Conversation Compressor

This skill compresses conversation history while preserving essential context for continued work.

When to Invoke

• •Automatic: Conversation exceeds 20 message exchanges
• •Manual: User explicitly says:
  • •"压缩对话" / "压缩上下文"
  • •"总结上文" / "summarize conversation"
  • •"squash context" / "compress history"
  • •"对话太长了" / "context too long"
• •Emergency: Context window approaching limit (warning signs: slower responses, truncated outputs)

Compression Process

Step 1: Analyze Conversation Structure

Identify and categorize:

• •User's Original Intent: What did the user initially want to achieve?
• •AI Actions: What steps were taken? (files read, modified, commands executed)
• •Key Decisions: Important choices made during the conversation
• •Current State: Where are we now? (completed/pending/blocked)
• •Technical Context: File paths, code snippets, error messages, configurations

Step 2: Extract Critical Information

Must Preserve:

• •File paths and line numbers referenced
• •Code changes (functions, classes, key logic)
• •Command outputs and error messages (if unresolved)
• •Configuration changes
• •Test results and verification status
• •Todo list items and their status

Can Summarize:

• •Explanatory text and educational content
• •Repeated attempts or iterations
• •Successful verifications (keep result only)
• •General discussion

Step 3: Generate Structured Summary

Use this exact format:

## Conversation Summary

**Original Request**: [One-line description of user's goal]

**Current Status**: [completed / in-progress / blocked / needs-verification]

**Actions Completed**:
1. [Action]: [Brief result] → [File/Location affected]
2. [Action]: [Brief result] → [File/Location affected]
...

**Key Technical Context**:
- **Modified Files**: [list with brief description of changes]
- **Running Processes**: [if any]
- **Environment**: [relevant config, ports, versions]
- **Unresolved Issues**: [errors or blockers needing attention]

**Current Todo Status**:
- [x] Completed: [task]
- [ ] Pending: [task]
- [ ] Blocked: [task] (reason: ...)

**Next Steps**: [What should happen next]

**Important Notes**:
- [Any critical info not covered above]

Step 4: User Confirmation

Before applying compression:

1. •Present the generated summary
2. •Ask: "是否确认压缩对话？这将保留上述关键信息，删除详细历史。"
3. •Wait for user confirmation (Y/n)
4. •If confirmed, replace conversation context with summary

Guidelines

Do's

• •✅ Keep all file paths exact and complete
• •✅ Preserve error messages if issue not resolved
• •✅ Maintain todo list with accurate status
• •✅ Include verification results (pass/fail)
• •✅ Note any external dependencies or requirements

Don'ts

• •❌ Remove unresolved errors or blockers
• •❌ Lose track of modified files
• •❌ Delete pending tasks without noting them
• •❌ Compress if user explicitly wants full history
• •❌ Auto-compress without asking (unless emergency)

Example

Before Compression (25 messages):

• •User: Help me fix this bug
• •AI: Let me check the file
• •[10 messages of investigation]
• •AI: Found the issue on line 45
• •[5 messages of explaining the fix]
• •AI: Fixed, please test
• •User: Works great!
• •[5 messages of discussion]

After Compression:

## Conversation Summary

**Original Request**: Fix bug in authentication module

**Current Status**: completed

**Actions Completed**:
1. Investigated auth.py → Found null pointer on line 45
2. Applied fix: Added null check → auth.py#L45-48
3. Verified fix works → User confirmed

**Key Technical Context**:
- **Modified Files**: auth.py (added null check at line 45)
- **Environment**: Python 3.9, FastAPI
- **Unresolved Issues**: None

**Current Todo Status**:
- [x] Completed: Fix auth bug

**Next Steps**: None - task complete

Emergency Compression

If context window is critically low:

1. •Skip user confirmation
2. •Auto-generate summary
3. •Notify user: "[系统自动压缩对话以节省上下文空间]"
4. •Include compressed summary in next response