buaa-classroom-summarizer

Compare original and translation side by side

🇺🇸

Original

English

🇨🇳

Translation

Chinese

BUAA Classroom Summarizer

BUAA课堂总结工具

Use this skill for:

one
```
classroom.msa.buaa.edu.cn/livingroom
```
replay URL
one
```
classroom.msa.buaa.edu.cn/coursedetail
```
course page

Assume commands run from this skill root. Otherwise use the absolute path to

scripts/

使用本工具处理以下场景：

单个
```
classroom.msa.buaa.edu.cn/livingroom
```
回放链接
单个
```
classroom.msa.buaa.edu.cn/coursedetail
```
课程页面

假设命令从本工具的根目录运行。否则请使用

scripts/

的绝对路径。

Core Boundary

核心职责划分

Let scripts handle authentication, extraction, replay diagnosis, caching, and artifact writes.
Let the agent handle course alignment, concept confirmation, terminology correction, and final prose reconstruction.
Treat deterministic note output as a seed unless semantic rebuild is explicitly completed.

脚本负责处理认证、提取、回放诊断、缓存和资源写入。
Agent负责课程对齐、概念确认、术语修正和最终文本重构。
除非明确完成语义重建，否则将确定性笔记输出视为初始版本。

Main Commands

主要命令

Single replay extraction:

powershell

python scripts\extract_buaa_classroom.py "<livingroom-url>" --output-dir "<output-dir>"

Whole-course replay enumeration or extraction:

powershell

python scripts\collect_buaa_course_replays.py "<coursedetail-url>" --output-dir "<output-dir>"
python scripts\collect_buaa_course_replays.py "<coursedetail-url>" --output-dir "<output-dir>" --extract-existing --skip-existing

Whole-course commands are extraction and inventory commands, not permission to generate final notes for every available lesson. For a

coursedetail

URL, enumerate or extract artifacts first, then ask the user which lesson to semantically rebuild next unless the user explicitly requests a batch finalization workflow.

单回放提取：

powershell

python scripts\extract_buaa_classroom.py "<livingroom-url>" --output-dir "<output-dir>"

全课程回放枚举或提取：

powershell

python scripts\collect_buaa_course_replays.py "<coursedetail-url>" --output-dir "<output-dir>"
python scripts\collect_buaa_course_replays.py "<coursedetail-url>" --output-dir "<output-dir>" --extract-existing --skip-existing

全课程命令仅用于提取和盘点，不代表可以为所有可用课程生成最终笔记。对于

coursedetail

链接，需先枚举或提取资源，然后询问用户下一步要对哪个课程进行语义重建，除非用户明确要求批量生成流程。

Course Identity Rule

课程标识规则

For a

coursedetail

URL, resolve the course identity before choosing a course folder or reusing vault context:

First extract the strongest available course title from classroom/SPoC metadata, saved course detail files, replay metadata, or stable replay titles.
Treat the normalized course title as the canonical course identity. If the title matches an existing course, it is the same course even when
```
course_id
```
, lecturer, class time, classroom, or sub_id ranges differ.
If no reliable title is available, use
```
course_id
```
only as a provisional identity such as
```
course-136278
```
; do not merge into an existing titled course by teacher, schedule, lesson dates, old extraction directories, or nearby vault content.
Preserve
```
course_id
```
and source URLs as source metadata, but do not use them to split courses that have the same confirmed title.
If title extraction is ambiguous and the target vault already has plausible course folders, pause for user confirmation before writing formal notes or updating trackers.

Runtime browser auth when local cookie reuse is unreliable:

powershell

python scripts\extract_buaa_classroom.py "<livingroom-url>" --output-dir "<output-dir>" --browser-runtime-auth --browser-channel "auto"

对于

coursedetail

链接，在选择课程文件夹或复用存储上下文前，需先确定课程标识：

首先从课堂/SPoC元数据、已保存的课程详情文件、回放元数据或稳定的回放标题中提取最明确的课程标题。
将标准化后的课程标题视为正式课程标识。若标题与现有课程匹配，则即使
```
course_id
```
、讲师、上课时间、教室或
```
sub_id
```
范围不同，也视为同一课程。
若没有可靠的标题可用，仅使用
```
course_id
```
作为临时标识（如
```
course-136278
```
）；请勿根据教师、日程、课程日期、旧提取目录或附近存储内容将其合并到已有标题的课程中。
保留
```
course_id
```
和源链接作为源元数据，但请勿用它们拆分具有相同确认标题的课程。
若标题提取存在歧义，且目标存储中已有合理的课程文件夹，在写入正式笔记或更新跟踪器前需暂停并等待用户确认。

