Ingest OpenClaw agent history into the Obsidian wiki. Use this skill when the user wants to mine their past OpenClaw sessions for knowledge, import their ~/.openclaw folder, extract insights from previous OpenClaw conversations, or says things like "process my OpenClaw history", "add my OpenClaw sessions to the wiki", "ingest ~/.openclaw", or "what have I worked on in OpenClaw". Also triggers when the user mentions OpenClaw session logs, MEMORY.md, daily notes, or ~/.openclaw/workspace.
Detected risks:
A workflow for mining a user's OpenClaw agent history and distilling durable knowledge into an Obsidian wiki. It triggers when someone wants to process their OpenClaw sessions, import their ~/.openclaw folder, or extract insights from past OpenClaw conversations, and it can run directly or via the wiki-history-ingest router. Before starting, it resolves configuration (following the Config Resolution Protocol in llm-wiki to obtain OBSIDIAN_VAULT_PATH and OPENCLAW_HISTORY_PATH), then reads the vault's .manifest.json to see what has already been ingested and index.md to understand existing content. Two modes are supported: append mode (the default), which processes only files absent from the manifest or modified since their ingested_at timestamp, and full mode, which reprocesses everything after a rebuild or on explicit request. The document maps the OpenClaw data layout under ~/.openclaw — global config, a workspace holding MEMORY.md, optional DREAMS.md, and dated daily notes, and per-agent session directories with sessions.json indexes and per-session JSONL transcripts — ranking sources by value (MEMORY.md highest, then daily notes, then transcripts) and instructing which paths to skip, such as credentials and runtime model config. The process surveys and computes a delta, parses MEMORY.md and recent daily notes for concept-level knowledge, and parses session JSONL safely, prioritizing assistant conclusions and user intent while applying a critical privacy filter that removes API keys, tokens, passwords, and private identifiers and never quotes transcripts verbatim. Extracted knowledge is clustered by stable topic rather than by session, then distilled into wiki pages routed to projects, concepts, skills, entities, or synthesis folders, each carrying summary and confidence/lifecycle frontmatter and provenance markers (extracted, inferred, ambiguous). Finally it updates the manifest, log.md, index.md, and hot.md to record what was ingested. The emphasis throughout is on capturing durable knowledge, not operational telemetry or chronology.
The user's OpenClaw agent history under ~/.openclaw — primarily MEMORY.md, dated daily notes, and per-session JSONL transcripts — distilling durable knowledge into an Obsidian wiki while skipping credentials and runtime config.
Append mode (default) processes only files new to the manifest or modified since their ingested_at timestamp; full mode reprocesses everything, for use after a wiki-rebuild or when explicitly requested.
MEMORY.md is highest signal (durable long-term facts), followed by recent daily notes, then session transcripts, with the sessions.json index used for inventory and labels.
Session transcripts are never ingested verbatim — API keys, tokens, passwords, credentials, and private identifiers are removed or redacted, and content is summarized rather than quoted.
It is clustered by stable topic rather than per session, routed to folders like projects, concepts, skills, entities, or synthesis, and each page gets summary, confidence/lifecycle frontmatter, and provenance markers.
Quick Setup:
.claude/skills/Repository
ar9av/obsidian-wiki