Memory

Savemaintenance

Try it

Deterministic reconciliation of conversation-log.md, FTS5 index, and saved conversations — one-shot fix or maintenance run Sister skill to the save skill, only get this if you are planning on getting and using the save skill by the same author

What it does

Deterministic reconciliation of conversation-log.md, FTS5 index, and saved conversations — one-shot fix or maintenance run

The skill document

savemaintenance — Memory System Reconcile Pipeline

One-shot tool for keeping the three-tier memory system coherent. Cross-references the conversation log against files on disk, rebuilds the FTS5 index, runs a full audit. Zero external dependencies.

The Problem

The three-tier memory system (log → FTS5 index → files) is designed for consistency, but things drift:

  • Files get deleted but their log entries remain (orphans)
  • Files get added outside the save workflow but never get indexed
  • The FTS5 index gets stale
  • The topic map and stub index need regeneration

savemaintenance is the repair pipeline for all of the above. Run it weekly, or after any bulk operation on saved conversations.

Usage

# Copy the pipeline to the workspace
cp -r {baseDir}/. ~/.openclaw/workspace/savemaintenance/

# Dry run first — see what would change
cd ~/.openclaw/workspace/savemaintenance && python3 full-reconcile.py --dry

# Full repair: backup → reconcile → rebuild → audit
./run.sh

# Or run individual steps
python3 full-reconcile.py --step backup
python3 full-reconcile.py --step reconcile
python3 full-reconcile.py --step rebuild
python3 full-reconcile.py --step audit

What It Does

1. Backup

Copies conversation-log.md and topic-index.json to timestamped backups.

2. Reconcile

  • Removes log entries where the .md file no longer exists
  • Adds entries for .md files not in the log (reads # Title from file content)
  • Sorts all entries by date descending
  • Updates the header count

3. Rebuild

Regenerates the FTS5 index at /dev/shm/memory-index.db from scratch.

4. Audit

Reports: orphan entries, missing files, index coherence, directory hygiene.

Scripts

FilePurpose
run.shEntry point — snapshot then full reconcile
full-reconcile.pyCore pipeline (backup → reconcile → rebuild → audit)
snapshot.pyManual snapshot/restore tool
README.mdFull documentation

Snapshots

python3 snapshot.py                # Auto-timestamped snapshot
python3 snapshot.py take      # Named snapshot
python3 snapshot.py list           # List all snapshots

Snapshots go to backups/ by default.

Edge Cases Handled

  • Empty log file — creates a fresh log from scratch
  • New files added between runs — adds them without removing existing entries
  • Deleted files still in log — entries are removed
  • Renamed files — detected as "dead entry + new orphan" pair
  • Dry run — preview changes without writing anything

Default Paths

PathDefault
Saved conversations~/.openclaw/workspace/saved/
Conversation log~/.openclaw/workspace/saved/conversation-log.md
Memory index/dev/shm/memory-index.db
Backup dir~/.openclaw/workspace/savemaintenance/backups/

Set OPENCLAW_WORKSPACE env var to override ~/.openclaw/workspace.

Dependencies

  • Python 3.8+ (standard library only — no pip packages)
  • SQLite3 FTS5 (bundled with Python)
  • The save skill from ClawHub (for memory-index.py)

Related skills

Save

Save conversations to memory index — FTS5 rebuild, log cross-ref, saved convos dispatch Long term memory for your ai agent

1 installs

persistent-skill-memory

Persist a deterministic skill index into the agent system prompt. One stdlib-only CLI (index/prompt/inject/verify/stats/hook), idempotent marker-block injection, no truncation, offline.

2 installs

xmemory

Persistent structured memory — save, recall, and update facts, decisions, people, and projects across sessions via the xmemory MCP server.

memory-system

A practical memory protocol for long-running AI agents: dual logs, encoding safety, backup sync, multi-agent facts

1 installs

Mem

Mem (mem.ai). Use this skill for ANY Mem request — reading, creating, updating, and deleting data. Whenever a task involves Mem, use this skill instead of calling the API directly.

2 installs