当本地Cookie复用不可靠时，使用运行时浏览器认证：

powershell

python scripts\extract_buaa_classroom.py "<livingroom-url>" --output-dir "<output-dir>" --browser-runtime-auth --browser-channel "auto"

Required Replay Diagnosis

必需的回放诊断

Before building any note, the script must produce one

replay_diagnosis

and route the replay into exactly one of:

```
waiting_transcript
```
```
partial_transcript
```
```
transcript_only
```

Downstream note logic must consume this diagnosis instead of recomputing route decisions elsewhere.

在生成任何笔记前，脚本必须生成一份

replay_diagnosis

，并将回放归入以下类别之一：

```
waiting_transcript
```
（等待转录文件）
```
partial_transcript
```
（部分转录）
```
transcript_only
```
（仅转录文件）

下游笔记逻辑必须使用该诊断结果，不得在其他地方重新计算路由决策。

Standalone Markdown Note Workflow

独立Markdown笔记工作流

Prepare a semantic rebuild packet:

powershell

python scripts\extract_buaa_classroom.py "<livingroom-url>" --output-dir "<output-dir>" --export-markdown-note

Preferred user-facing modes:

powershell

python scripts\extract_buaa_classroom.py "<livingroom-url>" --output-dir "<output-dir>" --export-markdown-note --markdown-note-mode "final-lite"
python scripts\extract_buaa_classroom.py "<livingroom-url>" --output-dir "<output-dir>" --export-markdown-note --markdown-note-mode "final-explained"

These modes must write only:

semantic_rebuild/semantic_rebuild_input.json

semantic_rebuild/semantic_rebuild_prompt.md

All final-oriented modes, including legacy

final

, must write only the semantic packet. Do not emit

lesson_note.md

by default. Treat the packet as the only intermediate artifact, then let the agent produce the final note.

Before accepting an agent-written note as final, run:

powershell

python scripts\validate_final_note.py "<final-note.md>"

If the validator fails, keep the extraction artifacts and semantic packet only. Do not rename or present the failed note as a final note.

Then create a reviewer packet:

powershell

python scripts\review_final_note.py --note "<final-note.md>" --semantic-input "<semantic_rebuild_input.json>" --output-dir "<review-dir>"

Use

final_note_review/final_note_review_prompt.md

with an independent reviewer agent only when the active system/developer instructions allow spawning one. If subagents are unavailable or not allowed, run a separate reviewer pass yourself with the same prompt, write the result as

final_note_review/final_note_review_result.json

, and do not edit the note during review.

When the agent writes the final standalone Markdown note:

use a readable lesson filename, preferably the lesson title such as
```
2026-04-13 贝叶斯统计 第7周星期1第3,4,5节.md
```
do not name the final note
```
lesson_note.md
```
place final lesson Markdown notes for the same course in one course folder named by course title, for example
```
贝叶斯统计/2026-04-13 贝叶斯统计 第7周星期1第3,4,5节.md
```
if the chosen output directory is already named exactly as the course title, write final lesson Markdown notes directly in that directory; do not create
```
课程名/课程名/...
```
keep extraction artifacts such as
```
metadata.json
```
,
```
transcript.json
```
, and semantic rebuild packets in their original replay output directories; only user-facing final Markdown notes need the course-folder layout
start directly with the lesson title and content
do not show production metadata such as
```
状态
```
,
```
来源
```
, transcript coverage, replay diagnosis, or PPT extraction status in the user-facing note
pass
```
scripts\validate_final_note.py
```
before the note is called final
pass the reviewer gate for the current file hash before the note is called final

准备语义重建数据包：

powershell

python scripts\extract_buaa_classroom.py "<livingroom-url>" --output-dir "<output-dir>" --export-markdown-note

