Herdr Game — Development skill for Claude Code

learn-herdr

Interactive browser exercises for learning [herdr](https://herdr.dev), the terminal workspace manager for AI coding agents.

The curriculum contains 34 hands-on lessons covering:

• the prefix key and help overlay;
• tabs, panes, workspaces, focus, zoom, resize, and navigation;
• Herdr's CLI and socket API;
• advanced multi-shortcut workflows;
• agent operations: list, prompt, wait, blocked-state handling, logical keys, get, and read.

Each player gets an isolated Herdr session. Lessons stage the topology they need and validate real Herdr state. The Agents chapter uses deterministic, credential-free `claude` and `codex` practice programs: Herdr still performs its real process detection, pane targeting, lifecycle reporting, prompting, waiting, key delivery, and transcript reading, but no provider account, API key, or paid model inference is used.

Development

Requirements:

• Node.js 20+
• pnpm
• Vercel credentials with Sandbox access, supplied through the environment or local Vercel OIDC configuration

pnpm install
pnpm dev

Open .

A prebuilt Sandbox snapshot is optional. Set `HERDR_SNAPSHOT_ID` to use one; otherwise the app provisions a cold session.

Verification

pnpm test
pnpm lint
pnpm build
pnpm smoke http://localhost:3000
pnpm ui-smoke http://localhost:3000

`pnpm smoke` crosses the real backend boundary. It provisions a session and verifies boot, terminal attach, prompt identity, state snapshots, lesson staging, tab and pane exercises, both practice-agent kinds, prompt/wait transitions, blocked-state recovery, transcript reading, detach, and teardown.

`pnpm ui-smoke` uses `agent-browser` at a 1686×1178 viewport and fails unless wterm has nonzero dimensions and nonblank rendered rows before and after a staged lesson transition. It also asserts that the loading state makes the terminal inert.

Architecture

• components/game.tsx — interactive curriculum and terminal UI
• `lib/lessons.t