AgentSkillsCN

physics-lesson-capture

从图片、PDF 或 DOCX 文件中提取课程内容，运行 OCR（DeepSeek-OCR），提取例题，总结课堂讨论，并为课后诊断做好准备。适用于教师提供课程资料或讨论课堂内容时使用。

SKILL.md
--- frontmatter
name: physics-lesson-capture
description: Capture lesson content from images/PDF/DOCX, run OCR (DeepSeek-OCR), extract example questions, summarize classroom discussion, and prepare lesson-first post-class diagnostics. Use when teachers provide lesson materials or discuss classroom content.

Physics Lesson Capture

Overview

Use this skill when the teacher provides lesson materials (images/PDF/DOCX) or describes classroom content. The goal is to OCR/parse the materials, extract example questions, and produce a structured lesson summary and discussion notes that can drive post-class diagnostics and homework.

Required Inputs

  • •lesson_id (e.g., L2403_2026-02-04)
  • •lesson topic (string)
  • •class (optional)
  • •sources: images, PDF, or DOCX files
  • •optional teacher notes: classroom misconceptions, emphasis KPs, homework focus

Workflow (Lesson-First)

  1. •

    Ingest sources

    • •Copy sources into data/lessons/<lesson_id>/sources/.
    • •Record a manifest file with filenames and types (see references/lesson_io.md).
  2. •

    Extract text

    • •PDF: if text-based, use pdfplumber; if scanned, render pages to images and OCR.
    • •DOCX: extract text via python-docx; OCR embedded images.
    • •Images: OCR directly.
    • •Use DeepSeek-OCR via SiliconFlow for OCR (see references/ocr_pipeline.md).
    • •Script: scripts/lesson_capture.py handles OCR + extraction end-to-end.
    • •If OCR quality is low, ask the teacher for a clearer scan or export to PDF.
  3. •

    Build artifacts

    • •Save raw OCR output under ocr/ and cleaned text under text/.
    • •Keep page references for traceability.
  4. •

    Extract example questions

    • •Split by numbering and option patterns (A/B/C/D).
    • •Create examples.csv with example_id, stem_text, options, source_ref, page.
    • •See references/extract_rules.md for heuristics.
  5. •

    Summarize classroom content

    • •Parse teacher notes (or request confirmation) into a structured class_discussion.md.
    • •Use template in references/discussion_template.md.
  6. •

    Knowledge point candidates

    • •Propose KP candidates based on examples.
    • •Require teacher confirmation before updating data/knowledge/knowledge_point_map.csv.
  7. •

    Lesson-first post-class diagnostic

    • •If requested, call generate_postclass_diagnostic.py with --discussion-notes and --student-notes.
    • •Only merge exam data when explicitly requested.
  8. •

    Write-back rules

    • •Never store raw student scores here.
    • •Only write confirmed summaries to mem0 using the teacher memory template.

Output Templates

  • •class_discussion.md (structured classroom summary)
  • •lesson_summary.md (lesson digest + key misconceptions)
  • •examples.csv (example questions index)
  • •lesson_review.md (visual review report for OCR + examples)

References

  • •references/lesson_io.md
  • •references/ocr_pipeline.md
  • •references/extract_rules.md
  • •references/discussion_template.md
  • •(Related) skills/physics-core-examples/SKILL.md

CLI (Quick Start)

bash
python3 skills/physics-lesson-capture/scripts/lesson_capture.py \\
  --lesson-id L2403_2026-02-04 \\
  --topic \"静电场综合\" \\
  --sources /path/to/lesson.pdf /path/to/example.png \\
  --discussion-notes /path/to/class_discussion.md

Optional:

  • •--lesson-plan to include teacher plan (PDF/DOCX/MD)
  • •--force-ocr to OCR even text-based PDFs
  • •--ocr-mode FREE_OCR|GROUNDING|OCR_IMAGE

Review report:

bash
python3 skills/physics-lesson-capture/scripts/lesson_review_report.py \\
  --lesson-id L2403_2026-02-04 \\
  --render-pdf

Confirm KP candidates from the review:

bash
python3 skills/physics-lesson-capture/scripts/confirm_kp_candidates.py \\
  --lesson-id L2403_2026-02-04

Apply confirmed KP candidates to knowledge_point_map.csv:

bash
python3 skills/physics-lesson-capture/scripts/apply_kp_to_question_map.py \\
  --lesson-id L2403_2026-02-04