Context Course

0. Welcome to the Context Course

1. Skills: Portable Knowledge for Code Agents

2. MCP: The Model Context Protocol

3. Plugins: Bundling Tools for Distribution

4. Subagents: Orchestrating Multi-Agent Workflows

5. Hooks: Observing and Guarding the Agent Lifecycle

6. Bonus: Nano Harness — Build Your Own Agent

Introduction to Nano Harness The Agentic Loop Deep Dive Tools and Sandboxing in Detail Hands-On: Extending Nano Harness Quiz: Nano Harness and Agent Internals

Join the Hugging Face community

and get access to the augmented documentation experience

Collaborate on models, datasets and Spaces

Faster examples with accelerated inference

Switch between documentation themes

to get started

Quiz: Nano Harness and Agent Internals

Test your understanding of agent loops, tools, and sandboxing.

Call model → return result immediately Call model → parse Python code → execute with tools → observe → repeat until final_answer() Model returns JSON → parse → store in database → done Keep calling the model until it returns the word 'done'

The model outputs English text like 'The answer is 42' The model outputs Python code that calls tools; agent executes that code in a sandbox The model outputs JSON which the agent parses The model outputs commands like 'list_dir' and the agent translates them

The message history grows with every loop iteration (model output + observation) The message history is cleared every step to save tokens The message history is stored in a database for later analysis The message history doesn't matter; only the latest model output matters

safe_path() allows reading any file on the system safe_path() prevents directory traversal by checking that paths stay within WORKSPACE safe_path() is optional; security isn't really needed safe_path() encrypts paths for transmission

exec_cmd() allows any shell command to run exec_cmd() maintains a whitelist of safe commands; anything else is blocked exec_cmd() has a timeout to prevent hanging commands exec_cmd() runs commands with admin privileges

Update on GitHub

←Hands-On: Extending Nano Harness

Quiz: Nano Harness and Agent Internals