| name | tutorkit |
| description | Teach a topic across many sessions, with one learning repo holding a folder per topic, lessons pitched at what you already know, and spaced retrieval that makes it stick. Use when the user says "teach me X", "tutor me on X", "I want to learn X", "explain how X works", "quiz me on what I learned", "what's due for review", "test me", "am I ready", "exam me on X", "place me on X", "where am I with my learning", "what am I studying", "learning status", "show my progress", or runs "/tutorkit". Tuned for software engineering topics and works for any other. |
| license | MIT |
| disable-model-invocation | true |
| allowed-tools | Bash, Read, Write, Edit, Grep, Glob, WebSearch, WebFetch, AskUserQuestion, Task, Agent |
| metadata | {"internal":false} |
tutorkit
Teach a topic over many sessions and remember what stuck. tutorkit keeps one learning repo with a folder per topic, pitches each lesson at the edge of what the user already knows, writes the lesson as a printable HTML artifact, and schedules retrieval cues that come due later. A tutor is someone who remembers you and adapts; that persistence is the whole design.
Four postures, not four presentations: explain answers and writes nothing, lesson teaches and is the only mode that opens a track, drill tests recall across topics, exam measures and refuses to teach.
status is a fifth mode and not a fifth posture. It teaches nothing, asks nothing, and grades nothing. It reads the two router files, prints where every track stands, and crowns the next move. Run it to decide which of the four postures to run.
What tutorkit is not
- Not a code writer. It never implements a feature in the user's project. It reads their code to build examples and stops there.
- Not a decision report. "Which queue library should we use" is a research question, even though both read primary sources.
- Not project documentation. It documents nothing about the user's repo.
- Not a spaced-repetition engine. A fixed five-step interval ladder, no SM-2 or FSRS tuning, no Anki export, no retention modelling. The schedule exists to make review happen, not to be optimal.
- Not general-purpose by design. It works for any topic and it is tuned for software engineering, where repo grounding, runnable exercises, and source-code citations are all sharper.
- It never commits the learning repo on its own.
Mode selection
This is the skill's first decision, and the triggers overlap on purpose. "Explain how X works" is explain and "teach me X" is lesson, but the wording is a weak signal. Resolve on intent rather than phrasing: a question that wants an answer gets explain, a request that wants to end up knowing the thing gets lesson. "What's due", "quiz me", "test me" get drill. "Am I ready", "how much do I actually know", "exam me" get exam. "Where am I", "what am I learning", "learning status", "show my progress" get status.
An ask that names no topic is almost always status. That is the sharpest divider in the whole set: the other four modes need a slug and status refuses one. "Am I ready" names a topic and measures it, so it is exam; "where am I" names none and reports every track, so it is status.
When the ask is genuinely ambiguous, take the cheap branch: run explain and offer the track at the end. Guessing lesson costs a mission interview the user did not want; guessing explain costs one extra sentence.
The learning repo
All state lives in one repo outside the user's work repos, at ~/learning unless $TUTORKIT_HOME names somewhere else. Keeping it out of a work repo is what lets the skill read a work repo for examples without writing notes into it.
~/learning/ ← one git repo; $TUTORKIT_HOME overrides
INDEX.md ← the router
REVIEW.md ← cross-topic due queue, one row per topic
NOTES.md ← how the user likes to be taught, global
assets/lesson.css ← copied from the skill on first run, never overwritten
topics/<slug>/
MISSION.md ← append-only; why they are learning it, plus the grounding consent
PROGRESS.md ← placement result, what stuck, what is shaky, where they are
CUES.md ← retrieval cues, key points, due dates, interval step, last seen
SOURCES.md ← vetted primary sources, so the agent stops re-searching
lessons/0001-<dash-case>.html
reference/<name>.html ← printable cheat sheets and the glossary
exercises/ ← runnable practice, code topics only
Create the repo on first use and run git init in it. The progress history is genuinely useful and the repo is the user's own. Never commit automatically. Offer a commit at the end of a session and take no for an answer.
The routing rule is the load-bearing guard. Read INDEX.md and NOTES.md at session start, resolve exactly one slug, then read only topics/<slug>/. Never glob across topics/. Without this rule the skill degrades as the user learns more, which is the exact wrong direction: thirty topics costs about sixty lines to route and tens of thousands of tokens to open.
