Aio Dream

Dream: Memory Consolidation

You are performing a dream — a reflective pass over your memory files. Synthesize what you've learned recently into durable, well-organized memories so that future sessions can orient quickly.

Environment

• Memory directory: !echo "${CLAUDE_MEMORY_DIR:-$HOME/.claude/projects/$(pwd | sed 's|/|-|g')/memory}"
• Session transcripts: !echo "${CLAUDE_PROJECT_DIR:-$HOME/.claude/projects/$(pwd | sed 's|/|-|g')}" (JSONL files — each file is one past session)

Phase 1 — Orient

1. ls the memory directory to see what already exists
2. Read MEMORY.md to understand the current index
3. Skim existing topic files so you improve them rather than creating duplicates
4. If logs/ or sessions/ subdirectories exist (assistant-mode layout), review recent entries there

Phase 2 — Gather recent signal

Look for new information worth persisting. Sources in rough priority order:

1. Daily logs (logs/YYYY/MM/YYYY-MM-DD.md) if present — these are the append-only stream
2. Existing memories that drifted — facts that contradict something you see in the codebase now
3. Session transcripts — the transcript directory contains JSONL files of past sessions. Grep them for relevant context:

   grep -rn "<narrow term>" <transcript-dir>/ --include="*.jsonl" | tail -50
   

   Use narrow, specific terms relevant to what you found in Phase 1 (e.g., a project name, a tool, a decision topic). Read surrounding lines when a match looks valuable.

Phase 3 — Consolidate

For each thing worth remembering, write or update a memory file at the top level of the memory directory.

Memory file format

---
name: {{memory name}}
description: {{one-line description — used to decide relevance in future conversations, so be specific}}
type: {{user | feedback | project | reference}}
---

{{memory content}}

Memory types

What NOT to save

• Code patterns, architecture, file paths, project structure (derivable from code)
• Git history, recent changes (use git log)
• Debugging solutions (the fix is in the code)
• Anything already in CLAUDE.md
• Ephemeral task details

Consolidation rules

• Merge new signal into existing topic files rather than creating near-duplicates
• Convert relative dates ("yesterday", "last week") to absolute dates so they remain interpretable
• Delete contradicted facts — if today's investigation disproves an old memory, fix it at the source
• For feedback and project types, structure as: rule/fact, then **Why:** and **How to apply:** lines

Phase 4 — Prune and index

Update MEMORY.md so it stays under 200 lines AND under ~25KB.

MEMORY.md is an index, not a dump — each entry should be one line under ~150 characters:

- [Title](file.md) — one-line hook

Never write memory content directly into MEMORY.md.

Pruning checklist

• Remove pointers to memories that are now stale, wrong, or superseded
• Demote verbose entries: if an index line is over ~200 chars, move detail to the topic file
• Add pointers to newly important memories
• Resolve contradictions — if two files disagree, fix the wrong one
• Remove empty or near-empty memory files

Output

Return a brief summary of what you consolidated, updated, or pruned. If nothing changed (memories are already tight), say so.

Format:

## Dream Summary

**Reviewed:** N memory files
**Updated:** file1.md, file2.md
**Created:** new-topic.md
**Pruned:** old-stale.md
**MEMORY.md:** added 2 entries, removed 1, now at X/200 lines

### Changes
- Merged feedback about X into existing feedback_testing.md
- Created project_migration.md for the ongoing DB migration context
- Removed user_old_role.md (outdated, user changed teams)
- Converted relative dates in 3 files to absolute dates