openclaw-history-ingest reads the files OpenClaw keeps under ~/.openclaw: MEMORY.md, daily notes, an optional dream diary and the per-session JSONL transcripts. It sorts them by value, works out which ones are new or changed since the last run, and distills the durable parts into wiki pages.
Pages are grouped by topic, carry a summary, confidence and provenance markers, and are routed to projects, concepts, entities, skills or synthesis folders. Secrets and private identifiers are removed before anything is written.
What You Get
- Wiki pages grouped by stable topic, each with a short summary, a confidence value and a lifecycle field
- Provenance markers that separate extracted, inferred and ambiguous claims
- An updated manifest that records every ingested file, its source type and the pages it produced
- A delta report before parsing starts, so you see what is new or changed
- Append mode for regular syncs and full mode for a complete re-ingest
Common Use Cases
- Importing a long-running OpenClaw memory file into a personal knowledge base
- Running a regular sync that picks up only new daily notes and sessions
- Rebuilding the wiki from scratch after a vault reset using full mode
What Makes It Unique
It knows the OpenClaw data layout and ranks the sources by value: the structured MEMORY.md first, daily notes second, transcripts third. It skips the credentials folder and agent runtime config entirely and applies a redaction filter to transcripts. When a newer session contradicts an existing page, it records the old claim under a history entry or an expiry field and keeps the page.
Under The Hood
The skill is a set of instructions for the agent. It reads files under the OpenClaw history folder and the vault, and writes new and updated Markdown pages and the manifest in the vault. It runs the obsidian-wiki command line tool through the shell to update the index, log and hot cache under a lock. If a QMD search collection is configured it also runs the qmd command to refresh the search index. It does not fetch web pages or call remote APIs.