Predict the paper → forge the study guide → drill it to retention.
ExamForge turns your own course material — lecture slides, professor transcripts (English or Arabic), past exam papers, model answers, lab manuals, notes — into a source-grounded study guide, an evidence-ranked prediction of what the exam will actually ask, and a spaced-repetition drill built from that prediction.
Built for biology-family courses (biotechnology, biology, biochemistry, bioinformatics, chemistry, microbiology, molecular biology) but adapts to the material you give it.
ExamForge exists because two things a student needs live in two different runtimes:
| Runtime | Gives you | Persistence | |
|---|---|---|---|
| ExamForge skill | claude.ai chat | One-shot: study guide + exam prediction (HTML/docx) | None (single session) |
| Enhanced Learn FASTER | Claude Code / Codex CLI | Ongoing coached spaced repetition, progress tracking, printable mock exams | Yes (.learning/ state across days) |
They share one intelligence: the same signal-weighting, "purple" explicit-marker extraction, Arabic-cue parsing, and faithfulness rules. Use the skill for a fast guide with zero setup; graduate to the CLI kit when you want the predicted paper drilled on a spaced schedule until exam day.
Stock Learn FASTER generates practice from concepts you've already learned in the tool and even searches online for exam examples. ExamForge inverts that: it ingests your professor's actual materials, extracts what the professor explicitly said is on the exam, predicts the real paper with confidence tags, and seeds the spaced-repetition scheduler with those questions — not generic ones.
Key features:
- Purple vs red — what the professor literally said is on the exam (fact, quoted, Arabic translated) is kept distinct from what ExamForge predicts is likely (inference, confidence-tagged).
- Evidence-weighted prediction — past papers strongest, then model answers, slide repetition, verbal emphasis, your insider tips; textbook last.
- Reads Arabic transcripts — catches spoken cues like "ده مهم للامتحان" and folds them into English output.
- Marks-driven — asks total marks + distribution and prioritizes accordingly.
- Conditional sections — calculations, visual/spot-ID tables, and protocol breakdowns activate only when the material contains them.
- Source-faithful — bound to your materials; outside knowledge only for central-but-poorly-explained points, always flagged.
Download dist/examforge.skill and use Save skill in the Claude app / claude.ai (availability depends on plan/org), or drop skill/examforge/ into a Claude Code skills directory. Then just say "predict my microbiology final from these slides" and attach files.
Needs uv. Install the bundled, enhanced kit from this repo (not Hugo's original — the ExamForge commands only exist here):
uv tool install learn-faster --from ./claude-code-kit
learn-faster # choose Exam-Oriented modeThen, in a project directory:
/ingest-materials # load your slides, transcripts, past papers, model answers
/predict-exam # source-grounded prediction + guide; seeds the review scheduler
/review # spaced-repetition drills on the predicted paper
/generate-exam # printable mock exam (PDF) from your material
The chat skill can also emit a learn-faster-handoff.md to seed the kit if you started in chat.
skill/examforge/ # claude.ai chat skill (SKILL.md + references)
dist/examforge.skill # packaged, installable skill
claude-code-kit/ # Learn FASTER (MIT © Hugo Lau), extended with ExamForge exam commands
docs/how-it-fits.md # how the two entry points share one prediction core
NOTICE.md # component attribution & license details
MIT. ExamForge's prediction intelligence is generalized from exam-prep prompts by Hossam Mahmoud. The spaced-repetition engine is Learn FASTER, MIT © 2025 Hugo Lau — bundled and extended under claude-code-kit/. See NOTICE.md for exact attribution and the list of modifications.