WorkBuddy Meeting Assistant

Common Pain Points in Meeting Minutes Preparation

After a meeting, organizing minutes is often a tedious process. Manually listening to recordings and typing text is time-consuming, and meeting records from different platforms like Tencent Meeting, Feishu, and DingTalk are scattered, requiring repeated switching and transferring. If teams use multiple collaboration tools, syncing minutes to DingTalk groups, Feishu documents, or WeChat Work adds extra steps. These pain points make it hard to retain meeting value and often lead to missed action items.

How the WorkBuddy Meeting Assistant Works

This skill acts as a pure orchestrator, automating the meeting minutes pipeline. It does not directly handle recording transcription but dispatches external tools for core tasks, focusing on scheduling, template processing, privacy gates, and deduplication.

Core Capabilities

• Multi-Source Extraction: Automatically pulls recordings or minutes from Tencent Meeting (tmeet), Feishu (lark-cli), DingTalk (dws), IMA (ima-mcp), or WeChat Work (wecom-cli). If the default source has no new meetings, it probes other sources and asks for user confirmation.
• Intelligent Processing: Features 15 scene templates (e.g., weekly sync, review, retrospective) that are auto-matched based on meeting title and content keywords. Templates generate structured minutes, including decisions and to-dos, with fact sources and confidence levels marked (✅/⚠️/❓).
• Privacy Gate: Provides a preview before distribution to prevent accidental sharing of sensitive content. For group message pushes (e.g., dingtalk-send, feishu-send), user confirmation is mandatory.
• One-Click Distribution: Supports pushing to DingTalk, Feishu (groups or documents), Tencent Docs, or WeChat Work, with deduplication based on source_id to avoid duplicate pushes for the same meeting.

Key Steps

1. Extraction: Uses tools like tmeet, dws, or lark-cli to pull raw data, producing raw_transcript and source_id. If a source is not ready, it guides authorization or degrades to manual paste.
2. Processing: Identifies the scene and applies a template. For example, saying "整理一下" triggers extraction and processing; saying "简版" switches to concise mode, outputting only to-dos and decisions.
3. Preview and Confirmation: At Step 4, it provides a minutes preview where users can modify the scene or content. This step is the privacy gate, ensuring sensitive information is not mis-sent.
4. Distribution: Executes push to the selected target platform. Document-creation operations (e.g., tencent-docs-create) bypass the gate, while group messages (e.g., dingtalk-send) require gate confirmation.

Applicable Boundaries and Considerations

• External Tool Dependencies: This skill requires installation and authorization of tools like tmeet, dws, lark-cli, tencent-docs, and wecom-cli. If not ready, it guides users through authorization (e.g., tmeet auth login) but does not skip automatically.
• Unhandled Tasks: Meeting management (e.g., scheduling, cancellation), document editing, and batch to-do operations are handled by other skills. For instance, DingTalk document editing uses dingtalk-unified, and Feishu task management uses lark-cli task.
• Degradation Mode: If a source is unavailable or has no recording, manual paste of minutes is supported as a fallback to prevent pipeline stalls. Missing content is marked as 【待补充】 in output.
• Deduplication Mechanism: Based on source_id recorded in processed.json, the same meeting minutes are not pushed twice. Intermediate checkpoints use checkpoint.json for resumable execution.
• Human-AI Co-Writing: After pushing to Tencent Docs, users can manually edit and have AI take over modifications (requires version ≥ V5.3.5), but changes do not flow back to the output snapshot.