Setup & Installation
Or with the ClawHub CLI, for registry-managed skill folders outside a full OpenClaw workspace:
Version History
v2.0.3: Built-in ONNX embedding, 6 MCP tools, dual database (corpus + cognition)
What This Skill Does
Persistent memory system for coding agents that combines BM25 full-text and vector semantic search over indexed code and documents. Stores cognitive facts like decisions and lessons learned across sessions via 6 MCP tools. Works out of the box with built-in ONNX embeddings, no external embedding service required.
Hybrid BM25 + vector search with RRF fusion finds relevant code for exploratory questions where grep requires knowing the exact identifier, and can cross language boundaries that grep cannot.
When to use it
- Finding how a feature works by concept rather than exact symbol name
- Searching an English codebase using a Chinese or Japanese query
- Recording architectural decisions so future agent sessions retain context
- Reducing context window consumption when a keyword appears hundreds of times
- Rebuilding the search index after adding new source files or docs
Example Workflow
Here's how your AI assistant might use this skill in practice.
User asks: how does the authentication flow work in this project?
- 1Agent calls recall with query 'authentication flow'
- 2index1 runs BM25 + vector hybrid search across indexed code and cognitive facts
- 3Returns ranked summaries of relevant files and any recorded decisions about auth
- 4Agent uses Grep for specific identifiers found in the summaries
- 5Agent composes answer from summaries and targeted grep results
Explanation of the auth flow with file locations, using ~460 tokens instead of thousands from raw grep output