| name | study |
| description | Run a structured AI-assisted study session on a curriculum module or topic. Use when the user wants to learn, review, or work through material — e.g. "/study distributed-systems", "study Paxos with me", or "let's work on the ML module". |
Study Session
Run one iteration of the repo's study loop (defined in milestones/roadmap.md) on the topic given in the arguments. If no topic is given, ask which module or concept to work on, listing the modules in curriculum/core/.
Setup
- If the topic maps to a module, read its file in
curriculum/core/ (or curriculum/specializations/) and anchor the session to its learning objectives and mastery checklist.
- Check
research/ for the user's existing notes or error logs on this topic and calibrate difficulty accordingly. Don't re-teach what their notes show they've mastered; target weaknesses.
The loop
Work through these stages, one at a time, conversationally:
- Explain — teach the concept in three passes: intuition, formalism, implementation. Name the standard misconceptions and edge cases explicitly. Cite where this sits in the module's primary resources so the user can deep-read later.
- Generate — produce 3–5 exercises at increasing difficulty: at least one conceptual, one quantitative/proof, one open-ended. Keep your rubric hidden.
- Attempt first — the user solves before you reveal anything. Give hints only when asked, smallest hint first. Never volunteer the answer.
- Critique — grade their solution against the hidden rubric: correctness (40), depth (25), rigor (20), clarity (15). State the minimum changes needed for full marks.
- Repair — for anything below full marks, append an entry to the user's error log (suggest
research/error-log.md if none exists): topic, mistake, root cause, scheduled re-attempt date (+48h).
