Tool

Claude Technical Tutor

Overview

A Socratic technical tutor for Claude Code with structured session note-taking.

Claude Technical Tutor is a Claude Code plugin that changes how Claude interacts with you during technical work. Instead of solving problems for you, it asks guiding questions, identifies broken assumptions, and names the underlying concepts — the way a good teacher would. It also captures structured notes as you work, so the things you learn don't evaporate when the session ends.

Why this exists: Most AI coding tools optimise for speed — they give you the answer and move on. This plugin optimises for learning. It's built for lab work, exam prep, and anyone who wants to actually understand what they're building rather than just shipping it.

Design Philosophy

Claude never does work for you unless explicitly asked. The default is guidance, not answers.
Explicit opt-in for recording. No notes are captured unless you start a session. Keeps Q&A clean and builds the habit of intentionally entering work mode.
Curation over capture. Claude filters for signal in real-time. Failures and challenges are weighted highest — they're the most valuable learning material.
Works with any markdown note tool. Obsidian, Logseq, OneNote, or just a folder. Output is plain .md files.

Installation

As a Plugin

claude plugin install claude-technical-tutor@wblv-dev

Manual

Clone the repo and copy into your Claude Code config directories:

git clone https://github.com/wblv-dev/claude-technical-tutor.git

cp -r claude-technical-tutor/skills/* ~/.claude/skills/
cp -r claude-technical-tutor/commands/* ~/.claude/commands/
cp -r claude-technical-tutor/rules/* ~/.claude/rules/

Then merge the CLAUDE.md contents into your own ~/.claude/CLAUDE.md and add the hooks from hooks/hooks.json to your ~/.claude/settings.json.

Configuration

Set your notes directory in ~/.claude/settings.json:

{
  "env": {
    "TUTOR_NOTES_DIR": "/path/to/your/notes"
  }
}

Defaults to ~/tutor-notes/ if not set.

Teaching Mode

When teaching mode is active (always on by default), Claude:

Asks 1–2 guiding questions before answering
Identifies the specific assumption that broke when you're wrong
Names the underlying concept — standards, RFCs, framework terms
Flags security issues in anything you build

Disable for a single turn with "just tell me" or "skip the teaching".

Session Note-Taking

The build-log system captures structured notes using a Task / Reason / Challenge rubric:

Task — what did we do?
Reason — why did we do it?
Challenge — what broke, and how did we fix it?

Every entry gets a signal level (high / medium / low) that drives synthesis priority. High-signal challenges — broken assumptions, debugging breakthroughs — surface first.

Commands

/project-start [topic]	Start a recording session with an optional topic
/note [hint]	Manually capture a T/R/C entry for the current moment
/recover	Retroactively generate notes from the session transcript
/project-end	Synthesize all entries into a final note and clean up

Workflow

/project-start Kubernetes networking lab
  ... work normally, Claude captures notes silently ...
  /note            (manual capture if Claude misses something)
/project-end       (synthesize, confirm filename, done)

If you forget /project-start:

... work normally, no notes captured ...
/recover           (mines session transcript retroactively)
/project-end       (synthesize as normal)

Components

The plugin uses multiple Claude Code extension points working together:

CLAUDE.md	Socratic teaching persona and core principles
Skill	Build-log capture logic, T/R/C rubric, signal levels, synthesis
Commands	Session lifecycle: start, note, recover, end
Hook	Warns on exit if a session wasn't wrapped
Rule	Format enforcement on note files

What Gets Captured

Claude watches for noteworthy moments and filters out noise:

Captured: new concepts, decisions with tradeoffs, debugging breakthroughs, broken assumptions, reusable patterns, gotchas that cost time.

Filtered: routine tool calls, typo fixes, dead-ends that taught nothing, things obvious from code or commit history.

Security

Notes never contain secrets, API keys, tokens, or credentials
Claude logs the shape of fixes ("rotated API key, updated env var"), not values
Security issues in your work are flagged proactively — hardcoded credentials, disabled TLS verification, injection vectors