Feature
Instead of rule-based compression (lines > 150), use Claude's understanding of the project to decide what to compress.
Current approach: Compress based on age and line count thresholds.
Proposed approach: Claude analyzes each memory entry and asks:
- Is this still relevant to current work?
- Would losing this cause a re-decision later?
- Is this derivable from code/git?
- How likely is this to be needed in the next 3 sessions?
Why this matters:
Rule-based compression sometimes archives things that are still important, and keeps things that are no longer relevant. Context-aware compression would be much more precise.
Feature
Instead of rule-based compression (lines > 150), use Claude's understanding of the project to decide what to compress.
Current approach: Compress based on age and line count thresholds.
Proposed approach: Claude analyzes each memory entry and asks:
Why this matters:
Rule-based compression sometimes archives things that are still important, and keeps things that are no longer relevant. Context-aware compression would be much more precise.