lark-vc

vc (v1)

CRITICAL, Before starting, you MUST use the Read tool to read ../lark-shared/SKILL.md, which contains authentication and permission handling.

CRITICAL, Before starting, you MUST use the Read tool to read references/vc-domain-boundaries.md. Failure to read this will lead to errors in command usage, meeting artifact decision-making, and domain boundary responsibility judgment:

1. Understand the relationship and division of responsibilities between Calendar & VC, and Meeting Artifacts & Documents.
2. Understand the relationship between meeting artifacts (Minutes and Notes), for example: Minutes and Notes are generated independently.
3. Understand the components of different meeting artifacts to decide which artifact data to use based on requirements.
4. Understand the standard process for meeting summarization, analysis, and information extraction.

Identity

All vc commands use --as user by default. +search and meeting get also support --as bot.

# BAD, Use calendar to check yesterday's meetings, as it will miss instant meetings
lark-cli calendar events search_event --query "standup" --start-time ...

# GOOD, Use vc +search to check finished meetings
lark-cli vc +search --query "standup" --start-time ...

Shortcuts (Recommended for priority use)

• Before using any Shortcut, you must read its corresponding reference document.

Intent Routing

Core Concepts

• Meeting: Lark video meeting instance, identified by meeting_id. Finished meetings support searching by keywords, time range, participants, organizer, meeting room, etc. (see +search).
• Note: Structured document generated after a video meeting, identified by note_id, containing note documents (summary, to-dos) and verbatim transcript documents. note_display_type distinguishes between normal notes (normal) and unified notes. For direct queries with a known note_id and unified original records, use lark-note.
• Minutes: Recording artifacts from Lark video meetings or audio/video files uploaded by users. Supports transcription of video/audio, including summaries, to-dos, chapters, and text records, identified by minute_token.
• MainDoc: The main document of AI intelligent notes, containing AI-generated summaries and to-dos, corresponding to note_doc_token.
• MeetingNotes: Note documents actively bound to the meeting by the user, corresponding to meeting_notes. Only returned via the --calendar-event-ids path.
• VerbatimDoc: Sentence-by-sentence text record of the meeting, including speaker and timestamp.

Artifact Selection Decision

Verbatim Transcript Routing: First check the note_display_type returned by vc +notes, do not just check if verbatim_doc_token is empty. Specific routing is subject to +notes and lark-note.

Why must "extraction/summarization" start from the original conversation record? AI notes are the model's secondary compression of the meeting and may omit discussion details, debate processes, and implicit decisions. When a user requests "extraction" or "re-summarization", they expect an independent analysis based on the original conversation, not a re-formatting of AI artifacts.

Core Scenarios

1. Search Meeting Records

1. Only supports searching finished meetings. For future meetings that have not started, use the lark-calendar skill.
2. Only supports searching meeting records using filter conditions such as keywords, time range, participants, organizer, meeting room, etc. For unsupported filter conditions, prompt the user.
3. When search results contain multiple pieces of data, be sure to pay attention to paginated data acquisition and do not miss any meeting records.
4. When there is only a natural language note title and no meeting clues, do not treat the title as a meeting keyword; switch to document search according to the intent routing above.

2. Organize Meeting Notes

Before choosing which artifact to read, confirm that you understand the difference between the AI summary link and the recording link. If unsure, read references/vc-domain-boundaries.md.

1. When organizing note documents, it is default to provide the note document, verbatim transcript, and Minutes link; there is no need to read the content of the note document or verbatim transcript.
2. When the user explicitly needs to obtain summaries, to-dos, or chapter artifacts, then read the document to obtain specific content.
3. When reading the content of intelligent notes (note_doc_token), the first <whiteboard> tag in the note document is the cover image (AI-generated summary visualization), which should be downloaded and displayed to the user:

# 1. Read note content
lark-cli docs +fetch --api-version v2 --doc <note_doc_token> --doc-format markdown
# 2. Extract the token of the first <whiteboard token="xxx"/> from the returned markdown
# 3. Download the cover image to the aggregation directory (same directory as verbatim transcript and recording, keep artifacts together)
#    Not all notes have a cover whiteboard; skip if there is no <whiteboard> tag
lark-cli docs +media-download --type whiteboard --token <whiteboard_token> --output ./minutes/<minute_token>/cover

Artifact Directory Specification: All downloaded artifacts for the same meeting (recording, verbatim transcript, cover image, etc.) are unified into the ./minutes/{minute_token}/ directory. This is consistent with the default landing points of minutes +download and vc +notes --minute-tokens, facilitating Agent aggregation. Explicit paths (such as cover images) need to be manually aligned to the same directory.

