memory-sync_skill

Overview

This skill scrapes OpenClaw JSONL session logs and reconstructs agent memory files to restore continuity across model switches. It supports fast keyword-based extraction or higher-quality LLM narrative summaries, with automatic secret sanitization at every step. Use it to verify coverage, backfill missing days, or run nightly automated syncs.

How this skill works

The tool reads session logs from a configured sessions directory, compares them to existing memory files, and identifies gaps or changed days. Backfill can run in two modes: simple extraction that derives topics, key exchanges, and decisions via heuristics; or LLM summarization that generates coherent 2–4 paragraph narratives and can incorporate existing hand-written notes. All extracted text is sanitized for secrets before any LLM processing, file writes, or CLI output.

When to use it

• Memory appears incomplete after switching models or backends
• Verifying daily memory coverage and detecting gaps
• Reconstructing lost or accidentally deleted memory files
• Nightly or scheduled automated memory sync via cron/heartbeat
• Running targeted catch-up for a date range after outages

Best practices

• Use simple extraction for fast initial backfills and resource-constrained systems
• Use LLM summarization (--summarize) plus --preserve for highest-quality narratives that retain your notes
• Run an initial --all backfill once, then switch to --incremental nightly runs for efficiency
• Prefer the default openclaw backend to avoid separate API keys and leverage existing configuration
• Enable --dry-run first to preview changes before writing files

Example use cases

• Nightly cron job: backfill --today --summarize --preserve to automatically update daily memory
• Recovery after downtime: backfill --since 2026-01-28 --summarize to catch up several days
• Quick check: compare to list missing dates before running a full backfill
• Regenerate a day with preserved notes: backfill --date 2026-02-05 --force --preserve --summarize
• Low-resource environment: backfill --today (simple extraction) to avoid API usage