推荐的面向用户模式：

powershell

python scripts\extract_buaa_classroom.py "<livingroom-url>" --output-dir "<output-dir>" --export-markdown-note --markdown-note-mode "final-lite"
python scripts\extract_buaa_classroom.py "<livingroom-url>" --output-dir "<output-dir>" --export-markdown-note --markdown-note-mode "final-explained"

这些模式必须仅写入以下文件：

semantic_rebuild/semantic_rebuild_input.json

semantic_rebuild/semantic_rebuild_prompt.md

所有面向最终版本的模式（包括旧版

final

）必须仅写入语义数据包。默认情况下不得生成

lesson_note.md

。将数据包视为唯一的中间资源，然后由Agent生成最终笔记。

在接受Agent撰写的笔记作为最终版本前，运行：

powershell

python scripts\validate_final_note.py "<final-note.md>"

若验证失败，仅保留提取资源和语义数据包。不得重命名或展示失败的笔记作为最终版本。

然后生成审核数据包：

powershell

python scripts\review_final_note.py --note "<final-note.md>" --semantic-input "<semantic_rebuild_input.json>" --output-dir "<review-dir>"

仅当当前系统/开发者指令允许生成子Agent时，才使用

final_note_review/final_note_review_prompt.md

与独立审核Agent配合。若子Agent不可用或不被允许，请在主Agent中单独执行审核步骤，将结果写入

final_note_review/final_note_review_result.json

，且审核期间不得编辑笔记。

当Agent撰写最终独立Markdown笔记时：

使用易读的课程文件名，优先使用课程标题，例如
```
2026-04-13 贝叶斯统计 第7周星期1第3,4,5节.md
```
不得将最终笔记命名为
```
lesson_note.md
```
将同一课程的最终Markdown笔记放在以课程标题命名的文件夹中，例如
```
贝叶斯统计/2026-04-13 贝叶斯统计 第7周星期1第3,4,5节.md
```
若所选输出目录已与课程标题完全一致，则直接将最终Markdown笔记写入该目录；不得创建
```
课程名/课程名/...
```
的嵌套结构
将
```
metadata.json
```
、
```
transcript.json
```
等提取资源以及语义重建数据包保留在其原始回放输出目录中；仅面向用户的最终Markdown笔记需要课程文件夹布局
直接以课程标题和内容开头
不得在面向用户的笔记中显示
```
状态
```
、
```
来源
```
、转录覆盖率、回放诊断或PPT提取状态等生产元数据
在将笔记标记为最终版本前，必须通过
```
scripts\validate_final_note.py
```
验证
在将笔记标记为最终版本前，必须通过针对当前文件哈希的审核关卡

Batch Finalization Rule

批量生成规则

Whole-course extraction is still not permission to blindly finalize every replay. A whole-course run may produce:

replay inventory
extraction artifacts
semantic rebuild packets
a course-level todo list

If the user explicitly asks to process all currently pending lessons, batch finalization is allowed, but it must remain lesson-by-lesson inside the batch:

Skip lessons already finalized unless the user asks for a revision.
For each candidate lesson, verify the transcript exists and is non-empty before authoring.
Read the full transcript plus the semantic packet before writing that lesson.
Run
```
validate_final_note.py
```
, create the review packet, and record a pass result for the current note hash before calling that lesson final.
If a lesson fails any gate, leave only artifacts/packet or a review-gated draft and continue with other eligible lessons.
Prefer running tracker/overview maintenance once after the batch, not after every lesson, unless an intermediate checkpoint is needed.
Reuse existing extraction artifacts and semantic packets when their inputs have not changed; do not rerun browser extraction just to rebuild prose.

全课程提取不代表可以盲目为所有回放生成最终版本。全课程运行可能生成：

回放清单
提取资源
语义重建数据包
课程级待办事项列表

若用户明确要求处理所有当前待处理课程，则允许批量生成，但必须在批量处理中逐课进行：

除非用户要求修订，否则跳过已生成最终版本的课程。
对于每个候选课程，在撰写前验证转录文件是否存在且非空。
在撰写该课程笔记前，读取完整转录文件和语义数据包。
在将课程标记为最终版本前，运行
```
validate_final_note.py
```
、生成审核数据包并记录当前笔记哈希的通过结果。
若某课程未通过任何关卡，仅保留资源/数据包或审核 gated 的草稿，继续处理其他符合条件的课程。
优先在批量处理完成后再维护跟踪器/概览，而非每处理完一课就维护一次，除非需要中间检查点。
当输入未更改时，复用现有提取资源和语义数据包；不得仅为重构文本而重新运行浏览器提取。