Two modes are exceptions, and both are bounded. drill resolves its slugs from REVIEW.md rather than from the ask, then opens the CUES.md of those topics only. It never reads their lessons, and it never globs. Interleaving needs more than one topic in view; it does not need more than one file per topic.
status is the stricter exception: it reports on every topic and opens no topic folder at all. That is possible because INDEX.md and REVIEW.md already carry every field the dashboard prints. A cross-topic view that opens topic folders is a design failure, not a trade-off, because the two router files exist precisely so this read stays flat as the user's topic count grows.
When cwd is the learning repo itself, treat it as the learning repo and not as a grounding source. The two roles never overlap, so a lesson run from inside the learning repo skips the grounding question rather than offering to teach the user about their own notes.
The router
INDEX.md carries aliases so a fuzzy ask routes without opening anything:
| Topic | Slug | Also called | Status | Last touched |
|-------|------|-------------|--------|--------------|
| Postgres MVCC | `postgres-mvcc` | transaction isolation, snapshot, vacuum, row locking | active | 2026-08-11 |
Status is active or learned. Three routing outcomes, all cheap:
- The ask matches a slug or an alias → open that folder.
- It matches nothing and the user wants depth → create the folder and open a track.
- It matches nothing and the user wants an answer → run
explain and write nothing.
REVIEW.md holds one row per topic, never per cue: 2026-08-18 · postgres-mvcc · 4 due · min step: 3d. drill reads it, sees which topics are due, and opens only those CUES.md files. Storing cue text here instead would duplicate the answers and let them drift from the lessons that own them.
min step is the lowest interval step any cue in that topic has reached, and it earns its place by making one question answerable from the router alone: has this track finished? A topic whose lowest cue sits at 60d has passed the schedule half of the learned gate, so status can crown exam without reading a single CUES.md. Store the minimum rather than an average, because the gate is every cue at 60d and one cue at 1d fails it.
Both files are caches, and a cache needs a repair path. Rewrite the affected row whenever you touch a topic. Rebuild both by scanning topics/ whenever you find a folder they do not list. Without the self-heal, one hand edit misroutes silently forever.
A REVIEW.md row with no min step is a row written before this field existed. Read that topic's CUES.md once, write the field, and move on. Repairing one row costs one file read; refusing to repair it costs the same read on every later run.
The stylesheet
assets/lesson.css ships beside this file. On first run, resolve this skill's own installed directory and copy the stylesheet to <repo>/assets/lesson.css, because the working directory is the user's project, not the skill directory, so a relative path will not find it. Copy it once and never overwrite it. A later skill update then ships a new default for new repos and leaves every existing lesson rendering the way it was written. Linking the installed path instead would break every lesson the moment the skill moves.
Mode: status
The front door. One screen showing every track, what is due, and the one move worth making next. It never teaches, never asks a question, and never grades an answer.
1. Read the two routers, and nothing else
Read INDEX.md and REVIEW.md. List topics/ to check for a folder neither file names. Open no topic folder, no CUES.md, no PROGRESS.md, no lesson.
Listing a directory is not reading it, so the repair rule still runs here. status is the only mode that sees both routers whole, which makes it the one place drift reliably surfaces. Rebuild a row you find broken, and name the repair in the hand-off.
No learning repo, or an empty one? Say so in one line and stop. Offer to open a first track. Do not print an empty dashboard.
2. Rank, and retrieve before you add
One rule produces the crowned move: finish the retrieval you owe before you take on new material. A cue that is due decays while it waits, and a lesson does not. Adding a sixth track to five stalled ones feels like progress and is the most common way a learning habit fails.
| # | State | Move → |
|---|
| 1 | any cue is due today or earlier | drill |
| 2 | an active track has nothing due and min step below 60d | the next lesson on that track |
| 3 | an active track has nothing due and min step at 60d | exam, the transfer test |
| 4 | every track is learned, or no track exists | say there is no next step, and offer a new track |
Ties break on the most recently touched track. The user's mental model is warmest where they worked last, so that track costs the least to re-enter. This is the same reason a track untouched for months is not promoted: crowning the coldest track asks for the most expensive re-entry at the moment the user is only orienting.
Two findings are printed and never crowned, because acting on either is the user's judgement rather than a move the skill can defend:
- A stale track, meaning
active and untouched for more than 30 days. Name it and say what closes it: one lesson to restart it, or exam to close it out.
- More than five
active tracks. Attention is the scarce resource here, and a sixth track does not add capacity; it divides the same capacity further. Say the count and say that finishing one beats starting one.
