| name | nuaa-vod-summarizer |
| description | Extract NUAA Feitian Cloud Classroom VOD replay metadata, lesson lists, subtitles, playback URLs, and courseware artifacts from ft.nuaa.edu.cn video-detail pages using a browser login session. |
NUAA VOD Summarizer
Use this skill for NUAA Feitian Cloud Classroom pages. Both URL formats are supported:
https://ft.nuaa.edu.cn/jy-application-vod-he-ui/#/video-detail?id=118467
https://ft.nuaa.edu.cn/jy-application-resourcemanage-ui/#/play-center?teclId=118690
The API base (/jy-application-vod-he or /jy-application-resourcemanage) is auto-detected from the URL.
Assume commands run from this skill root. Otherwise use the absolute path to scripts/.
Core Boundary
- Let scripts handle browser login reuse, API discovery, replay metadata extraction, subtitle export, and artifact writes.
- Let the agent handle transcript diagnosis, terminology correction, semantic reconstruction, and final Markdown prose.
- Deterministic seed notes must not masquerade as finished course notes.
- Do not call a note final when subtitles/transcripts are missing, empty, or obviously partial.
Main Commands
Probe one lesson (quick test)
python scripts\extract_nuaa_vod.py "<video-detail-url>" --output-dir "<output-dir>" --browser-runtime-auth --manual-wait-seconds 60
Extract ALL transferable lessons (batch)
python scripts\extract_nuaa_vod.py "<video-detail-url>" --output-dir "<output-dir>" --browser-runtime-auth --manual-wait-seconds 60 --batch --skip-existing
Extract a specific lesson by ID
python scripts\extract_nuaa_vod.py "<video-detail-url>" --output-dir "<output-dir>" --subtitle-course-id 1547370 --manual-wait-seconds 10
Output Structure
<output-dir>/
course.json # course-level metadata
replay_inventory.json # full lesson list with transfer flags
raw/
page.json # parsed URL info
captured_responses.json # all captured API responses
storage_state.json # browser storage state
responses/ # individual response files
lessons/
<lesson_id>/
metadata.json # lesson metadata
raw/
subtitle.json # raw subtitle API response
transcript.txt # cleaned transcript with timestamps
semantic_rebuild/
semantic_rebuild_input.json # packet for semantic rebuild
semantic_rebuild_prompt.md # prompt template
note.md # generated Markdown note (after rebuild)
Supported Platforms
Two UI/API pairs are supported, auto-detected from the URL:
| UI host | API base | Main route |
|---|
ft.nuaa.edu.cn/jy-application-vod-he-ui | /jy-application-vod-he | #/video-detail?id=<teclId> |
ft.nuaa.edu.cn/jy-application-resourcemanage-ui | /jy-application-resourcemanage | #/play-center?teclId=<teclId> |
Known API families (both platforms share similar paths under their API base):
/v1/subject_vod_list — lesson list (vod-he)/v1/group_subject_vod_list — lesson list (resourcemanage)/v1/course_vod_videoinfos — lesson video metadata/v1/course_vod_urls — playback URLs/v1/course_vod_subtitle — AI subtitles (requires courTransferFlag=true)/v1/courseware/upload/list — courseware list/v1/courseware/downLoad — courseware download/v1/course/verify — course access check
Workflow
- Open the supplied URL (
video-detail or play-center) in a persistent Chromium context.
- If needed, complete NUAA login in the browser window.
- Capture NUAA VOD API responses while the page loads.
- Auto-detect API base from URL markers.
- Enumerate lessons and identify which have subtitles (
courTransferFlag).
- For each target lesson, fetch
<api_base>/v1/course_vod_subtitle.
- Save subtitle JSON and cleaned transcript per lesson.
- Prepare
semantic_rebuild/ packet only when transcript material exists.
- Generate Markdown notes from transcripts via semantic rebuild.
Required Replay Diagnosis
Before building any note, classify each lesson into exactly one of:
waiting_transcript — subtitles not yet available (no courTransferFlag, or fetch returned empty)partial_transcript — subtitle exists but is clearly incomplete, garbled beyond repair, or covers only a fraction of the lessontranscript_only — has complete transcript but no note written yet
Downstream note logic must consume this classification instead of recomputing route decisions.
Semantic Rebuild Rules
- Perform course-alignment check before treating a note as final.
- Correct obvious ASR term errors when course context makes the intended term clear.
- Keep the lesson time axis visible. Every major section should reference a time range or coarse
MM:SS-MM:SS marker.