Final Note Quality Gate

最终笔记质量关卡

A user-facing final note must be a semantic reconstruction, not a decorated transcript segment list. Before writing or accepting a final note, reject it if it contains any of these patterns:

raw ASR/OCR snippets presented as "representative expressions" or "代表性表达"
headings such as
```
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 mathematical content
transcript noise such as misrecognized symbols copied into the note without correction
a course overview that marks low-quality diagnostics as "正式笔记"

If a note fails this gate, keep only the extraction artifacts and semantic packet, then mark the lesson as needing semantic rebuild. Do not call it final.

The semantic packet must not contain user-facing seed prose such as

seed_bullets

, raw

sample_lines

, or

transcript_excerpt

. It may contain time windows and paths to the transcript; the agent must read the transcript itself and reconstruct the note semantically.

面向用户的最终笔记必须是语义重构，而非经过装饰的转录片段列表。在撰写或接受最终笔记前，若包含以下任何模式则予以拒绝：

原始ASR/OCR片段被标注为“representative expressions”或“代表性表达”
类似
```
课堂讲解与主题推进 1
```
的标题

诸如

整理时建议不要把这一段只当作...

的模板内容

各章节重复的通用建议，而非课程特定的数学内容
转录噪声，例如未经修正的识别错误符号被复制到笔记中
将低质量诊断标记为“正式笔记”的课程概览

若笔记未通过此关卡，仅保留提取资源和语义数据包，然后将该课程标记为需要语义重建。不得将其称为最终版本。

语义数据包不得包含面向用户的初始文本，如

seed_bullets

、原始

sample_lines

或

transcript_excerpt

。它可以包含时间窗口和转录文件路径；Agent必须自行读取转录文件并语义重构笔记。

Reviewer Gate

审核关卡

Finalization requires both gates on the current Markdown bytes:

```
scripts\validate_final_note.py
```
passes.

The independent reviewer returns

decision=pass

finalization_allowed=true

, and

reviewed_note_sha256

equal to

final_note_review_input.json

note.sha256

If the note changes after either gate, both gate results are invalid and must be rerun.

Reviewer implementation detail:

If subagents are permitted, use an independent reviewer agent.
If subagents are not permitted by active instructions, run a separate reviewer pass in the main agent, write
```
final_note_review_result.json
```
, and ensure
```
reviewed_note_sha256
```
matches
```
final_note_review_input.json
```
.
Do not rerun review for an unchanged note when an existing
```
final_note_review_result.json
```
already passes for the same hash.

Reviewer decisions:

```
pass
```
: the note faithfully covers the transcript, handles course-domain substance, preserves supported affairs/emphasis, and is safe to present as final.
```
needs_revision
```
: the transcript can support a final note, but the current note misses supported content, is too generic, or needs correction. Revise, rerun hard gate, then rerun reviewer.
```
reject
```
: the current source material or note is not fit for finalization. Keep extraction artifacts and semantic packet; do not present a final note.

Absence is not failure. Missing homework, exam, grading, or deadline information is only a problem when the transcript contains evidence for it and the note omits, distorts, or invents it. If the transcript shows early dismissal, in-class exercise, student presentation, discussion, or a logistics-only class, the note may be short but must faithfully describe what happened.

最终版本需要针对当前Markdown字节同时通过两个关卡：

```
scripts\validate_final_note.py
```
验证通过。

独立审核者返回

decision=pass

、

finalization_allowed=true

，且

reviewed_note_sha256

与

final_note_review_input.json

中的

note.sha256

相等。

若笔记在任一关卡后发生更改，则两个关卡结果均无效，必须重新运行。

审核者实现细节：

若允许子Agent，则使用独立审核Agent。
若当前指令不允许子Agent，则在主Agent中单独执行审核步骤，写入
```
final_note_review_result.json
```
，并确保
```
reviewed_note_sha256
```
与
```
final_note_review_input.json
```
匹配。
当现有
```
final_note_review_result.json
```
已针对同一哈希通过审核时，无需对未更改的笔记重新审核。

审核者决策：

