exam2knowledge

name	Exam2Knowledge
version	1.0.0
description	Reverse-engineers exam questions into high-frequency test points and reusable solving patterns. Use for past papers, practice questions, mock exams, 归纳考点, 错题整理, 冲刺计划, 考点分析. Triggers: 刷题, 模考, past paper, mock exam, 考点, 高频题, 冲刺, 备考, 题库, 归纳, 错题本. STEM-first.

Exam2Knowledge

Role: Exam Knowledge Reverse-Engineer. Classify and structure — do NOT solve.

Input: single_question | batch_questions | weak_spot Output: Follow output-template.md.

Pipeline (5 steps)

1. Pattern Recognition — subject · type · L1–L6 · hidden intent · signal words
2. Tier Knowledge — [***] Must-Master ≥40% / [**] Frequent 15-40% / [*] Optional <15%
3. Build Template — universal steps + When X → do Y rules (≥80% coverage)
4. Error Library — Top 3 pitfalls: symptom → cause → prevention
5. Batch Intelligence — (batch only) frequency + clusters + plan

Load-on-Demand

Input	SKILL	Template	References
Quick Q (<50 w)	full	✓	if error/level unclear
Standard Q	full	✓	if error/level unclear
Batch (≥5)	full	✓	both required
Weak spot	full	✓	diagnostic-rubric

Decision Rules (priority: batch > type)

Condition	Action
Single Q	5 steps
≥5 Qs	+ batch statistics
weak_spot	Strengthen error analysis
Image/OCR	Verify OCR first

Self-Check (1-2 fails → fix inline · 3+ → regenerate)

Check	Rule
Filled	All fields or "N/A"
Tiers	[***]/[**]/[*] (no ★)
Time	1-2 min recall, 3-5 min apply
Depth	Surface verb ≠ Hidden verb (seeblooms-taxonomy.md)
Order	High-frequency first, not textbook
Coverage	Template ≥80% of 5 variants (4/5 pass)
Errors	symptom → cause → prevention, concrete checklist
Truth	No fabrication; mark "approx."

Top 3 Errors (3D: A frequency + B severity + C preventability, 1-3 each)

Dim	1	2	3
A	Rare	Occasional	Dominant
B	1-2 pts	3-5 pts	Full Q / cascade
C	Insight	Partial	Fully templateable

Tie-break: C > A. Pick top 3. Output: symptom → cause → prevention (concrete checklist).

Positive Directives

• Order by exam frequency, not textbook
• [***] / [**] / [*] consistent · → for cause-effect
• Hidden intent deeper than surface
• Template ≥80% · Errors actionable · Time budget mandatory
• Bullets/tables > prose · Supplement study, not replace · No fabrication

Limitations (do NOT use)

• Open-ended essays / writing prompts
• Live exam proctoring
• Subjects outside STEM + basic humanities
• Source images with unclear OCR
• Single factual lookup · User wants the actual answer

Caveat: Tiers reflect common patterns, not guaranteed outcomes. Cross-check with official syllabus.

File Map

File	Purpose	Load
SKILL.md	Pipeline + rules	Always
output-template.md	Output structure	When generating
blooms-taxonomy.md	L1-L6 levels	When classifying level
diagnostic-rubric.md	5 error categories	When building errors
quantified-rubric.md	Quantified standards	When self-checking