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

Unit 5: Hooks Hook Events and the Agent Lifecycle Quiz 1: Hook Fundamentals Hands-On: Agent Activity Dashboard with Gradio Quiz 2: Hooks in Practice

6. Bonus: Nano Harness — Build Your Own Agent

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 1: Hook Fundamentals

Test your understanding of hooks, the events they cover, and how they differ across Claude Code, Codex, OpenCode, and Pi.

Question 1: What is a hook?

A hook is a model prompt that reminds the agent to run a command A hook is a user-defined handler that runs at a deterministic point in the agent's lifecycle, independent of the model's cooperation A hook is a way to package skills, MCP servers, and agents into a distributable bundle A hook is a kind of subagent that runs in parallel with the main agent

Question 2: How do hook surfaces differ across platforms?

Claude Code uses PascalCase JSON events (PreToolUse, Stop); Codex uses a smaller PascalCase JSON set behind a feature flag; OpenCode uses TS/JS plugin modules; Pi uses TS/JS extensions with lower_snake_case events like tool_call All four platforms use the exact same settings.json format for hook events OpenCode exposes hooks through JSON configuration, like Claude Code Only Claude Code supports HTTP hooks; Codex, OpenCode, and Pi cannot send events to an HTTP receiver

Question 3: Which lifecycle events matter most?

PreToolUse fires after the tool returns, so you can log the result UserPromptSubmit fires when the user submits a prompt, before the model sees it — ideal for injecting extra context Stop fires only at session end, not per turn PostToolUse in Codex fires for every possible tool path, just like in Claude Code

Question 4: How do hooks block or modify actions?

A Claude Code command hook exits with code 2 and writes a reason to stderr to block the action A Claude Code hook returns HTTP 500 to block the action An OpenCode hook writes to stdin to block a tool call Hooks cannot block or modify tool calls; they are observe-only

Question 5: When should you use hooks instead of skills?

Hooks replace skills: once you have hooks, you don't need to write skills anymore Hooks are the right surface when you need deterministic behavior — like 'always run the linter after an Edit' — that shouldn't depend on the model remembering Hooks are mainly for user interface theming Hooks only make sense for single-agent workflows

Summary

If you got 4-5 correct, you understand the core hook model well enough to start wiring real guardrails and observability. If not, review the event lifecycle and the platform comparison before continuing.

Key Takeaways

Hooks are deterministic runtime handlers, not prompts or plugins
Claude Code, Codex, OpenCode, and Pi expose similar lifecycle moments with different configuration surfaces
Hooks are the right tool when behavior must happen every time rather than relying on the model to remember

Next Steps

Next, wire hooks into a real dashboard so you can see agent activity live.

Update on GitHub

←Hook Events and the Agent Lifecycle Hands-On: Agent Activity Dashboard with Gradio→

Quiz 1: Hook Fundamentals Question 1: What is a hook?Question 2: How do hook surfaces differ across platforms?Question 3: Which lifecycle events matter most?Question 4: How do hooks block or modify actions?Question 5: When should you use hooks instead of skills?Summary Key Takeaways Next Steps