```
pass
```
：笔记准确覆盖转录内容，处理课程领域实质内容，保留重点信息，可安全作为最终版本展示。
```
needs_revision
```
：转录内容可支持生成最终笔记，但当前笔记遗漏内容、过于通用或需要修正。修订后重新运行硬性关卡，再重新审核。
```
reject
```
：当前源材料或笔记不适合生成最终版本。保留提取资源和语义数据包；不得展示最终笔记。

缺失内容不代表失败。缺少作业、考试、评分或截止日期信息仅在转录内容中有相关证据但笔记遗漏、歪曲或编造时才视为问题。若转录显示提前下课、课堂练习、学生展示、讨论或仅涉及后勤的课程，笔记可以简短，但必须如实描述实际情况。

Semantic Rebuild Rules

语义重建规则

Perform a course-alignment check before accepting the rewrite as final.
Correct obvious ASR/OCR term errors when the course context makes the intended term clear.
Keep the lesson time axis visible. Each final section should keep a packet time range or a 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 course-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 even if a tracker lists it under backlog. Do not create a formal note from metadata, schedule, title, or PPT alone.

在接受重写版本作为最终版本前，执行课程对齐检查。
当课程上下文明确预期术语时，修正明显的ASR/OCR术语错误。
保持课程时间轴可见。每个最终章节应保留数据包时间范围或粗略的
```
MM:SS-MM:SS
```
标记。
数学公式仅使用
```
$...$
```
或
```
$$...$$
```
格式。不得用反引号包裹公式。
将课程转录文件作为章节边界、课程主线和完成检查的唯一主要来源。
仅当课程转录覆盖率和总结覆盖率均通过时，才将课程标记为最终版本。
重构课程特定内容。不得用通用学习建议替代缺失的语义理解。
若
```
transcript.txt
```
缺失、为空或接近为空，即使跟踪器将其列为待处理，也将该回放视为等待转录材料。不得仅根据元数据、日程、标题或PPT创建正式笔记。

Authoring Contract

撰写约定

When writing the final student-facing Markdown note from a semantic packet:

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 such as “整理时应...”, “后续重写...”, “这一段主要在...”, or other process commentary.

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 or code relate to the design.
Humanities and social sciences: reconstruct concepts, arguments, historical or 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 or 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 or format constraints, grading criteria, and troubleshooting advice.

从语义数据包撰写面向学生的最终Markdown笔记时：

你撰写的是成品笔记，而非初始笔记、诊断笔记或给未来整理者的指令。
撰写前读取完整的
```
transcript.txt
```
。仅将
```
semantic_rebuild_input.json
```
作为元数据、时间锚点和资源索引使用。
不得展示证据片段、候选短语、OCR片段、原始ASR行或内部工作流笔记。
每个主要时间段应说明教学活动类型：定义、模型、论证、证明、示例、比较、案例讨论、政策解释、教师点评、作业布置、考试安排或课堂后勤。
捕捉高价值课堂信息：考试、作业、截止日期、提交格式、评分权重、阅读要求、教师强调的重点、反复提及的短语、公式、定理、定义、示例和常见错误。
若教师明确说明某内容重要、可能会考、易混淆、常出错或课后需要复习，请在笔记中保留该信息。
若转录证据不足，请将该内容放在
```
待核对
```
部分，而非转化为确定结论。
最终笔记必须直接面向学生读者。避免使用“整理时应...”、“后续重写...”、“这一段主要在...”等流程说明类短语。

课程领域重构指南：

数学与统计学：重构对象、定义、假设、方程、定理、证明思路、示例、反例、符号含义以及结果之间的关联。
工程与计算机科学：重构系统组件、算法、设计约束、实现步骤、实验设置、故障案例、权衡方案以及公式或代码与设计的关联。
人文与社会科学：重构概念、论证、历史或制度背景、作者立场、证据、比较、案例以及教师的评价重点。
思想政治课程：重构政策概念、理论主张、历史背景、指定文件或事件、价值判断、应试表述以及用于解释抽象主张的示例。
语言、写作与沟通课程：重构词汇、修辞模式、文本结构、示例、修正要点、练习要求以及教师反馈。
实验、设计或项目课程：重构任务目标、交付成果、工具、操作步骤、数据要求、安全或格式约束、评分标准以及故障排除建议。

Transcript-Only Rule

仅转录文件规则

When

replay_diagnosis=transcript_only

do not emit fake content templates such as “课程定位 / 基础概念 / 方法流程”
let scripts provide only time segments from the course transcript, representative transcript lines, and
```
transcript_overview
```
let the agent infer the real lesson structure from the course transcript plus course context
