---
name: "study_skills"
description: "Designs evidence-based study plans from cognitive-science research: spaced-repetition schedules, active-recall practice, and interleaved problem sets. Triggers when someone asks for help studying, preparing for an exam or certification, learning a subject deeply, or building a review system — e.g. 'help me study for my physics final', 'make me a study plan', 'I keep forgetting what I learn'."
metadata: { "includeInPrompt": true }
---

# Study Skills

## Purpose
Turn any learning goal into a study plan that uses the techniques cognitive science actually supports: retrieval practice (test yourself, don't reread), spaced repetition (spread sessions over time with widening gaps), and interleaved practice (mix topics instead of drilling one block at a time). Retire the low-utility habits — rereading, highlighting, cramming — that research shows barely move the needle.

## Workflow
1. **Get the target.** Ask: what material, what date does it need to be retained until (exam date, or "forever" for durable learning), and how many sessions per week are realistic.
2. **Generate the schedule.** Run `bin/study_schedule.py --exam YYYY-MM-DD --topics "t1,t2,t3" --reviews N` for an exam-dated plan, or adapt the worked templates in `references/weekly-plans.md` for open-ended learning. Inter-study gaps default to ~15% of the retention interval (the Cepeda et al. 2008 optimum).
3. **Build active-recall materials.** For each topic, produce 5–15 questions the learner answers *from memory* before checking the source: definitions, worked-problem prompts ("solve this without notes"), compare/contrast ("when do you use X vs Y?"), and "why" questions (elaborative interrogation). Blank-page recall first, source-check second.
4. **Interleave, don't block.** Practice sessions mix problem types so every item forces *strategy choice* — the learner must decide which method applies, not just execute the one from today's lesson. Within a session, no two consecutive items should be the same topic.
5. **Enforce the review discipline.** Each review is pass/fail: cold recall → check → re-test misses. If a topic fails recall, its next review resets to 1 day out (SM-2 quality < 3 rule). Do not drop a topic after one successful recall — that is exactly when people stop and exactly when retention collapses (Karpicke 2009).
6. **Warn off the traps.** Rereading and highlighting feel productive but are rated low utility (Dunlosky et al. 2013); extra same-session problems after mastery (overlearning) add almost nothing (Rohrer & Taylor 2007). Cut them from the plan and say why.

## Output Contract
A finished plan always contains:
- Dated sessions (from `study_schedule.py` or a worked template) covering every topic
- For each topic: an initial-learn session ("self-test until one correct recall") plus ≥3 spaced review sessions
- A question bank of active-recall prompts per topic
- Sessions labeled by method (cold recall → check → re-test), not just by topic
- A one-paragraph "why this works" note citing the research basis, not vibes

## Operating Rules
1. **Accuracy rule:** every factual claim in a plan must trace to a verified source (see `references/research-brief.md`). Never invent statistics, dates, or quotes. Unverifiable = cut it or frame as opinion.
2. Exam date is required for dated plans. Without one, schedule by "retain forever" defaults: reviews at ~1 day, ~1 week, ~1 month.
3. One topic = un-interleavable. If the user gives a single topic, split it into sub-skills (definitions, worked problems, edge cases) so sessions still force discrimination.
4. Keep plans honest about load: a plan the learner won't do is worse than a smaller one they will. Prefer fewer sessions spaced correctly over many sessions crammed.
5. Never present rereading/highlighting/cramming as effective techniques, even if the user asks for them — explain what to do instead.