Skip to main content

wrong-answer-print

Integration skill for Z: drive wrong-answer printer, interactive wrong-list excel update, HWPX generation, 1-shot LLM solution caching, and Daily Test constructed response (A6/A7) reassignment pipeline.

Source facts

Repository
park-kyungchan/brain
Last source activity
July 21, 2026 at 16:21
Detected SKILL.md language
English
Stars
0
Forks
0

Install options

The review-first prompt is selected by default. You can switch to a direct command or download a local copy.

Review the source files

Read SKILL.md and any companion files shown by SkillsMP before deciding whether to install.

Showing SKILL.md

SKILL.md
Source instructions ยท Read-only preview
name
wrong-answer-print
description
Integration skill for Z: drive wrong-answer printer, interactive wrong-list excel update, HWPX generation, 1-shot LLM solution caching, and Daily Test constructed response (A6/A7) reassignment pipeline.
# Wrong-Answer Print & Reassessment Pipeline Skill Integrates the wrong-answer printer workspace (`Z:\homes\onemath\๊ณต์œ ํด๋”\ํ•™์›์„œ๋ฒ„\โ˜…์˜ค๋‹ตํ”„๋ฆฐํŠธ`) with the Brain repository, caching per-student error histories to reassign items as short/extended constructed response items ($A6/A7$) in Daily Tests. ## Invariants & Operating Doctrine 1. **Interactive Excel Update**: When user requests wrong-item additions in natural language, automatically update `โ˜…KHJ_์˜ค๋‹ต๋ฆฌ์ŠคํŠธ_๋ฐ•๊ฒฝ์ฐฌ.xlsx` via `wrong_answer_print.py`. 2. **Automated HWPX Generation**: Win32com HWP integration generates `.hwpx`. Omits expensive LLM visual auditing; the human user performs final visual review in the Hangul app. 3. **Deterministic 1-Shot Solution Caching**: If a solution for problem image `{num}.png` is missing, multimodal LLM extracts correct answer, step-by-step solution, target concepts, and targeted misconceptions ($M$) once into `memory/wrong-answers/solutions/{course}_{book}_{num}.json` (0 LLM token cost on subsequent runs). 4. **Daily Test Reassessment Bridge**: Feeds collected wrong-answer data into `wrong_answer_dt_bridge.py` to generate Daily Test constructed response items ($A6$: terminology / $A7$: solution process) adhering to $T \times A \times Q \times M$ item design axes. ## Command Workflow ### Step 1. Interactive Wrong Entry & HWPX Build ```bash python tools/py/wrong_answer_print.py --student "๊น€์ •ํ˜" --course "m3s1" --book "ssen" --nums 33 38 42 ``` * Use `--dry-run` to update Excel and student JSON history without launching HWP. ### Step 2. Solution Extraction & 1-Shot Caching ```bash python tools/py/wrong_answer_solver.py --course "m3s1" --book "ssen" --num 33 ``` * SHA256 source image matching guarantees 0 LLM API calls when cache exists. ### Step 3. Reassessment INTENT-DRAFT Generation ```bash python tools/py/wrong_answer_dt_bridge.py --student "๊น€์ •ํ˜" --out "intent_draft_reassessment.json" ``` ## Authority References - Solution Cache Schema: `memory/wrong-answers/schemas/wrong_answer_solution.schema.json` - Student Wrong History Schema: `memory/wrong-answers/schemas/student_wrong_history.schema.json` - Excel SSoT: `Z:\homes\onemath\๊ณต์œ ํด๋”\ํ•™์›์„œ๋ฒ„\โ˜…์˜ค๋‹ตํ”„๋ฆฐํŠธ\โ˜…์˜ค๋‹ตํ”„๋ฆฐํŠธ ํ”„๋กœ๊ทธ๋žจ\๋ฐ•๊ฒฝ์ฐฌ\โ˜…KHJ_์˜ค๋‹ต๋ฆฌ์ŠคํŠธ_๋ฐ•๊ฒฝ์ฐฌ.xlsx`
View on GitHub