← 返回
未分类

Smart Memory Keeper

解决 AI /new 后失忆问题的记忆管理 skill。三层加载机制(热/温/冷),session 启动时只取当前需要的记忆,省 token。 包含:任务状态恢复、每日日记、项目索引、Dream 定期整理。纯文件系统,无需外部服务。 用户说"先这样"、"暂停"、"记住这个"、里程碑完成时触发。 Memory ma...
解决 AI /new 后失忆问题的记忆管理技能。三层加载机制(热/温/冷),session 启动时仅加载当前所需记忆,节省 token。包含:任务状态恢复、每日日记、项目索引、定期整理 Dream。纯文件系统,无外部依赖。触发条件:用户说「先这样」「暂停」「记住这个」或里程碑完成时。Memory ma...
porkapple porkapple 来源
未分类 clawhub v1.5.2 1 版本 99798.8 Key: 无需
★ 1
Stars
📥 476
下载
💾 0
安装
1
版本
#latest

概述

Memory Keeper

Core goal: After /new, the AI picks up exactly where you left off — no re-explaining needed.

Part 0: First-Run Initialization

Trigger: memory/tasks.md exists but is empty.

  1. Tell the user: "Your tasks.md is empty. Let me scan your recent work history and build your memory from scratch."
  2. Read MEMORY.md + most recent session history (openclaw sessions list --limit 1)
  3. Extract in-progress tasks, statuses, next steps
  4. Present to user for confirmation — never write without approval
  5. On confirmation → write to memory/tasks.md
  6. If nothing found → ask user what they're working on

Post-Install Setup (one-time)

See references/install-snippets.md for the three things to set up:

  1. Append memory management section to AGENTS.md
  2. Append daily journal check to HEARTBEAT.md
  3. Initialize memory/tasks.md

> These three configs form a complete memory loop: HEARTBEAT.md creates journals, AGENTS.md restores context on startup, tasks.md persists work state.

Part 1: Task State

See references/formats.md for format specs, quality bar for "Next", and the 3-file rule.

When to update tasks.md — trigger immediately, don't wait:

TriggerAction
-----------------
Milestone reached (version published, bug fixed, module done)Update status/next, or move to Completed
Pause signal ("that's it", "pause", "done for today")Self-check gaps → fill → update Next → tell user "State saved"

Don't trigger for: Pure Q&A, reviewing files, discussing plans without executing.

Multiple in-progress tasks: List all on startup, let user choose. Don't auto-select.

Part 2: Daily Journal

See references/formats.md for journal template and when-to-write triggers.

  • Group by topic, not chronological
  • Skip empty sections entirely
  • Watch List is the most important section

Part 3: Project Index

See references/formats.md for MEMORY.md entry format.

Update immediately after: project added/removed, version released, Git remote/path changed.

Part 4: Dream Consolidation

Version check (every heartbeat)

openclaw --version

Parse version as YYYY.M.R:

  • >= 2026.4.8: OpenClaw has native Dreaming — use it instead
  • Check if native Dream is enabled: look for plugins.entries.memory-core.config.dreaming.enabled: true in openclaw.json, or run openclaw memory promote --dry-run (exit 0 = active)
  • Already enabled → skip our Part 4 entirely, no message
  • Not enabled:
  • Check for upgrade from old version: memory/dream-state.json exists AND lastDream != "2000-01-01"
  • Upgrade detected → tell user: "检测到你升级到了 2026.4.8+,建议开启原生 Dreaming(/dreaming on)。开启后旧 Dream 状态文件会自动清理。"
  • Fresh install (no upgrade) → tell user once: "OpenClaw v2026.4.8+ has built-in Dreaming. You can enable it with /dreaming on in any session."
  • After user enables native Dreaming → delete memory/dream-state.json
  • Then skip our Part 4
  • < 2026.4.8: use our built-in Dream consolidation (see references/dream-guide.md)

Trigger conditions (only for < 2026.4.8)

Run during heartbeat when both are true:

  • lastDream > 7 days ago
  • sessionsSinceLastDream >= 3

For full details — four phases, drift correction, state file — see references/dream-guide.md.

Rules

  • Never record secrets, tokens, or credentials in tasks.md or journals
  • Keep project index at the top of MEMORY.md for fast scanning
  • Absolute dates only: always YYYY-MM-DD — never "next week", "tomorrow"
  • Archive journals by age: move journals older than 30 days to memory/archive/; startup loads last 7 days only
  • MEMORY.md under 200 lines: Dream consolidation handles trimming — never hard-truncate

版本历史

共 1 个版本

  • v1.5.2 当前
    2026-05-03 06:39 安全 安全

安全检测

腾讯云安全 (Keen)

安全,无风险
查看报告

腾讯云安全 (Sanbu)

安全,无风险
查看报告

🔗 相关推荐

ai-agent

self-improving agent

pskoett
捕获经验教训、错误及修正内容,以实现持续改进。适用于以下场景:(1)命令或操作意外失败;(2)用户纠正Claude(如“不,那不对……”“实际上……”);(3)用户请求的功能不存在;(4)外部API或工具出现故障;(5)Claude发现自身
★ 4,099 📥 826,416
ai-agent

Self-Improving + Proactive Agent

ivangdavila
自我反思+自我批评+自我学习+自组织记忆。智能体评估自身工作、发现错误并持续改进。
★ 1,396 📥 322,588
ai-agent

Agent Browser

rez0
用于 AI 代理的浏览器自动化 CLI。当用户需要与网站交互(包括浏览页面、填写表单、点击按钮、截图等)时使用。
★ 835 📥 307,249