---

name: agent-memento description: Tick-Driven Autonomous Production Factory for LLMs. A framework for long-running agents using Cron/Heartbeats and physical Markdown checklists to prevent OOM and context decay. metadata: {"clawdbot":{"emoji":"🧠","requires":{"bins":["bash","openclaw"],"env":[]}}}

Agent Memento (Tick-Driven Autonomous Execution)

“Don't build AI that tries to remember everything. Build systems that make AI read the blueprint anew every 5 minutes.”

当我们在聊天框里让大模型“写一个游戏”时，开头总是美好的。但随着代码超过 2000 行，模型开始：

• 失忆症 (Amnesia)：忘记了自己定义的接口，开始凭空捏造。
• 幻觉重构 (Hallucination)：自信满满地输出一堆根本没有跑过的“完美”代码。
• 上下文核爆 (Resource Overload)：为了修一个小 Bug，强行覆盖整个大文件，导致内存耗尽 (OOM)。

核心原则：把“全知全能的超级 Agent”降级为无数个“只有 5 分钟记忆的、无情的 Tick 机器”。

The Methodology：双态智能与外部状态机

1. 状态剥离（物理外置大脑）：Agent 不在内存池中记忆项目历史，长线记忆与当前状态全部下放至物理文件（MASTER_PLAN.md 验证清单）。
2. 架构师态（The Architect）：人类与大模型在主 Session 里讨论，将宏大叙事翻译成 MASTER_PLAN.md 里的每一个 [ ]。
3. 泥瓦匠态（The Tick Worker）：Cron 脚本每隔几分钟自动唤醒一个全新的、绝对专注的短命进程。它只寻找清单里第一个未打钩的 [ ]，修好它，执行物理验证，然后离线。

系统化实施方法 (How to Deploy)

当用户说要启动 Memento 流水线时，使用以下命令：

bash skills/agent-memento/scripts/init_memento.sh <ProjectName>

铁血纪律 (Execution Sandbox)

作为一个被 Memento 唤醒的 Tick Worker（或要求你写 Tick 脚本时），你必须遵循：

1. 防爆底线 (Surgical Edits Only)：严禁输出完整文件要求覆写。必须通过精准定位工具修改。
2. 拒绝脑内验算 (Evidence-based Audit)：跑脚本看真实退出码，没有 Exit Code 0，不许在 [ ] 上画一个 x。
3. 不要静默空转 (No Silent Ghosting)：卡死时，把死亡原因写进 TICK_STATUS.md 中。

Security & Transparency

Data Access — Exactly What Is Read

• Reads <project-dir>/docs/MASTER_PLAN.md to determine the current task.
• Reads source code files within the target project directory as needed to accomplish the task.

Data Written

• Creates project scaffolding (MASTER_PLAN.md, TICK_STATUS.md, *_tick.sh).
• Updates [ ] to [x] in MASTER_PLAN.md via tick worker.
• Edits source files within the target project directory.

Persistence & Autonomous Scheduling

• The init_memento.sh script generates a bash script intended to be run via cron (e.g., */5 * * * *). The user must manually add this to their crontab to enable autonomous execution.
• You can stop the agent at any time by removing the entry from crontab -e or killing the background process.

Credentials

• Operates entirely locally. Relies on the default openclaw agent command, utilizing the underlying system's existing LLM configurations. No new keys are requested or stored.

Network Access

• No outbound network access is initiated by the core framework scripts themselves, except when the user's specific project tasks require installing packages (e.g., npm install).