Group Chat Summarizer
Intelligent group chat summarization that transforms messy conversations into actionable insights.
Quick Start
Basic Usage
User: "Summarize this chat log"
[User pastes chat log]
→ Generate standard summary (Basic/Standard/Detailed based on length)
With Options
User: "Give me a detailed summary of yesterday's team chat"
→ Use detailed format with full timeline
User: "Extract only action items from this discussion"
→ Skip narrative, output only action items table
User: "What's the sentiment of this conversation?"
→ Include sentiment analysis section
Supported Platforms
The skill automatically detects and parses:
- China: Feishu (飞书), DingTalk (钉钉), WeChat Work (企业微信), QQ
- International: Discord, Slack, Microsoft Teams, Telegram
- Generic: Plain text, CSV exports, JSON logs
See references/platform_formats.md for format details.
Summary Levels
Basic (几句话)
- Best for: Quick catch-up, 50-100 messages
- Output: 3-5 bullet points of key decisions
Standard (几个要点)
- Best for: Daily standups, 100-300 messages
- Output: Topic threads + key decisions + action items
Detailed (完整脉络)
- Best for: Important meetings, 300+ messages
- Output: Full timeline + all sections + risk analysis
Output Structure
All summaries follow this structure (see references/output_template.md):
- Basic Info - Group name, time range, participants, message count
- Topic Timeline - Chronological thread of discussion topics
- Key Points - Decisions made, risks, important information
- Action Items - Tasks with owner, deadline, status
- Follow-up Suggestions - Recommended next steps
- Notes - Special mentions, absences, reminders
Special Features
Action Item Extraction
Automatically identifies:
- Task descriptions
- @mentioned owners
- Deadline phrases ("by Friday", "next week", "ASAP")
- Status indicators ("done", "pending", "blocked")
Sentiment Analysis
Detects conversation tone:
- 😊 Positive - collaborative, supportive
- 😐 Neutral - factual, informational
- 😟 Negative - conflicts, complaints, concerns
- ⚠️ Controversial - disagreements, unresolved debates
Risk & Controversy Detection
Identifies:
- Blocked items or impediments
- Resource constraints mentioned
- Disagreements without resolution
- Missing information or dependencies
Workflow
- Parse - Detect platform format and extract messages
- Analyze - Identify topics, participants, timeline
- Extract - Pull out decisions, action items, key info
- Generate - Create structured summary
- Enhance - Add sentiment, risks, follow-ups
Platform-Specific Notes
Feishu/DingTalk/WeChat Work
- Supports exported chat logs
- Handles Chinese date/time formats
- Recognizes @mentions and reply threads
Discord/Slack
- Supports JSON exports
- Handles threaded conversations
- Recognizes emoji reactions as sentiment signals
API Integration
When platform APIs are available:
- Use
scripts/fetch_messages.py to retrieve chat history - Requires appropriate authentication tokens
- Respects rate limits and privacy settings
Examples
Example 1: Work Group Daily Summary
Input: 127 messages from product-tech team
Output:
- 3 topics discussed (Q2 planning, UI review, technical concerns)
- 4 action items identified with owners
- 1 risk flagged (frontend timeline)
- Follow-up: Interface doc due Thursday
Example 2: Interest Group Discussion
Input: 89 messages about weekend hiking plan
Output:
- Topic: Hiking route selection → Decision: Xiangshan Trail
- 5 participants confirmed
- Action: @Alice to book bus by Wednesday
- Note: @Bob unavailable this weekend
Best Practices
- For long chats (>500 messages): Suggest breaking into time periods
- For sensitive content: Remind users about privacy when sharing logs
- For action items: Always verify @mentions are correctly assigned
- For follow-ups: Suggest specific dates based on context, not generic
Limitations
- Cannot access chats without user-provided logs or API tokens
- May miss context from edited/deleted messages
- Complex threaded discussions may need manual clarification
- Very long messages (>2000 chars) may be truncated in analysis
References