majiayu000 memory-optimization Skill
This skill prevents "Context Overflow" and reduces API costs.
1. Summarization Strategies
- Rolling Summary: Keep the last 10 messages + a summary of everything before.
- Entity Extraction: Extract key facts ("User chose Redis", "App is on port 3000") and store in
semantic/.
2. Context Pruning
- Irrelevant Files: Only keep file definitions in context if actively editing. Use
grepfor others. - Old Logs: Truncate logs > 50 lines.
3. The "Forget" Protocol
- When a task is done:
- Summarize: Write outcome to
AUDIT_LOG.md. - Clear: Remove task-specific context from working memory.
- Retain: Keep only global constraints and active file list.
- Summarize: Write outcome to
4. Semantic Memory Integration
- Use Qdrant or PGVector (via MCP) to store embeddings of documentation.
- Retrieval: Retrieve only the Top-3 relevant chunks for the current query.