memory-management

Promoting auto-memory into enforced rules involves analyzing MEMORY.md entries to find proven debugging patterns and moving them into CLAUDE.md or .claude/rules/ with scoped YAML frontmatter paths for reliable, context-aware loading. This prevents valuable learnings from being lost to truncation or left as outdated notes.

The best way to manage CLAUDE.md line budget is to measure memory health against capacity limits, consolidate stale content, and extract recurring solutions into standalone portable skills. This keeps the file within budget while preserving proven patterns as reusable, portable assets.

Yes, you can scope rules to load only when relevant files are open by applying structured rule scoping via YAML frontmatter paths. This ensures enforced instructions activate contextually, preventing unnecessary token consumption and keeping the agent focused on the active workflow.

Extracting recurring debugging solutions into reusable skills requires analyzing MEMORY.md to identify recurring patterns and packaging them as standalone portable skills. This avoids re-solving the same problem and frees up memory capacity by removing the need to keep the full solution in auto-memory.

Your agent loses debugging solutions from auto-memory because valuable learnings get buried or lost to truncation when MEMORY.md grows too large. Without a deliberate lifecycle for reviewing and promoting agent knowledge across memory layers, recurring solutions remain unstructured and are not enforced as durable rules.

You should persist a fix to auto-memory immediately when you know a debugging pattern will recur, using command-driven memory curation. This explicitly captures the solution before it is lost, making it available for later promotion to enforced rules or extraction as a portable skill during a memory review.