Note-related Documents, Choose based on user intent:

• note_doc_token → AI Intelligent Notes (AI summary + to-dos)
• meeting_notes → User-bound meeting notes (documents actively associated with the meeting by the user, only returned via --calendar-event-ids path)
• When the user says "verbatim transcript", "complete record", "who said what" → Route according to note_display_type, see +notes for details
• When the user says "notes", "summary", "note content", return both note_doc_token and meeting_notes (if any)
• When user intent is unclear, display all document links for the user to choose, rather than deciding for the user
• If the user provides a local audio/video file and says "convert to notes" or "convert to verbatim transcript", do not start directly from vc +notes; first use minutes +upload to generate minute_url, then extract minute_token to call vc +notes --minute-tokens

3. Note Documents and Verbatim Transcript Links

1. Note documents, verbatim transcript documents, and associated shared documents are returned using document Tokens by default.
2. When only basic information such as document name and URL is needed, use lark-cli drive metas batch_query to query.

# Learn command usage
lark-cli schema drive.metas.batch_query

# Batch get basic document information: query up to 10 documents at a time
lark-cli drive metas batch_query --data '{"request_docs": [{"doc_type": "docx", "doc_token": "<doc_token>"}], "with_url": true}'

3. When document content is needed, use lark-cli docs +fetch --api-version v2.

# Get document content
lark-cli docs +fetch --api-version v2 --doc <doc_token> --doc-format markdown

4. Query Participant Snapshot (Read Operation)

When a user asks questions about participant snapshots such as "who attended this meeting", "who are the participants in this meeting", or "did so-and-so attend", use vc meeting get --with-participants: This is the participant server-side snapshot API, which does not rely on the bot identity to join the meeting, and finished meetings can also be queried:

lark-cli vc meeting get --params '{"meeting_id":"<meeting_id>","with_participants":true}'

Selection Judgment Table:

Resource Relationships

Meeting
├── Note ← identified by note_id, note_display_type: normal / unified
│   ├── MainDoc (AI Intelligent Note Document, note_doc_token)
│   ├── MeetingNotes (User-bound meeting note document, meeting_notes)
│   ├── VerbatimDoc (Verbatim Transcript, verbatim_doc_token) ← normal path
│   ├── UnifiedTranscript (Unified original record) ← unified path, note +transcript (lark-note)
│   └── SharedDoc (In-meeting shared document)
└── Minutes ← identified by minute_token, +recording obtained from meeting_id
    ├── Transcript (Text record)
    ├── Summary (Summary)
    ├── Todos (To-dos)
    ├── Chapters (Chapters)
    └── Keywords (Recommended keywords)

Minutes Boundary: +notes is responsible for note content, verbatim transcripts, and AI artifacts; for basic Minutes information, please prioritize +recording and lark-minutes.

Note Domain Boundary: vc +notes is the entry point for locating notes from meeting clues (meeting_id / calendar_event_id / minute_token), returning note_id and note_display_type.

• Already have note_id → lark-note.
• Already have doc_token and the goal is to read the body → lark-doc.
• Only have natural language note title → Document search / Docx body reading; only enter lark-note if there is an explicit vc-node-id.

API Resources

lark-cli vc <resource> <method> [flags]

meeting

• get, Get meeting details (topic, time, participants, note_id)

# Get basic meeting information (excluding participants)
lark-cli vc meeting get --params '{"meeting_id": "<meeting_id>"}'

# Get basic meeting information (including participants)
lark-cli vc meeting get --params '{"meeting_id": "<meeting_id>", "with_participants": true}'

minutes (Cross-domain, see lark-minutes for details)

• get, Get basic Minutes information (title, duration, cover); to query Minutes content, use +notes --minute-tokens <minute-token>

Outside the Scope of This Skill

• Query future meeting schedules → lark-calendar
• Agent real join/leave, real-time in-meeting events → lark-vc-agent
• Verbatim transcript query with only note document title → Document search / Docx body reading; only enter lark-note if there is an explicit vc-node-id
• Convert local audio/video files to notes/transcripts → lark-minutes (return to vc +notes after upload)
• Minutes search/download/upload/rename/replace speaker → lark-minutes