clawvault_skill

Overview

This skill implements a local, structured memory system for AI agents with a memory graph, context profiles, checkpoint/recover tools, and Obsidian integration. It keeps all data on-disk (no cloud sync) and provides semantic search, observational compression, task tracking, canvas dashboards, and Tailscale-based cross-vault networking. Use it to prevent context loss, repair broken sessions, and surface graph-aware context for agent prompts.

How this skill works

clawvault stores memories as markdown files in a vault directory and maintains a ledger and graph index to build relationships from wiki-links, tags, and frontmatter. It instruments sessions with checkpointing, session observation (LLM compression), and repair tools that update transcripts safely with backups. The tool offers CLI commands for remembering items, semantic/keyword search, task management, canvas generation, and serving the vault over Tailscale for cross-vault queries.

When to use it

• When an agent needs durable, structured memory stored locally with graph relationships
• To avoid context death by checkpointing and recovering OpenClaw sessions
• When you want graph-aware context retrieval using profiles (planning, incident, handoff)
• To compress and summarize session transcripts or generate weekly reflections
• To manage tasks, generate canvas dashboards, or integrate with Obsidian views
• When you need cross-vault search and sharing over a Tailscale network

Best practices

• Initialize a dedicated vault path and set CLAWVAULT_PATH to avoid accidental writes
• Use checkpoint frequently during long sessions and enable the OpenClaw hook for auto-checkpointing
• Run graph refresh after bulk edits to keep context retrieval accurate
• Provide GEMINI_API_KEY if you want observe/reflect LLM compression to run
• Keep backups before running repair-session; use --dry-run first to inspect changes

Example use cases

• An agent captures decision notes and later retrieves related project context via graph-aware context profiles
• Recovering a broken OpenClaw session where tool_result blocks were orphaned or calls aborted
• Generating an Obsidian canvas dashboard for sprint planning filtered by owner and include-done flag
• Compressing long session transcripts into scored observations and weekly reflections
• Serving vault contents over Tailscale to allow secure cross-vault semantic search with peers