in-the-loop

in-the-loop

Stay engaged while Claude Code works. in-the-loop is a Claude Code plugin that presents contextual quiz questions about the code being written — both when Claude stops and while it's actively working.

Why

When an AI assistant is writing code for you, it's easy to zone out. But you still own that code — you'll debug it, maintain it, and extend it. in-the-loop keeps you thinking about what's happening by asking short, relevant quiz questions tied to the actual work being done.

How It Works

in-the-loop installs two Claude Code hooks:

• Stop hook — asks a question when Claude finishes a task, based on the full transcript (tools used, files touched, commands run)
• PostToolUse hook — asks a question mid-task after meaningful tool uses (Edit, Write, Bash), based on the specific action that just happened

Questions are generated by Claude Haiku using context from the current session. If generation times out (15s for Stop, 12s for PostToolUse), a relevant fallback question is shown instead.

Questions appear as system messages in the Claude Code UI — they don't block or interrupt the workflow.

Install

From a Claude Code session:

/plugin install from /path/to/in-the-loop

Or once published to a marketplace:

/plugin install in-the-loop

Configuration

Use the built-in skill to view or change settings:

/in-the-loop:config
/in-the-loop:config --frequency often
/in-the-loop:config --difficulty advanced

Frequency

Controls how often questions appear. The intervals differ between Stop and PostToolUse since tool uses are much more frequent:

Setting	On Stop	On Tool Use
every	every stop	every 5th tool use
often	every 2nd	every 8th
sometimes (default)	every 3rd	every 12th
rarely	every 5th	every 20th

Difficulty

• beginner — fundamental concepts, assumes learning programmer
• intermediate (default) — design decisions, trade-offs, best practices
• advanced — edge cases, security, performance, architecture

Example Question

After Claude runs git checkout -b feature/auth:

┌───────────────────────────────────────────────┐
│  🧠 In The Loop                               │
├───────────────────────────────────────────────┤
│                                               │
│  Why create a feature branch instead of       │
│  committing directly to main?                 │
│                                               │
│  A) Main doesn't support new files            │
│  B) To isolate changes and enable code        │
│     review before merging  ✓                  │
│  C) Git requires branches for commits         │
│  D) Feature branches are faster               │
│                                               │
│  ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─  │
│  Branches isolate work so it can be           │
│  reviewed and tested independently.           │
└───────────────────────────────────────────────┘

Project Structure

.claude-plugin/
  plugin.json                     Plugin manifest
hooks/
  hooks.json                      Hook definitions (Stop + PostToolUse)
skills/
  config/
    SKILL.md                      /in-the-loop:config command
src/
  types.ts                        TypeScript interfaces
  hooks/
    on-stop.ts                    Stop hook handler
    on-tool-use.ts                PostToolUse hook handler
  core/
    frequency-controller.ts       When to show questions
    question-generator.ts         AI question generation via Haiku
    question-formatter.ts         Terminal box-drawing UI
    state-manager.ts              Config and session state persistence
    transcript-reader.ts          Claude Code transcript parsing
  prompts/
    question-system-prompt.ts     System prompts by difficulty/mode
tests/
  frequency-controller.test.ts
  question-formatter.test.ts
  question-generator.test.ts
  transcript-reader.test.ts

State is stored in ~/.in-the-loop/ (config + per-session counters).

Development

npm install
npm run build
npm test

License

MIT