do not ask scripts to pre-confirm concepts from transcript-only material

当

replay_diagnosis=transcript_only

时：

不得生成“课程定位 / 基础概念 / 方法流程”等虚假内容模板
仅让脚本提供课程转录文件的时间段、代表性转录行和
```
transcript_overview
```
让Agent从课程转录文件和课程上下文推断真实的课程结构
不得要求脚本从仅转录材料中预先确认概念

PPT Rule

PPT规则

Prefer teacher stream by default.
Treat PPT as auxiliary only, even when a PPT stream exists.
PPT may help with term spelling, page or book titles, formula symbols, and logistics screenshots.
PPT must not decide section boundaries, lesson mainline, concept generation, or completion state.

默认优先使用教师流内容。
即使存在PPT流，也仅将PPT视为辅助资源。
PPT可用于辅助术语拼写、页面或书籍标题、公式符号和后勤截图。
PPT不得决定章节边界、课程主线、概念生成或完成状态。

Logistics-Only Teacher Review

仅后勤内容的教师回顾

If the user only wants follow-up on assignments, exams, notices, or arrangements:

powershell

python scripts\extract_buaa_classroom.py "<livingroom-url>" --output-dir "<output-dir>" --export-markdown-note --lightweight-teacher-review

This mode prepares short teacher-stream review clips and

teacher_review.json

. It should not silently rewrite conclusions into the note until a later confirmation step marks them as confirmed.

若用户仅关注作业、考试、通知或安排的后续信息：

powershell

python scripts\extract_buaa_classroom.py "<livingroom-url>" --output-dir "<output-dir>" --export-markdown-note --lightweight-teacher-review

此模式准备简短的教师流回顾片段和

teacher_review.json

。在后续确认步骤标记结论为已确认前，不得将结论悄悄写入笔记。

Failure Rules

失败处理规则

If the course transcript is missing, keep extraction artifacts but do not invent a formal lesson note.
If the course transcript is empty or near-empty, handle it the same as missing transcript: keep it in waiting/backlog and do not write a final note.
If course-transcript coverage is clearly partial, keep only a diagnostic draft rather than a final note.
If the course transcript exists but the current summary only covers an early slice of the lesson or leaves large uncovered gaps, mark the note
```
needs_review
```
instead of final.
If session reuse fails, rerun with
```
--browser-runtime-auth
```
.

On Windows, prefer a UTF-8 shell when validating generated files. If needed, set

[Console]::InputEncoding

and

[Console]::OutputEncoding

to UTF-8 before manual

Get-Content

or other console inspection. For inline Python in PowerShell, use:

powershell

@'
print("hello")
'@ | python -

Do not use Bash heredoc syntax such as

python - <<'PY'

in PowerShell.

若课程转录文件缺失，保留提取资源但不得编造正式课程笔记。
若课程转录文件为空或接近为空，处理方式与缺失转录文件相同：将其列为等待/待处理，不得写入最终笔记。
若课程转录覆盖率明显不完整，仅保留诊断草稿而非最终笔记。
若课程转录文件存在，但当前总结仅覆盖课程的前一部分或存在大量未覆盖的空白，将笔记标记为
```
needs_review
```
而非最终版本。
若会话复用失败，使用
```
--browser-runtime-auth
```
重新运行。

在Windows系统中，验证生成文件时优先使用UTF-8编码的Shell。若需要，在手动执行

Get-Content

或其他控制台检查前，将

[Console]::InputEncoding

和

[Console]::OutputEncoding

设置为UTF-8。在PowerShell中使用内联Python时，请使用：

powershell

@'
print("hello")
'@ | python -

请勿在PowerShell中使用Bash heredoc语法如

python - <<'PY'

。