Study Mode

A Claude Code plugin that turns Claude into a strict tutor instead of a chatbot — and enforces it.

When you ask Claude to help you learn something — finetune a LoRA, pick up Go, run an experiment to understand it — normal Claude just does it. It writes all the code, runs ahead in auto mode, and hands you output you don't understand. You end up with a working repo and zero new skill.

Study Mode flips that. Claude becomes a demanding teacher: it paces one concept at a time, verifies you actually understand before moving on, gathers real resources, reviews your code — and refuses to write code for you, even when you ask.

That refusal isn't a polite suggestion in a system prompt you can argue with. It's a PreToolUse hook that denies Claude's file-writing tools while the mode is on. You can't talk it into doing the work for you. That's the whole point.

Why not just the built-in "Learning" output style?

Claude Code ships a soft Learning style that asks you to fill in some code. It's a persona — insist hard enough and Claude writes the code anyway. Study Mode adds the two things that make it actually work:

1. Hard enforcement — a hook blocks Write/Edit/NotebookEdit, not just a prompt.
2. Checkpoint pacing — a session state file tracks where you are so Claude can't run ahead; it must verify understanding (/quiz) before advancing.

Install

/plugin marketplace add codeprakhar25/claude-study-mode
/plugin install study-mode@study-mode      # choose "user" scope to use it everywhere

Restart Claude Code, then type study on. Requires node on your PATH (the hooks run on node).

git clone https://github.com/codeprakhar25/claude-study-mode
cd claude-study-mode
bash install.sh --statusline   # symlinks into ~/.claude/plugins, wires the status badge (needs jq)

Usage

study on            # activate study mode (strict level)
study lite          # activate, allow ≤3-line pseudocode hints
study off           # back to normal Claude

/study learn Go     # start a learning session — builds a concept path, gathers resources
/quiz               # prove you understand the current concept before moving on
/progress           # on-demand diagram: current concept map + lifetime summary

Levels

In both levels, you write all the code; Claude reads and critiques it but never writes it.

How it works

Read / Grep / Glob / WebSearch are never blocked — so Claude can review your code and pull up canonical docs.

State

All study state lives centrally under $CLAUDE_CONFIG_DIR — nothing is written into your repos, so there's no .study/ dir to gitignore.

• Global toggle: $CLAUDE_CONFIG_DIR/.study-state.json ({active, level}) plus a hardened .study-active sidecar the statusline reads.
• Per-project session: $CLAUDE_CONFIG_DIR/study/sessions/<slug>.json — {topic, plan, checkpoint, passed, cwd}, keyed by project path.
• Lifetime ledger: $CLAUDE_CONFIG_DIR/study/history.jsonl — append-only, one line per concept you pass, across every project. This is what /progress summarizes. Honest record only — concepts, topics, dates; no streaks or XP.

While the mode is on, the central study/ dir is the only place Claude may write (a legacy ./.study/ is still accepted so older sessions keep resuming). Everything else is denied.

License

MIT — see LICENSE.