Dream

• Session was routine with no new learnings
• Recent /dream already ran (< 24 hours)

4-Phase Process

Phase 1 — Orient

List the memory directory to see what exists:

Glob pattern: memory/*.md
Read: memory/MEMORY.md

Understand current structure by reading MEMORY.md and scanning topic files for:

• What patterns are already captured
• What might be stale or contradicted
• Gaps in coverage

Phase 2 — Gather Signal

Priority order for finding worth-saving information:

1. Daily logs — Check memory/logs/YYYY/MM/YYYY-MM-DD.md if they exist
2. Recent session transcript — Look for corrections, learnings, new patterns
3. Existing memories that may have drifted — Re-read topic files you've visited recently
4. Grep for contradictions — Search for patterns like "but actually", "correction", "wait —"
5. Dreaming daemon insights — Read P:/.claude/state/dreaming-insights.json (or .md)
   • Check principle_stats for high-count violations (context_reuse, grounded_changes, etc.)
   • Check patterns for high_violation_principle entries
   • Treat high-count principles as lesson candidates for Phase 3

Key guidance: Don't exhaustively read transcripts. Look only for things you already suspect matter. Bias toward action over completeness.

Phase 3 — Consolidate

Write or update memory files at the top level using established conventions:

From memory_management.md:

• Use lowercase_with_underscores for topic file names
• One-line descriptions linking to topic files (index, not dump)
• Add new topic files to MEMORY.md table

Merge signal into existing topics rather than creating duplicates:

• Don't create new_topic.md if existing_topic.md covers it
• Extend existing files, don't fragment

Convert relative dates to absolute dates:

• "yesterday" → "2026-03-23"
• "last week" → "2026-03-17"
• "recently" → specific date when known

Delete contradicted facts:

• If MEMORY.md says X but you learned Y, remove X entirely
• Don't keep contradicting entries side-by-side

Daemon-derived insights:

• High-count principle violations from dreaming-insights.json are lesson candidates
• Example: context_reuse with count=10 → write/update context_reuse.md in memory/ with violation pattern, examples, and corrective action
• Use the graduate pass: if a daemon insight appears repeatedly, promote it to a hook rule or validator

Phase 4 — Prune and Index

Keep MEMORY.md under 200 lines:

• Count current lines before editing
• If adding content would exceed limit, extract to topic file first

Maintain index quality:

• One-line descriptions only — it's an index
• Remove stale pointers to deleted files
• Demote verbose entries to topic files
• Add new memories as topic files with table entries

Resolve contradictions:

• If two topic files disagree, pick the correct one
• Delete or update the wrong entry
• Note the correction in your summary

Output

Return a brief summary of what was consolidated:

## Dream Summary

**Consolidated:**
- [List of files written/updated]

**Pruned:**
- [List of files or entries removed]

**Learned:**
- [2-3 key learnings from this session's work]

Memory Locations

Integration

SessionStart hook (SessionStart_memory_monitor.py) warns when MEMORY.md > 180 lines. Consider running /dream when you see this warning.

Related skills:

• /reflect — Captures session learnings to CKS
• /learn — Novelty-weighted lesson capture
• /gto — Gap analysis for projects