mindgardener
by widingmarcus-cyber
Coding Agents & IDEsBrief
Local-first long-term memory for autonomous agents.
Description
What This Skill Does
Local-first long-term memory for OpenClaw agents. Extracts entities from conversations into a wiki knowledge graph, scores events by surprise value, detects conflicts between new and old information, and assembles token-budget context. Everything is stored as markdown files with no database or external API required.
Unlike OpenClaw's built-in memory_search, which only queries existing memory, MindGardener creates and structures memory from conversations, adds surprise scoring, conflict detection, and graph traversal.
When to Use It
- Persisting project facts across multiple agent sessions
- Flagging when new information contradicts stored knowledge
- Building a navigable knowledge graph from conversation history
- Injecting relevant context automatically at session start
- Syncing per-agent memories to a shared store in multi-agent workflows
Example Workflow
User asks: remember everything from today's planning session
- 1
Run
garden extractto pull entities and facts from the conversation - 2
Run
garden surpriseto score unexpected or high-value facts - 3
Run
garden consolidateto merge new facts into the knowledge graph and detect conflicts - 4
Run
garden inject --output RECALL-CONTEXT.mdat the start of the next session to assemble relevant context
RECALL-CONTEXT.md populated with relevant facts within the token budget, ready to inject into the agent's context window
Installation
clawhub install widingmarcus-cyber/mindgardenerRun the CLI command, or paste the prompt into Claude Code, Cursor, Codex or whichever coding agent you use.