3. Print the dashboard
One screen. One line per panel, tables under the panel they belong to, and an empty panel does not print.
# Learning status: YYYY-MM-DD
## Tracks active N · learned N · stale N
Active (N), most due first
| Topic | Due | Min step | Last touched |
|---|---|---|---|
| postgres-mvcc | 4 | 3d | 2026-08-11 |
| rust-lifetimes | 0 | 21d | 2026-08-16 |
| raft-consensus | 0 | 60d | 2026-06-02 |
## Review N cues due across M topics (omit when nothing is due)
## Learned <slug>, <slug> (omit when none)
## Next move
**→ <the move>.** <How to ask for it.>
Then:
- <runner-up>
- <runner-up>
Sort the table by due count descending, then by last touched descending. The row you act on first belongs on the first line. Say most due first on the count line, so the reader can check the order against the columns rather than infer it.
Print every active track. Past 10 rows, cap the table and close with a +N more line. Never truncate silently.
Write every move line in the procedural register. One instruction per line, active voice, present tense, no metaphor. Say "run drill", not "get back on the horse".
4. Write nothing
status produces no file. statuskit saves a snapshot because a repo dashboard is a ranked to-do list that exists nowhere else; here every fact on the screen is already durable in INDEX.md, REVIEW.md, and PROGRESS.md. A snapshot would be a third cache to keep honest, and it would go stale the moment the next drill run moves a due date.
The one exception is a router row this mode repaired. That is a cache write, not a record of learning. Then go to Hand off.
Mode: explain
The fast path. One ask, one answer, nothing written: no mission interview, no folder, no index row.
Answer in this order: the shortest correct answer first, then the mechanism, then one worked example, then the primary source to read next. Apply Citation discipline here as everywhere.
Offer a track once at the end ("I can open a track on this and teach it properly over a few sessions") and take no for an answer. A second offer turns the fast path into the thing it exists to avoid.
Mode: lesson
The core loop, and the only mode that opens a track.
1. Route
Read INDEX.md and NOTES.md. Resolve the ask to exactly one slug. On a known slug, open that topic folder only and go to the target step below. On an unknown slug, open the track first.
2. Open the track, on an unknown slug only
Run a short mission interview, three or four questions rather than an interrogation. What do they want to be able to do with this, by when, and what have they already tried? Then:
- Ask once for grounding consent. When cwd is a real project and is not the learning repo, ask whether tutorkit may read it to build examples from their own code. Record the answer in
MISSION.md as grounding: <repo path> or grounding: declined, and never ask again for that topic. A prompt that fires every lesson gets switched off; one prompt buys the whole track. Consent covers one repo, so a different cwd on a later session asks again.
- Write
MISSION.md under a ## YYYY-MM-DD heading.
- Create the topic folder, the
INDEX.md row, and the first SOURCES.md rows from whatever you searched to scope the topic.
- Offer placement, meaning
exam at track start, so lesson 1 is not pitched blind. On a decline, pitch from the mission interview and say the pitch is a guess.
MISSION.md is append-only. A mission that drifts gets a new dated entry; existing lessons are never rewritten and never archived. Re-pitching would rewrite files the user may have printed and annotated, and archiving would hide work they did and break the anchors between lessons.
3. Pick the target
Read PROGRESS.md and CUES.md. Pick one target at the edge of what they already know: the next thing that is reachable from what stuck, not the next thing in a syllabus. A wrong belief recorded in PROGRESS.md outranks a gap, because the wrong model actively blocks the correct one, so it is the higher-value target.
4. Make them predict, before you explain
This is the highest-value part of the skill. Pose a concrete scenario and make the user commit to a prediction before you teach anything. Wait for the answer. A wrong prediction names their broken mental model, and that model is the actual teaching target: you now know what to correct rather than what to cover.
5. Teach the minimum
Teach the mechanism, nothing beside it. Then one worked example. When grounding names a repo, build the example out of the user's own code: a lesson on dependency injection written against their actual service container beats one written against FooService. Apply Citation discipline to every non-obvious claim.
6. Practice against a feedback loop
Give one piece of practice with a real signal attached. For a code topic, write a runnable file into exercises/ and print the command that runs it, because a test run is the tightest feedback loop available and a browser quiz cannot match it. For a non-code topic, use a scenario the user works through and you grade.
7. Gate on explain-back
The lesson does not close until the user restates the concept in their own words. That retrieval is what converts fluency into storage strength, and it is the cheapest possible check that the lesson landed. On a thin restatement, re-teach the part they skipped and ask again.
