reflect:consolidate

Reflect: Consolidate — Project Memory Cleanup

Merge orphaned worktree memory directories for the current project into a single, tidy .agents/MEMORY.md. This is a project-level operation — it does not touch the global knowledge base.

For global indexing (GraphRAG + QMD), use /reflect:ingest after consolidating.

When to Use

• After deleting git worktrees
• When multiple worktree sessions left orphaned memory dirs
• To keep .agents/MEMORY.md current and deduped
• Before running /reflect:ingest (consolidate first, then index)

What It Does

1. Discovers orphaned worktree memory dirs for the current project
2. Reads all MEMORY.md files from those dirs
3. Deduplicates entries (fuzzy — same concept = skip)
4. Routes skill-worthy content to existing skills
5. Writes consolidated .agents/MEMORY.md (200-line max)
6. Ensures agent config references it
7. Proposes cleanup of orphaned dirs

What It Does NOT Do

• Does NOT index into GraphRAG or QMD (that's reflect:ingest)
• Does NOT scan other tools (Codex, Copilot, Gemini) — project-scoped only
• Does NOT archive originals to ~/.learnings/ — that's ingest's job

Pipeline

Step 1: Discover Project Identity

python3 {{HOME_TOOL_DIR}}/skills/reflect/scripts/memory_discovery.py project-id
# -> e.g. "shotclubhouse"

Step 2: Find Orphaned Memory Dirs

python3 {{HOME_TOOL_DIR}}/skills/reflect/scripts/memory_discovery.py discover --provider claude
# Lists all worktree memory dirs for the current repo
# Excludes current session's memory dir

Step 3: Read All MEMORY.md Files

Read from each matched directory. Extract individual entries (sections, bullets, paragraphs).

Step 4: Deduplicate and Categorize

Group by section, remove redundant entries:

• Same concept = skip
• Same error + same solution = skip
• More detailed version = keep the detailed one

Step 5: Route Skill-Worthy Content

Check if entries match existing skill topics (check {{HOME_TOOL_DIR}}/skills/ and .claude/skills/). Propose adding matching content to those skills.

Step 6: Write .agents/MEMORY.md

Consolidated, deduped, within 200-line limit:

# Project Memory

> Auto-consolidated by /reflect:consolidate. 200-line max.
> Detailed learnings route to skills. Global index via /reflect:ingest.

## Architecture & Patterns

## Build & Deploy

## Gotchas & Workarounds

## Testing

## Environment & Config

Sections are dynamic — empty sections removed, new sections created as needed.

Step 7: Ensure Agent Config Reference

Check if .claude/CLAUDE.md or .claude/AGENTS.md contains @.agents/MEMORY.md. If not, add it at the top.

Step 8: Propose Orphan Cleanup

Show list of orphaned dirs and ask for confirmation:

# After user approval:
python3 {{HOME_TOOL_DIR}}/skills/reflect/scripts/memory_discovery.py cleanup /tmp/reflect-cleanup-dirs.txt --execute

Step 9: Report

## Consolidation Complete

- **Orphaned dirs processed**: 7
- **Entries consolidated**: 42
- **Duplicates removed**: 15
- **Routed to skills**: 3
- **.agents/MEMORY.md**: 127 lines
- **Dirs cleaned up**: 7

Next: Run `/reflect:ingest` to index into GraphRAG + QMD.

Stats (Quick Check)

Check orphaned memory status without making changes:

python3 {{HOME_TOOL_DIR}}/skills/reflect/scripts/memory_discovery.py stats
# -> Repo: shotclubhouse
# -> Orphaned memory dirs: 7
# -> Total lines across all: 283

Safety

• NEVER auto-apply without user approval
• NEVER delete orphaned dirs without explicit confirmation
• Always show what will be written to .agents/MEMORY.md before writing
• Preserve original structure of existing .agents/MEMORY.md