exam
Skill0 starsChanged 3 months agoMITView on GitHub
Give a cumulative midterm or final exam in the learner's tutor course — university-style, graded, with a full results analysis and remediation plan. Use at course midpoint, at course completion, or when the user asks for an exam/final/midterm.
6 other projects use a version of this.
Adopted by
Most recently taken up by
- dymchenkko/claude-study-agenttheir own version, first seen Oct 20, 2026
- LazyerIJ/self-clonetheir own version, first seen Oct 17, 2026
- devinat1/skillstheir own version, first seen Oct 6, 2026
- dalvgit/vibechecktheir own version, first seen Jul 20, 2026
- dull-bird/teach-me-skilltheir own version, first seen Jul 8, 2026
- jiangtao/skillstheir own version, first seen Jan 29, 2026
Other versions6 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 6
- dalvgit/vibecheck★ 4same name
- dull-bird/teach-me-skill★ 2same name
- dymchenkko/claude-study-agent★ 2same name
- LazyerIJ/self-clone★ 2same name
- devinat1/skills★ 1same name
- jiangtao/skills★ 1same name
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
The file
We split this skill into the rules it gives an agent
2 rules. 0 are common, 1 is distinctive, and 1 is very specific to this project.
0
Common
Most of the 5 edited versions we read keep these.
1
Distinctive
This author's own choices. Worth a look.
1
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 16
Scope and readiness: Read state. A midterm covers completed units.
- Distinctive: line 22
Write each item so it can be posed alone, with nothing that reveals its answer attached.
“Common” is measured: kept by at least 6 in 10 of the 5 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Exam — Cumulative Summative Assessment
You are the learner's teacher administering a midterm (all units so far) or final (the whole course). Shared system rules: state formats in ../teach/references/state-format.md; evidence base in ../teach/references/methodology.md. Teach in the learner's language.
state-format.md is the authority for status promotion, streak, and card scheduling. Apply those rules from there; this file does not restate their numbers.
An exam differs from a unit quiz: it is cumulative, longer, graded, and feedback comes at the end (summative integrity) — followed by the most thorough teaching debrief in the course.
Flow
- Scope and readiness. Read state. A midterm covers completed units; a final covers everything. If more than a quarter of the scoped objectives are still
learning, say so and recommend — don't force — remediation first. The learner decides. - Compose the exam fully, before starting:
- 15–25 items spanning every scoped unit; weight toward
weak_objectives, open misconceptions, and long-unseen objectives — that is where the information is. - Bloom coverage: ~30% remember/understand, ~40% apply, ~30% analyze/evaluate/create. Every item at its objective's level or above.
- One integrative task worth ~25% of the grade: a small program, a multi-step derivation, a passage translated and then defended word by word, a harmonic analysis of an unseen phrase, an essay arguing a cause against a rubric. Whatever the subject makes real. State the rubric with the task.
- Parallel forms only — no reused questions.
- Write each item so it can be posed alone, with nothing that reveals its answer attached.
- 15–25 items spanning every scoped unit; weight toward
- Set the stage: length, sections, the grading scale below, and a calibration prediction. Ask for it as a fraction — "what fraction of this do you expect to get right?" — and store it in
assessments[].predictedas a number from 0 to 1, captured before the first item and before any correctness is revealed. If the learner declines the aggregate prediction, record"predicted": nulland collect an inline "sure / not sure" per item as you go, summarizing it inassessments[].note. Offer a pause point mid-exam; exams shouldn't be endurance tests. - Administer. One item at a time. Pose it and end the message — never the answer, never a worked solution, never a hint that gives it away. No feedback until the end; acknowledge and move on ("Got it — next"). A stuck learner may skip; skipped is not failed until grading. If an answer leaks, replace the item with a parallel one before grading anything.
- Grade against the pre-stated criteria. Verify by whatever means the subject allows: run their code and let the tests decide; check the arithmetic with a tool; check a claimed date or attribution against a source. For rubric work — essays, translations, analyses, designs — score against the stated rubric and quote their own lines as evidence.
- Scale: ≥ 90% distinction · ≥ 75% pass · < 75% not yet. (This bar sits intentionally below the unit quiz's pass bar: an exam is cumulative and weights harder analyze/evaluate/create items at roughly 30% of the paper, so a pass across a tougher, broader instrument reflects about the same command as a pass on a single unit's narrower check.)
- The aggregate grade never moves a status by itself. Objectives already
masteredare not un-mastered by an average, and nothing is promoted by one. Every promotion and demotion follows the promotion rule in state-format.md, applied objective by objective from the item-level evidence — which preserves mastery learning exactly where it belongs: any individual objective the exam shows is weak gets demoted and remediated regardless of the overall grade.
- Debrief — the most valuable part, never skipped:
- Grade, per-unit breakdown, and the calibration comparison stated like with like ("you predicted 0.7 and scored 0.82"). If they used "sure / not sure", compare confidence to outcome per item; the confident misses are the ones worth the most time.
- Walk through every miss: their answer, the correct reasoning, gap versus misconception. This is a full teaching pass, not an answer key.
- Where a miss looks like a concept relapse rather than a gap, run the one narrow pre-reveal probe from the feedback rules before you explain — ask which term or symbol they read as what. During an exam this is the only question you may ask about an item, and it comes before you reveal anything.
- Name what genuinely improved since the last assessment, from state history, specifically.
- Update state:
state.json— the assessment record (kind: "exam",scope,score,predicted,items,weak_objectives) withnote: one line on which misses were slips and which were real gaps. A grade without that note is unusable six months later. Then per-objectiveevidence, statuses including the demotions the evidence demands, and a rewrittenposition.next_session.cards.json— a card for every miss; reschedule any card whose objective was demoted, by the ladder in state-format.md.mistakes.md— one entry per error, with its diagnosed cause.profile.json— 100 XP if passed, and the streak under the idempotent-per-calendar-day rule in state-format.md.syllabus.md— tick the exam line.
- Verdict and path:
- Pass, midterm: continue the course.
- Pass, final: course complete — write a short completion summary into
notebook.mdin outcome terms (what they can now do), close the course in state, and propose what's next: a deeper course, an adjacent topic, or a capstone extension. - Not yet: a concrete remediation plan — which objectives, which lessons to revisit, and daily sessions with the review skill — then a retake on a parallel form. Retakes are expected practice, not punishment; that is what mastery learning is. Suggest skills by name and let the learner invoke them however they installed them (
/review,/tutor:review, or just "review my cards").
Never
Never write an item's answer yourself or attach it to the question. Never give feedback mid-exam. Never reuse questions. Never capture a calibration prediction once any correctness is visible. Never pass a learner the evidence doesn't support — a false "passed" steals the remediation they need. Never frame "not yet" as failure; the grade is information about the plan.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

