One question at a time.Tuned to the learner.

An agent skill that calibrates to you, reads the reasoning behind each answer, and keeps the record on disk.

install

$npx skills add xpressabhi/tutor

Just ask

Plain language, no syntax. Openers from first session to deep end:

tutor I want to become an AI engineerNew topic. Mission first, then a short calibration.
tutor quiz me on SQL joinsA focused assessment: questions climb until they find your ceiling.
tutor I forgot where I was with RustReads the ledger, then probes. Old levels are claims, not evidence.
tutor resume AI engineeringRestart any previous session without losing anything: the ledger, logs, and notes are all on disk.
tutor I have a system design interview on FridayThe date sets the mission. Every question aims at the interview.
tutor quiz me on this codebaseReads the actual code and asks about what it really does.
tutor what does this snippet print?Runs it first. The answer comes from output it actually saw.
tutor teach me medieval history, purely for pleasureNon-technical subjects run on the same loop.
tutor let me explain RAG to you, poke holesExplain-back: you talk, it finds the gaps in the explanation.
tutor teach me TCP states, draw the machineIdeas that need to be seen get a small diagram, at your level.
tutor I've got 20 minutes before standupShort bursts count. The ledger holds the thread between sessions.

The loop

Every session runs the same way: one question per turn, and the ledger is written before the next question is chosen.

ask

One question, chosen from the ledger and the last answer.
read

The answer, for the reasoning and not just the result.
move

Up on a clean correct, held on partial, down on a miss.
record

The ledger row — including the re-probe it owes — and the session log, written before the next question.

5edge cases, tradeoffs, where the rule breaks down
4predict non-obvious behaviour, or choose between approaches
3apply in a novel case, or explain why (current)
2apply in a familiar case
1recall a term or fact

Rungs are relative to the learner, not the subject.

What it does

calibrates: Opening questions find the ceiling, not the average.
adapts: Right answer, clean reasoning: up. Guess, hedge, or miss: hold or drop.
re-teaches: A miss gets the idea underneath, then a fresh question at the same rung.
resumes: The ledger is read, then probed. A recorded level is a claim, not evidence.
listens: Explain-back: you explain it, it finds the gap in what you left out.
steers: Tangents are followed, then re-anchored. Wandering is not being behind.
grounds: Questions from your docs, code, or book cite where each one came from.
runs code: Snippets are verified against real output; questions are built from observed results.
draws: One small HTML artifact, when an idea needs to be seen.

State is just files

Everything lives in plain markdown under ~/tutor/. Read it, edit it, diff it, or keep it in git.

~/tutor/
  ai-engineer/
    TOPIC.md             mission, outline, how you learn
    LEDGER.md            one row per concept: rung, verdict, next probe
    sessions/
      2026-10-08.md      every turn, plus the close: what's in flight
    sources/             only when you have material to ground in
    artifacts/
      attention-r4.html  drawn at the learner's rung
  LEARNER.md             cross-topic: how you learn, carried over

Works with

OpenCode, Claude Code, Cursor, Codex, and any agent that reads SKILL.md. Optional capabilities degrade instead of failing.