memory-optimization Skill

Goal

This skill prevents "Context Overflow" and reduces API costs.

Steps

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 grep for others.
• Old Logs: Truncate logs > 50 lines.

3. The "Forget" Protocol

• When a task is done:
  1. Summarize: Write outcome to AUDIT_LOG.md.
  2. Clear: Remove task-specific context from working memory.
  3. Retain: Keep only global constraints and active file list.