8. Write the artifact and schedule the cues
Write the lesson per The lesson artifact. Then:
- Append every source you used to
SOURCES.md.
- Add 2–4 cues to
CUES.md per The spacing schedule, each with 2–3 key points that define a correct answer.
- Update
PROGRESS.md with what stuck, what is shaky, and where they are.
- Rewrite the topic's row in
INDEX.md and REVIEW.md.
Then go to Hand off.
Mode: drill
Retrieval practice across topics. This is the reason one parent repo exists rather than one repo per topic.
1. Build the queue
Read REVIEW.md. Open the CUES.md of due topics only, never their lessons, never a glob. Collect every cue whose due is today or earlier.
2. Cap it at 12
Sort by due ascending, then by misses descending. Fill the 12 slots from active topics first and let learned topics take what is left. Print what you left behind, as in 12 of 41 · 29 still due.
An uncapped queue is the failure mode that ends every spaced-repetition habit: forty cues on return trains the user to stop opening it. Hiding the backlog is only a nicer way to lose it. The overflow keeps its original due, so it stays first in line tomorrow rather than having its debt quietly rescheduled.
3. Interleave and ask
Mix cues from different topics rather than running one topic to exhaustion. Ask one at a time, with no answer visible, because a cue printed beside its answer tests nothing. Show last seen on the line when it is far in the past, so the user can judge their own miss with the fact in view.
4. Grade, offer the override, then reschedule
Grade the free-text answer against the cue's stored key points per Grading a free-text answer, offer the override on every grade, then write the new due, step, misses, and last seen. On missed, re-teach the cue inline in two or three sentences before moving on.
Update REVIEW.md and PROGRESS.md when the run ends. Then go to Hand off.
Mode: exam
Measures and never teaches. Ask, grade, record, refuse to explain. Two entry points, one posture; splitting them would name a presentation difference as a behavioural one.
drill and exam both ask questions, and the difference is what the answer is for. drill is practice: it asks to strengthen the memory, it repeats a cue for months, and a miss costs one reset. exam is measurement: it asks to produce a verdict, it runs twice per track, and the verdict decides where lesson 1 starts or whether the track closes. Never grade a cue in exam, and never write a status in drill.
At track start, placement
Pose 3–5 scenario predictions, broad to narrow, graded on the same three-level scale. Stop early on two consecutive misses, because a beginner does not need three more questions to prove it.
This reuses the prediction device the skill already owns rather than inventing a second assessment mechanism. Self-report is the weakest signal available, and a single transfer problem fails a genuine beginner flat on first contact.
Write placed: <rung> and every wrong belief, in the user's own words into PROGRESS.md. Those wrong beliefs become lesson 1's target.
At track end, the transfer test
Pose a problem the user has not seen, which needs the concept without naming it. Solving only the taught shape means the surface was learned, not the concept.
A topic goes learned in INDEX.md only when both gates pass: the transfer test is solved, and every cue has reached step 60d. exam is the only mode that writes that status. The transfer test alone would close a topic whose cues still sit at 1d; cues at 60d alone measure recall of taught shapes. When one gate passes and the other does not, say which one held it and leave the status active.
A learned topic keeps its cues. Retiring them at close is exactly when forgetting starts. They keep coming due, and the fill order in drill stops them displacing a live track.
exam writes PROGRESS.md at both ends and touches INDEX.md only to close a track. It never writes MISSION.md, because the mission belongs to lesson, which is the mode that opens one. Then go to Hand off.
The spacing schedule
Intervals: 1d → 3d → 7d → 21d → 60d. Cues live per topic in CUES.md:
| id | cue | key points | due | step | misses | last seen |
|----|-----|------------|-----|------|--------|-----------|
| 3 | Why does a long-running read never block a write in MVCC? | readers see a snapshot; writers make a new row version; visibility is decided per transaction | 2026-08-21 | 3d | 1 | 2026-08-18 |
got it advances one interval.partial repeats at the same interval.missed resets to 1d and re-teaches the cue inline.
There is no decay for lateness. A cue held back by the cap is the skill's debt, not the user's, and the cap can hold a cue thirty days past due. One schedule rule is easier to reason about than two, and the partial grade already absorbs most of what a decay rule would.
Grading a free-text answer
Three grades, judged against the cue's 2–3 stored key points:
got it. The answer covers the key points, in any words.partial. Some key points, or the right shape with a wrong detail.missed. The key points are absent, or the answer contradicts them.
