Clean Lecture Notes

Overview

Use this skill to turn noisy lecture transcripts, meeting transcripts, class recordings, Feishu/Lark docs, local ASR output, and optional courseware into reliable cleaned transcripts and review-ready study notes.

Default to preserving the original transcript. Unless the user explicitly asks to overwrite a source document, create new cleaned transcript and study-note artifacts.

Workflow

1. Identify the source:
   • Local transcript file: read it directly.
   • Local audio or video: use the user's available transcription or meeting-minutes tooling first, then continue from the transcript.
   • Feishu/Lark Minutes URL: if Feishu/Lark tools are available, fetch the transcript before cleaning.
   • Feishu/Lark docx transcript URL: if lark-doc is available, fetch the source doc before cleaning.
   • Courseware URL or file: extract its title, module order, core diagrams, and assignments; use it as the study-note spine.
2. For transcript cleanup only, read references/transcript-cleaning.md and produce a cleaned transcript.
3. For study notes, read both references/transcript-cleaning.md and references/study-note-workflow.md.
4. For Feishu/Lark output, read references/feishu-output.md; treat Feishu/Lark as optional integration, not a hard dependency.
5. Before final delivery, read references/quality-checklist.md.
6. When local note artifacts exist, run scripts/check_note_artifacts.py; for final study notes, include --final-note.

Output Contract

• Cleaned transcript: remove filler words, repeated fragments, ASR noise, broken punctuation, and fragmented paragraphs while preserving meaning, examples, names, numbers, assignments, caveats, and event order when relevant.
• Study notes: include a Mermaid mindmap or knowledge map, time-boxed review path, core conclusions, active recall questions, conceptual modules, practice/project migration items, interview-ready wording when useful, and sources.
• Emphasis: bold the first occurrence of core concepts, key methods, important metrics, and high-value sentences; do not bold entire paragraphs.
• Perspective: final study notes must use a learner review and transfer perspective. Do not write them as class minutes, a meeting recap, or "the teacher said..." narration.
• Pending information: cleaned transcripts may keep speaker labels and pending checks; final study notes must not include a standalone 待确认 section. Put unresolved factual uncertainty in the source and quality audit section.
• Feishu/Lark output: default to creating new docs, verify outlines after writing, and ensure the top 思维导图 section is a Mermaid flowchart or equivalent visual whiteboard rather than a plain text tree.

Optional Integrations

• Use Feishu/Lark tools only when they are available in the current Codex environment or the user explicitly asks for Feishu/Lark writeback.
• When Feishu/Lark tooling is unavailable, produce local Markdown artifacts and clearly state that remote writeback was skipped.
• For visual study-note overviews, prefer Mermaid flowchart TD or flowchart LR.

Safety Rules

• Never write secrets, API keys, access tokens, .env values, private Feishu/Lark links, or credentials into notes, logs, examples, or commits.
• Do not overwrite or delete source transcripts unless the user explicitly asks for it.
• Mark uncertain ASR corrections briefly instead of turning guesses into facts.
• Do not add analytics, telemetry, or unrelated network calls.

Final Response

Report:

• the local output paths or Feishu/Lark document URLs created,
• whether the source transcript remained unchanged,
• which validation commands were run,
• any unresolved terminology or source-quality caveats.