Tencent Meeting Minutes Assistant

Problem

Teams often have meeting notes but not a complete workflow from preparation to follow-through: agendas are assembled late, prior decisions are scattered, in-meeting decisions lack anchors, action items remain unowned, sensitive content is hard to handle, and cross-meeting reviews rely on manual digging. This skill connects Tencent Meeting data to an AI workflow for teams that need stable minutes, task boards, and follow-up materials.

How it works

• Connection layer: It relies on the Tencent Meeting MCP and mcporter call chain. Before reading real meeting data, it performs a connection check to avoid fabricating results without authorization.
• Workflow layer: It organizes work across pre-meeting, in-meeting, post-meeting, insight, and scenario stages. Pre-meeting outputs include agendas, material packs, and attendee briefs; in-meeting outputs include real-time minutes, decision markers, quote excerpts, and clarification lists; post-meeting outputs include minutes, action items, and task boards; later stages produce weekly reports, retrospective materials, and cross-meeting insights.
• Delivery layer: Output follows templates and passes qa checks for placeholders, decision traceability, and sensitive data. It can coordinate redaction, export, sync, and follow-up emails. The auto orchestrator can chain multiple steps with resume and retry support.
• Structure constraints: Before generating minutes or agendas, it classifies the meeting type, such as review/decision, recurring sync, customer negotiation, or project retrospective, to avoid forcing one format onto every meeting.

Boundaries

• Real Tencent Meeting data requires completed authorization and service connection. An offline template path is available when disconnected, but it cannot verify real data.
• Smart summaries, participation signals, and relevance judgments depend on transcripts and context, and can be affected by anonymous speakers, missing recordings, or time-window limits.
• Official multilingual support is limited; out-of-scope languages require a translation layer and are not guaranteed with official accuracy.
• Scheduled tasks are not “meeting-ended event triggers” but closer to periodic polling; external push depends on the corresponding connectors.