StudyKit

Turn any project directory into a spaced-repetition study workspace, driven by Claude skills.

StudyKit installs a small set of Claude Code skills plus a zero-dependency Node engine into a project directory. You study one certification per project; your learning state lives as flat markdown + JSONL on disk, so any session resumes exactly where you left off with a single command — no manual context reload.

• 📚 Two phases, one driver — a daily study loop (review → learn → recall) that graduates to an exam-prep loop (targeted revision + full mock exams), switched only with your confirmation.
• 🎯 Deterministic spaced repetition — a seeded selector resurfaces the topics you're weakest on and haven't seen in a while.
• 🧮 Honest progress — coverage, accuracy, streaks, and pace vs. your target date, computed from your real attempt log.
• ✍️ Open-ended grading — rubric-based judging of Feynman teach-backs and "explain X" answers, fed back into spaced repetition.
• 🗂️ Multi-cert overview — one read-only dashboard across every cert you're studying.
• 🪶 Zero runtime dependencies — pure Node (>=18), tested with the built-in node --test.

Table of contents

• Install
• Quickstart
• Skills
• What /sk:init creates
• How it works
• Multiple certifications
• Development
• Releasing
• License

Install

The npm package is studykit (lowercase, an npm requirement); the project name is StudyKit.

Install the skills into a study project — a directory for one exam or topic:

# from npm
npx studykit@latest init ~/study/aws-saa

# from a git checkout
node bin/studykit.cjs init ~/study/aws-saa

init only installs skills into <project>/.claude/skills/. It never creates or touches study/ — that belongs to /sk:init.

Quickstart

Open the study project in Claude Code, then:

| Step | Command | What it does | |------|---------|--------------| | 1 | /sk:init | Once per project. Researches the exam blueprint, writes profile.md / syllabus.md / knowledge/, captures your background, pass mark, and mock size. | | 2 | /sk:learn | Every session. Runs the daily loop one question at a time and resumes the exact step + question across sessions. | | 3 | /sk:stats · /sk:summary | Anytime. The numbers, or a narrative with pace vs. your target date (sk:summary also saves study/progress.md). | | 4 | Exam-prep | When coverage is high or the exam is near, /sk:learn proposes switching phases (you confirm), then runs revision + blueprint-weighted, resumable, scored mock exams. | | 5 | /sk:judge | Grade an open-ended explanation against a rubric; tracked topics get recorded so the result feeds spaced repetition. |

Studying more than one cert? See Multiple certifications.

Skills

| Skill | Role | | ----- | ---- | | sk-init | One-time setup: builds the exam blueprint + knowledge base | | sk-learn | The daily driver: study loop + exam-prep loop, with exact resume | | sk-stats | Read-only progress metrics (totals, coverage, weak/stale topics) | | sk-summary | Narrative progress + pace vs. target; writes study/progress.md | | sk-judge | Rubric grader for open-ended answers (Feynman teach-back, "explain X") | | sk-certs | Read-only overview across multiple cert projects | | _engine | Zero-dependency Node helpers (state · select · stats · scan) |

What /sk:init creates

<study-project>/
  study/
    state.json            # machine cursor (the engine owns this schema)
    profile.md            # exam meta + structure + your background + target date
    syllabus.md           # topic | domain | status | last_studied
    knowledge/<topic>.md  # knowledge base, grows as you study
    daily/YYYY-MM-DD.md   # human-readable daily record
    progress.md           # latest snapshot (written by /sk:summary)
    results.jsonl         # append-only attempt log the engine reads

How it works

• Markdown is the human source of truth; results.jsonl is the machine index. Only the engine reads the attempt log; skills load the small aggregate it returns.
• State-first resume. Every skill reads the tiny state.json cursor first, then loads only the slice it needs — reads stay token-bounded no matter how long your history grows. Resume reads state.json only, never markdown, even mid-mock-exam.
• Deterministic, topic-level selection. select.cjs is seeded and resurfaces topics (older-within-window and previously-wrong weighted heavier); sk-learn then generates a fresh question per topic. There is no question bank.
• Two phases, one driver. state.phase is study or exam-prep. The switch is always user-confirmed and reversible; the schema change is additive (no migration needed).
• Binary scoring everywhere. Multiple-choice and rubric-judged open-ended answers both resolve to one binary results.jsonl line; rich rubric detail stays human-facing in daily/.
• Zero dependencies. Pure Node CommonJS (>=18), tested with the built-in node --test.

Multiple certifications

StudyKit keeps 1 project = 1 cert, so each study/ stays isolated and token-bounded. To study several, create sibling projects under one parent and use the read-only overview:

npx studykit@latest init ~/study/aws-saa
npx studykit@latest init ~/study/terraform
# run /sk:init in each, study as usual, then from any project:
#   /sk:certs ~/study
#   -> combined table: phase, coverage, streak, days-to-target, next action

/sk:certs reads each cert's state.json + stats aggregate only — never raw attempt logs — and suggests which cert to study next. It mutates nothing.

Development

npm test          # node --test skills/_engine/test/

The engine is pure Node CommonJS with zero runtime dependencies. Every .cjs ships with a node:test suite; keep it green.

Releasing

See docs/release-process.md for the full publish runbook: semantic versioning, the pre-publish audit, git tagging, GitHub push, and npm publish (including the 2FA one-time-password step).

License

MIT © Thuan Bui