Portable Markdown skills that turn documents or code into a StudyVault, teach and quiz you, and build a learning plan from your recorded study evidence. Use them with an agent that supports skills, a chat application that accepts files, or an API that accepts text instructions. Obsidian is optional.
Adapted from upstream tutor-skills. The original MIT copyright and license are preserved in LICENSE.
| Skill | Purpose | Input | Output |
|---|---|---|---|
| tutor-setup | Organize learning material | Documents, pasted text, accessible web content, or source code | Linked Markdown notes, a dashboard, and practice questions |
| tutor | Assess, teach, and plan | Notes plus available study and answer history | Feedback, concept progress, and an adaptive learning plan |
The portable interface is each skill's SKILL.md and its linked resources. Platform-specific slash commands, question widgets, shell tools, and automatic file discovery are optional conveniences.
- Choose a skill and provide its
SKILL.mdto your assistant as an instruction file or prompt. - Keep the skill's resource folders alongside it. If your assistant cannot follow local links, attach or paste the referenced files needed for your request.
- Provide the source material or existing StudyVault and ask for the outcome you want.
For example:
Follow tutor-setup/SKILL.md. Turn these lecture notes into a StudyVault.
Follow tutor/SKILL.md. Quiz me on operating systems, one question at a time.
Follow tutor/SKILL.md. Use my study logs and answer history to plan 30 minutes a day this week.
In a Korean conversation:
tutor 스킬에 따라 이전 학습 인증 시간, 퀴즈와 개념 설명의 답변 시간,
정확도를 살펴보고 이번 주 하루 30분 학습 계획을 세워줘.
| Available capability | How the skills use it | Without it |
|---|---|---|
| File access | Read notes and save progress | Accept pasted/attached content and return Markdown for you to save |
| Persistent storage | Reuse evidence across sessions | Return the updated record; provide it again next session |
| Question widget | Present supported quiz inputs | Ask in ordinary chat; free-text explanations are supported |
| Reliable timestamps | Record per-answer elapsed time | Use clearly labeled self-reported timing or leave timing unknown |
| PDF/web extraction | Read accessible source content | Request readable text; do not silently install tools |
Markdown links and standard headings work without Obsidian. Hosts may differ in context limits, persistence, file support, and instruction following; copying the skills does not add missing capabilities.
No installer is required. If your agent discovers skill directories, copy skills/tutor-setup and skills/tutor into the directory documented by that host. Include all resources and a copy of LICENSE with each skill. There is no default Claude Code directory.
From this repository, the Bash helper installs both skills into an explicitly chosen directory:
bash install.sh /absolute/path/to/your/skills-directoryThe helper copies all skill resources (including optional scripts) and the MIT license, refuses existing destinations, and adds an ownership marker for uninstall. It requires Bash and standard cp/mkdir utilities, such as those supplied by Git Bash on Windows. To upgrade, back up any local changes, uninstall the previous helper-installed copy, then install again.
On Windows, File Explorer copying works too. Or use PowerShell with a directory of your choice:
$skillTarget = 'C:\path\to\your\skills-directory'
New-Item -ItemType Directory -Path $skillTarget -Force | Out-Null
foreach ($skillName in 'tutor-setup', 'tutor') {
$destination = Join-Path $skillTarget $skillName
if (Test-Path -LiteralPath $destination) { throw "Already exists: $destination" }
}
foreach ($skillName in 'tutor-setup', 'tutor') {
$destination = Join-Path $skillTarget $skillName
Copy-Item -LiteralPath (Join-Path '.\skills' $skillName) -Destination $destination -Recurse
Copy-Item -LiteralPath '.\LICENSE' -Destination (Join-Path $destination 'LICENSE')
}Run these commands from the repository directory. Open a new assistant session if your host discovers skills only at startup. Use its native invocation syntax if available, or ask it to follow the skill by name/file.
Create notes. tutor-setup organizes documents into topic notes and practice questions, or source code into architecture notes, module guides, and onboarding exercises. It records source references, maps prerequisites, and links the notes from a dashboard.
Gather evidence. tutor handles diagnostic quizzes, weak-area practice, concept explanations, and study-time records. Study logs label their source as self_reported, timer, or artifact; none is automatically an externally verified record. Study time indicates effort, not proof of understanding. Records live in StudyVault/learning/evidence.json, or can be exported in chat for you to retain.
Infer learning state. When you request a learning plan, the tutor combines recorded study time, graded accuracy, per-answer time when available, recency, and evidence coverage. Accuracy, evidence confidence, fluency, and study effort remain separate signals. Timing is compared only across comparable tasks; interruptions, accessibility needs, unknown timestamps, and assisted answers affect interpretation. A long or short answer time alone does not determine mastery. Quiz recall and explanations of concepts can provide different evidence about understanding.
Adapt the plan. The tutor explains its estimate and confidence, prioritizes prerequisites and weak concepts, allocates your available time, and includes a later reassessment. High accuracy with slow reliable responses calls for fluency practice; repeated errors despite substantial study call for a different explanation or prerequisite review. Missing evidence calls for a short diagnostic and a provisional plan. Estimates are transparent planning heuristics, not validated measurements of intelligence or learning ability. Plans are saved to StudyVault/learning/plan.md when file writing is available.
New observations update the next plan. Earlier records stay available so an agent can distinguish actual improvement from an isolated correct answer. The learning model documents the evidence format and decision rules. An optional Python helper computes summaries with the standard library; agents without Python apply the same rules directly. No Python runtime is needed for ordinary prompt-based use.
Run the learning-model regression checks with Python 3 (standard library only):
python tests/test_learning.pyFor copies installed by the Bash helper, use the same target directory:
bash uninstall.sh /absolute/path/to/your/skills-directoryThe uninstaller removes only recognized tutor-setup and tutor directories, refuses symlinks or missing ownership markers, and leaves the parent directory intact. Back up changes inside installed skill folders first. StudyVaults stored outside those folders are untouched. Remove manually copied skills manually after checking their exact paths.
MIT, including the upstream copyright notice: Copyright (c) 2026 tak.