agent-memory
由 dennis-da-menace 提供
AI & LLMs简介
Persistent memory system for AI agents.
描述
该技能的功能
Persistent memory layer for AI agents using a local SQLite database. Stores facts, experience-based lessons, and named entities across sessions. Provides structured recall by tag, context, or search query.
Unlike in-context memory that resets each session, this persists structured facts and lessons to disk and makes them searchable by tag or context.
何时使用
- Resuming an agent session with full context from previous runs
- Recording lessons after a failed automation to avoid repeating mistakes
- Tracking team members and their roles across multiple projects
- Storing API quirks or domain facts discovered during a task
- Loading relevant background before starting a new conversation
示例工作流
User asks: 'Remember that the staging API requires a special header, and log that we learned this the hard way'
- 1
Call mem.remember() with the API header fact and a tag like 'staging-api'
- 2
Call mem.learn() with action='used staging API without header', outcome='negative', insight='staging API requires X-Custom-Header'
- 3
On next session start, call mem.recall('staging API') to surface the stored fact
- 4
Call mem.get_lessons(context='staging-api') to retrieve the recorded lesson
Agent starts future sessions aware of the staging API requirement without being told again
安装
clawhub install dennis-da-menace/agent-memory运行 CLI 命令,或把这段提示词粘贴到 Claude Code、Cursor、Codex 或你在用的任意编程智能体里。