本文へスキップ
TakoAPI
スキル一覧に戻る

persistent-memory

jakebot-ops 作

Coding Agents & IDEs
669ダウンロード数はclawskills.shが公開している値です
0閲覧数
0件のいいねサインインするとこのスキルにいいねできます
—まだ評価がありません
あなたの評価

概要

Three-layer persistent memory system (Markdown + ChromaDB vectors + NetworkX knowledge graph) for long-term agent.

説明

このスキルでできること

Three-layer memory system using Markdown files, ChromaDB vector embeddings, and a NetworkX knowledge graph that persists agent knowledge across sessions. One command installs all dependencies and configures OpenClaw to index directive files like SOUL.md and AGENTS.md that its built-in memory search ignores by default.

Combining vector search with a knowledge graph lets the agent surface semantically related facts and traverse concept relationships, while the OpenClaw integration closes a gap where agents silently ignore their own directive files.

使いどころ

  • Recalling an architectural decision made weeks ago without fabricating details
  • Looking up an exact repo URL or contact handle before taking an action
  • Reviewing what the agent logged during a previous work session
  • Keeping agent behavior consistent with workspace rules after restarts
  • Building up an institutional knowledge base of people, repos, and infrastructure over time

ワークフロー例

入力

User asks: 'What did we decide about the authentication approach last month?'

エージェント
  1. 1

    Runs memory_search or auto_retrieve.py with the query before responding

  2. 2

    Vector search returns top-3 semantically similar chunks from MEMORY.md and daily logs

  3. 3

    Graph traversal surfaces related concepts linked to the matching chunks

  4. 4

    Agent cites the source file and section in its answer

出力

Agent returns the recorded decision with the date it was logged and the file it came from, without fabricating details

インストール

clawhub install jakebot-ops/persistent-memory

CLIコマンドを実行するか、Claude Code、Cursor、Codexなどお使いのコーディングエージェントにプロンプトを貼り付けてください。

リンク