dream-consolidation

Memory consolidation cycle — Orient, Gather, Consolidate, Prune. Run periodically to keep memory files clean, deduplicated, and under size limits.

SKILL.md

GitHub

Dream Consolidation

A 4-phase memory maintenance cycle that mirrors biological memory consolidation.

When to Trigger

  • After major project milestones
  • When MEMORY.md exceeds 200 lines
  • Every 2 weeks of active use
  • When the user says "consolidate memory" or "dream"

Phase 1: Orient

Read ALL memory files to understand current state:

  • Global memory (~/.claude/memory/*)
  • Project memory (~/.claude/projects//memory/)
  • MEMORY.md index file

Assess: line counts, staleness, duplicates, outdated entries.

Phase 2: Gather

Identify what to keep, compress, and discard:

  • Keep: Active project details, current priorities, hard rules
  • Compress: Session summaries → archive file with key takeaways
  • Discard: Completed tasks, stale priorities, resolved issues
  • Deduplicate: Info in MEMORY.md that's already in topic files

Phase 3: Consolidate

Execute the changes:

  • Archive old session summaries to archive file
  • Update MEMORY.md with compressed references
  • Backfill learning-log.md with missing entries
  • Clean up session-handoff.md
  • Extract cross-cutting technical lessons into MEMORY.md

Phase 4: Prune

Verify the results:

  • MEMORY.md under 200 lines
  • No duplicate information across files
  • All topic files still accurate and accessible
  • Archive file preserves important details
  • Report: before/after line counts, what changed

Key Principles

  • Deduplication through indirection: MEMORY.md points to topic files, doesn't repeat them
  • Cross-cutting consolidation: Scattered lessons get one reusable section
  • Nothing important lost: Archive preserves details, MEMORY.md keeps pointers