Bumblebiber/hmem-migrate-o

/hmem-migrate-o — O-Entry Migration to Project-Bound Structure

Prerequisite: This skill uses move_nodes and rename_id from the hmem-curate MCP server. Tell the user to activate it via /mcp before proceeding.

This is a one-time migration from the old flat O-entry format to the new 5-level hierarchy where each O-entry is permanently bound to a P-entry (O0048 belongs to P0048).

What Changes

Before (old format):

• New O-entry created per session, loosely linked via links field
• Flat structure: L2 (exchange) -> L4 (user msg) -> L5 (agent msg)
• active flag on O-entries to track current session

After (new format):

• One O-entry per project, ID matches P-entry (O0048 <-> P0048)
• 5-level hierarchy: L2 (session) -> L3 (batch) -> L4 (exchange) -> L5 (user/agent)
• No active flag needed — O is derived from the active P
• P0000/O0000 exist as "Non-Project" catch-all

---

Step 1: Pre-Migration Check

Understand what you're working with before changing anything.

hmem --version   # must be >= 5.3.0

Then use read_memory() to see current O-entries. Check:

• How many O-entries exist?
• Which P-entries are they linked to?
• Are any O-entries already using the new format (have L3 depth=3 nodes)?

If the version is too old, run /hmem-update first.

Step 2: Backup

Create a backup of the hmem database before migrating. This is critical — the migration rewrites entry IDs.

# Find the active hmem file
HMEM_FILE=$(find ~/.hmem -name "*.hmem" -not -path "*/Agents/*" | head -1)
echo "Backing up: $HMEM_FILE"
cp "$HMEM_FILE" "${HMEM_FILE}.pre-migration-backup"

# Also backup agent-specific hmem files
find ~/.hmem/Agents -name "*.hmem" -exec sh -c 'cp "$1" "$1.pre-migration-backup"' _ {} \;

Verify backups exist before continuing.

Step 3: Run Migration

The migration script handles everything:

hmem migrate-o-entries

This will:

1. Create P0000 "Non-Project" and O0000 catch-all if they don't exist
2. For each existing O-entry:
   • Find its linked P-entry via the links JSON field
   • Rename it to match the P-entry's sequence number (e.g., O0042 linked to P0048 becomes O0048)
   • Handle ID conflicts by moving blockers to temporary IDs (O9XXX range)
   • Tag all migrated entries with #legacy
3. Tag unlinked O-entries as #legacy (they'll go to O0000 eventually)
4. Clear all active flags from O-entries

Read the output carefully. The script reports every rename and conflict. Example output:

=== hmem O-Entry Migration ===

Found 105 O-entries to process.

  O0001 -> O0048 (P0048 hmem-mcp)
  O0020 -> O0051 (P0051 BookCast)
  O0168 -> O0043 (P0043 EasySAP)
  O0002 -> #legacy (conflict for O0048)
  O0173 -> #legacy (no P-link)

Migration plan: 3 renames, 102 legacy tags.

Step 4: Verify

After migration, verify the structure is correct:

read_memory()                    # Should show O-entries with correct IDs
read_memory(id="O0048")          # Should exist if P0048 exists
load_project(id="P0048")         # Should load without errors

Also check:

• The statusline still shows project + exchange counter
• New exchanges land in the correct O-entry (send a test message)

echo '{}' | hmem statusline     # Should show project info

Multi-Device Sync (after migrating on the server)

If the migration was run on the sync server (e.g., Strato), other devices don't need to run the migration themselves. Instead, delete the local DB and pull fresh from the server:

# 1. Delete local hmem file
AGENT_DIR=~/.hmem/Agents/DEVELOPER
rm "$AGENT_DIR/DEVELOPER.hmem"

# 2. Pull from server — MUST specify --hmem-path explicitly!
# Without it, hmem-sync writes to ~/.hmem/memory.hmem (wrong file).
# hmem-mcp reads from Agents/DEVELOPER/DEVELOPER.hmem — they must match.
hmem-sync pull --hmem-path "$AGENT_DIR/DEVELOPER.hmem" --force

IMPORTANT: Always use --hmem-path when pulling after a DB delete. Without it, hmem-sync auto-detects ~/.hmem/memory.hmem as the target, but hmem-mcp reads from Agents/DEVELOPER/DEVELOPER.hmem. This mismatch causes "No memories found" after pull.

Do this on every device/agent that syncs with the server. After pulling, verify with read_memory() that the O-entries have the correct IDs.

---

Step 5: Post-Migration Cleanup

The migration tags old O-entries as #legacy. These still use the old flat format but are readable via dual-format support. Options:

1. Keep for now — they don't hurt anything, just take space
2. Mark irrelevant — update_memory(id="O0042", irrelevant=true) for entries you don't need
3. Delete later — run /hmem-curate to review and clean up #legacy entries
4. Auto-purge — irrelevant entries older than 30 days are automatically deleted

Recommendation: keep them for a week to make sure everything works, then curate.

Step 6: Remove Backup (after verification)

Once you're confident the migration worked:

find ~/.hmem -name "*.pre-migration-backup" -exec rm {} \;

Troubleshooting

Migration failed mid-way: Restore from backup and retry.

cp "${HMEM_FILE}.pre-migration-backup" "$HMEM_FILE"

O-entry has wrong project: Use move_nodes to relocate sessions/exchanges:

move_nodes(node_ids=["O0048.3"], target_o_id="O0051")

Statusline shows wrong count: The statusline cache refreshes every 30s. Wait or:

rm /tmp/.hmem_statusline_cache

New exchanges still use old format: Make sure cli-log-exchange.ts is the updated version (v5.3+). Run hmem --version to confirm.

Notes

• This migration is idempotent — running it twice won't break anything (already-correct entries are skipped)
• The migration only reassigns IDs and tags; it does NOT restructure internal node hierarchy
• Old #legacy entries are readable via dual-format support in getOEntryExchanges()
• New exchanges automatically use the 5-level format via the rewritten Stop hook