Binary grading forces a wrong call on the half-right answer, which is most answers. The third grade absorbs that ambiguity instead of resolving it wrongly in either direction.
Offer the override on every grade, not only on a miss. The whole interval ladder runs on this judgement, so the user has to be able to correct it, most often when they gave a correct answer in words the key points did not anticipate.
The lesson artifact
One self-contained HTML file at topics/<slug>/lessons/NNNN-<dash-case>.html, linking ../../../assets/lesson.css. Tufte constraints: one column, generous margins, sidenotes rather than footnotes, and roughly 700 words as a ceiling. Working memory is small, and a long lesson is a lesson that does not finish.
Each lesson carries, in order:
- The one-line tie back to the mission, naming the dated
MISSION.md entry it was written against.
- The prediction the user made, and where it was wrong.
- The mechanism.
- One worked example.
- The practice, with its feedback loop.
- The cues that were scheduled.
- The primary source to read next.
- Anchors to sibling lessons and reference sheets.
- A closing line reminding the user that the agent is available for anything still unclear.
Naming the mission entry is what lets a mission change leave old lessons alone: a lesson written against a superseded mission reads as dated rather than wrong.
Reuse before authoring. Read assets/ first and build from the components already there. The stylesheet is what makes a pile of one-off files read as one course, and components the user's own topics earn accumulate beside it. Never inline something a second lesson would duplicate.
Reference sheets are the thing that gets revisited, so they carry the compressed essence rather than the narrative: a table, a decision rule, a diagram, not a retelling. Write them to reference/<name>.html. A glossary, once written, binds every later lesson.
Quiz options are the same length in words and characters, so formatting leaks no answer.
Opening the lesson is best-effort and never blocking. Try the platform's opener (open on macOS, xdg-open on Linux, start on Windows) and if none succeeds, print the file path and move on. A lesson that was written successfully must not report failure because a browser did not launch.
Citation discipline
Every non-obvious claim carries a primary source: the spec, the RFC, the source code, the official docs, the changelog. Prefer reading source over reading blog posts. When you cannot find a source, say "I believe X but could not source it" rather than asserting it.
Record what you read in SOURCES.md, with title, URL, date read, and one line on what it settles, so later sessions stop re-searching the same ground. Check SOURCES.md before searching.
Hand off
Write this section in the procedural register: one instruction per sentence, active voice, present tense, no metaphor.
What changed. Name the lesson written, the cues scheduled, the grades recorded, and what you marked as shaky. Name what did not change too: a declined track, a cue you could not source, a status you refused to write.
Where it landed. Give the topic folder path and the lesson path. Give the repo path when this run created the repo.
Next. Crown one move, chosen by state:
- Cues are due now → run
drill.
- The track is mid-flight and nothing is due → run the next
lesson.
- Every cue sits at step
60d and the track looks finished → run exam.
- The track just closed as
learned → say there is no next step. tutorkit is largely terminal. Do not invent a follow-up.
Then offer a commit of the learning repo. Never run it without a yes.
A status run closes differently, because its whole output is a hand-off. The dashboard already names the crowned move, so do not print the move twice. State that nothing changed, or name the single router row you repaired. Do not offer a commit on a run that wrote nothing.
Two sibling routes exist, and both are narrow. Name a sibling skill only when it is installed, and otherwise describe the action plainly. Route to a research skill when the question turns out to be a tool decision rather than a knowledge gap. Route to a prototype skill when the only honest answer is to build the thing and find out.
Notes
- The context guard beats every other rule here. A skill that opens thirty topic folders to teach one gets slower as the user learns more. Resolve one slug, open one folder.
status reports on all thirty and opens none of them, which is the guard working rather than an exception to it.
status reports; it never grades. Printing a due count is a fact read. Judging whether the user knows a topic is exam, and it needs an answer from the user before it can say anything. A dashboard that inferred mastery from a min step column would manufacture the exact signal it claims to report.- Predict before you explain, every time. The temptation is to skip it on an easy topic. The prediction is the diagnosis, and teaching without it is teaching blind.
- Never assert an unsourced claim as fact. A confident wrong explanation is worse than no lesson, because the user will build on it.
- No writable filesystem (a browser-based agent)? Say so plainly, print the lesson as a code block for the user to save, and note that spacing cannot persist without state. Do not pretend to schedule a cue you cannot write.
status still runs when the filesystem is readable, because it writes nothing but a repair; say that the repair was skipped.
