you need" -- one tool + one loop = an agent.

Harness layer: The loop -- the model's first connection to the real world.

Problem

A language model can reason about code, but it can't touch the real world -- can't read files, run tests, or check errors. Without a loop, every tool call requires you to manually copy-paste results back. You become the loop.

Solution

+--------+      +-------+      +---------+
|  User  | ---> |  LLM  | ---> |  Tool   |
| prompt |      |       |      | execute |
+--------+      +---+---+      +----+----+
                    ^                |
                    |   tool_result  |
                    +----------------+
                    (loop until stop_reason != "tool_use")

One exit condition controls the entire flow. The loop runs until the model stops calling tools.

How It Works

1. User prompt becomes the first message.

type ToolInput = Record<string, any>;

type ToolSpec = {
  name: string;
  description: string;
  input_schema: Record<string, unknown>;
};

const tool: ToolSpec = {
  name: "bash",
  description: "agent loop",
  input_schema: { type: "object", properties: {} }
};

async function handleS01Step(input: ToolInput) {
  await runCommand(input.command);
  return tool.name;
}

1. Send messages + tool definitions to the LLM.

type ToolInput = Record<string, any>;

type ToolSpec = {
  name: string;
  description: string;
  input_schema: Record<string, unknown>;
};

const tool: ToolSpec = {
  name: "bash",
  description: "agent loop",
  input_schema: { type: "object", properties: {} }
};

async function handleS01Step(input: ToolInput) {
  await runCommand(input.command);
  return tool.name;
}

1. Append the assistant response. Check stop_reason -- if the model didn't call a tool, we're done.

type ToolInput = Record<string, any>;

type ToolSpec = {
  name: string;
  description: string;
  input_schema: Record<string, unknown>;
};

const tool: ToolSpec = {
  name: "bash",
  description: "agent loop",
  input_schema: { type: "object", properties: {} }
};

async function handleS01Step(input: ToolInput) {
  await runCommand(input.command);
  return tool.name;
}

1. Execute each tool call, collect results, append as a user message. Loop back to step 2.

type ToolInput = Record<string, any>;

type ToolSpec = {
  name: string;
  description: string;
  input_schema: Record<string, unknown>;
};

const tool: ToolSpec = {
  name: "bash",
  description: "agent loop",
  input_schema: { type: "object", properties: {} }
};

async function handleS01Step(input: ToolInput) {
  await runCommand(input.command);
  return tool.name;
}

Assembled into one function:

type ToolInput = Record<string, any>;

type ToolSpec = {
  name: string;
  description: string;
  input_schema: Record<string, unknown>;
};

const tool: ToolSpec = {
  name: "bash",
  description: "agent loop",
  input_schema: { type: "object", properties: {} }
};

async function handleS01Step(input: ToolInput) {
  await runCommand(input.command);
  return tool.name;
}

That's the entire agent in under 30 lines. Everything else in this course layers on top -- without changing the loop.

What Changed

Component	Before	After
Agent loop	(none)	while True + stop_reason
Tools	(none)	bash (one tool)
Messages	(none)	Accumulating list
Control flow	(none)	stop_reason != "tool_use"

Try It

cd learn-claude-code
tsx agents/s01_s01_agent_loop.ts

1. Create a file called hello.ts that prints "Hello, World!"
2. List all TypeScript files in this directory
3. What is the current git branch?
4. Create a directory called test_output and write 3 files in it