- Keep math as
$...$ or $$...$$ only (do not wrap formulas in backticks).
- Treat the course transcript as the only primary source for section boundaries, lesson mainline, and completion checks.
- Only mark a lesson final when transcript coverage and summary coverage both pass.
- Reconstruct course-specific substance. Do not substitute generic learning advice for missing semantic understanding.
- If
transcript.txt is missing, empty, or near-empty, treat the replay as waiting for transcript material. Do not create a formal note from metadata, schedule, title, or courseware alone.
Authoring Contract
When writing the final student-facing Markdown note:
- You are writing the finished note, not a seed note, diagnostic note, or instruction to a future organizer.
- Read the full transcript.txt before writing. Use
semantic_rebuild_input.json only as metadata, time anchors, and artifact index.
- Do not expose evidence snippets, candidate phrases, OCR fragments, raw ASR lines, or internal workflow notes.
- Every major time block should explain what teaching move happened: definition, model, argument, proof, example, comparison, case discussion, policy explanation, teacher comment, assignment, exam arrangement, or class logistics.
- Capture high-value classroom signals: exams, homework, deadlines, submission format, grading weight, reading requirements, teacher-emphasized key points, repeatedly stressed phrases, formulas, theorems, definitions, examples, and common mistakes.
- If the teacher explicitly says something is important, likely to be tested, easy to confuse, often wrong, or needs review after class — preserve it in the note.
- If transcript evidence is weak, write the item under "待核对" instead of turning it into a confident conclusion.
- The final note must face the student reader directly. Avoid phrases like "整理时应...", "后续重写...", "这一段主要在...", or other process commentary.
Final Note Quality Gate
A user-facing final note must be a semantic reconstruction, not a decorated transcript segment list. Before accepting as final, reject if it contains:
- Raw ASR/OCR snippets presented as "代表性表达" or representative expressions
- Headings like "课堂讲解与主题推进 1"
- Boilerplate like "整理时建议不要把这一段只当作..."
- Repeated generic advice across sections instead of course-specific content
- Transcript noise (misrecognized symbols) copied without correction
- Low-quality diagnostics marked as formal notes
If a note fails this gate, keep only extraction artifacts and semantic packet, then mark the lesson as needing semantic rebuild. Do not call it final.
Course-Domain Reconstruction Guidance
- Math and statistics: reconstruct objects, definitions, assumptions, equations, theorems, proof ideas, examples, counterexamples, symbol meanings, and links between results.
- Engineering and computer science: reconstruct system components, algorithms, design constraints, implementation steps, experiment setup, failure cases, trade-offs, and how formulas/code relate to the design.
- Humanities and social sciences: reconstruct concepts, arguments, historical/institutional background, author positions, evidence, comparisons, cases, and the teacher's evaluative emphasis.
- Ideological and political courses: reconstruct policy concepts, theoretical claims, historical context, named documents/events, value judgments, exam-oriented formulations, and examples used to explain abstract claims.
- Language, writing, and communication courses: reconstruct vocabulary, rhetorical patterns, text structure, examples, correction points, practice requirements, and teacher feedback.
- Lab, design, or project courses: reconstruct task goals, deliverables, tools, operation steps, data requirements, safety/format constraints, grading criteria, and troubleshooting advice.
Transcript-Only Rule
When a lesson is classified as transcript_only:
- Do NOT emit fake content templates (e.g., "课程定位 / 基础概念 / 方法流程").
- Let scripts provide only time segments from the transcript, representative transcript lines, and transcript overview.
- Let the agent infer the real lesson structure from the transcript plus course context.
- Do not ask scripts to pre-confirm concepts from transcript-only material.
PPT / Courseware Rule
- Prefer teacher audio/subtitle stream by default.
- Treat PPT and courseware as auxiliary only, even when available.
- PPT may help with term spelling, page/book titles, formula symbols, and logistics screenshots.
- PPT must not decide section boundaries, lesson mainline, concept generation, or completion state.
Failure Rules
- If authentication fails or no VOD API responses are captured, keep
raw/page.json and explain that a logged-in browser session is required.
- If subtitle data is absent for a lesson, mark it as no-subtitle and skip.
- If subtitle text is empty or near-empty, do not produce a final note.
- If transcript coverage is clearly partial, keep only a diagnostic draft rather than a final note.
- Treat courseware/PPT as auxiliary only; it must not replace transcript evidence.
- If session reuse fails, rerun with
--browser-runtime-auth.
On Windows, prefer PowerShell commands. For inline Python, use:
@'
print("hello")
'@ | python -
