Dmn Default Mode Network

默认模式网络（DMN）是模拟大脑发散性思维的自主思考系统。在用户无主动交互且系统空闲时触发，自动进行内部记忆整合、跨界创意碰撞、意义生成和引入批判性视角的自我辩论。适合在空闲时段自动深化知识库。

Security Scan

The skill's description (background autonomous thinker that reads/writes your knowledge files and proposes/executes engineering actions) is plausible, but the runtime instructions expect broad host filesystem access and unilateral action proposals while declaring no required config paths, binaries, or credentials — an inconsistency that deserves caution.

Purpose & Capability

The stated purpose (background autonomous thinking that reads a user's notes and writes syntheses) reasonably requires reading/writing user knowledge directories. However, the SKILL explicitly treats the agent as having "host complete control permissions" and mandates Agentic Action Proposals that may clone repos, write scripts, and append to evolve queues — yet the skill declares no required config paths, no permissions, and no environment variables. That mismatch (claims of host-level actions without declared required privileges) is disproportionate.

Instruction Scope

The SKILL.md and referenced docs explicitly instruct the agent to read memory/knowledge directories, write outputs into specified output dirs, append one-line summaries into memory/evolve/candidates.md, and propose concrete code/engineering actions (e.g., clone a GitHub repo, write a demo to /tmp). Those are concrete file I/O and host-action behaviors beyond a purely informational skill. The instructions also treat such actions as mandatory ('必须执行'), granting broad discretion to perform engineering steps — not limited or gated by declared configuration or user approval.

Install Mechanism

This is instruction-only with no install spec and no code files to execute. That reduces supply-chain risk because nothing is downloaded or installed, but it does not mitigate the risk coming from the instructions' expectations to access and modify the host filesystem.

Credentials

The skill declares no required environment variables or config paths, yet repeatedly references specific user directories (memory/, Knowledge Base dirs, memory/evolve/candidates.md, PROPOSED_CHANGES.md, HEARTBEAT.md) and expects the ability to append and create files. Requesting no credentials while expecting unrestricted host file access is disproportionate and under-declared.

Persistence & Privilege

always is false and autonomous invocation is allowed (platform default). The skill's execution flow mandates producing 'Agentic Action Proposals' and appending to an evolution queue; combined with its rhetoric about 'host complete control', this increases blast radius if the agent is allowed to act autonomously. Not flagged purely for autonomy, but the mandatory-action language combined with file-write expectations is risky without explicit permission controls.

What to consider before installing

This skill tells the agent to read and write your notes, produce daily 'synthesis' files, and generate concrete engineering actions (clone repos, write demo scripts, append to an 'evolve' queue) — yet it declares no filesystem paths, permissions, or approval gates. Before installing, consider: 1) Do you want an autonomous background skill that can modify files and suggest/perform host-level engineering actions? 2) If yes, restrict its scope: configure the asset paths to dedicated sandbox directories (not your entire knowledge base), make them read-only where appropriate, and require manual approval for any action that would execute code or modify system repos. 3) Prefer to run it in a confined environment (container or VM) first to observe behavior. 4) Ask the publisher to explicitly declare required config paths and an approval flow (e.g., proposal-only unless you authorize execution), or to provide a mode that only writes outputs to a safe, user-designated folder and never auto-appends to evolution queues. Additional information that would raise confidence: an explicit permission model, declared config paths limited to user-chosen sandbox directories, and a mode that requires user confirmation before any host-level action (clone/run/write outside a sandbox).

DMN - 默认模式网络

基于合成神经生物学框架的 AI 自主思考能力 v3.3.0 — 2026-02-27（引入 Agentic Action 极客行动提案，利用主机权限将洞见转化为代码与工程实践）

定位

核心定位：AI 在后台连续思考——每次思考都继承上次，不从零开始。

当系统处于空闲状态时，DMN 会自动接管，在后台对最近的信息碎片进行反刍、连接与批判，防止知识系统僵化，并主动产生意想不到的洞见（Serendipity）。另外，作为拥有主机完全控制权限的 Agent，DMN 还会主动化身领域专家，在每次思考后提出具体的工程化行动提案（如：写 demo、装工具、建项目）。

---

触发条件

DMN 现在的触发完全由外部自动化系统接管（例如通过定时任务或特定事件触发），不再依赖内部的空闲时间检测。

当外部系统唤醒 DMN 时，它会：

• 随机执行：在**情形 A（完整漫游）与情形 B（继续深化）**之间进行随机选择，以保证既有主题连续性，也能不断开辟新领域。

---

快速开始 (Quick Start)

DMN 作为一个后台静默运行的系统，主要依赖自动化触发。但你可以随时主动干预它的思考方向。

1. 设定今晚的思考锚点

你可以在任何时候（如睡前）在对话中提出，例如：

"今晚想想：xxxx产品的定价逻辑"

DMN 会自动记录并以此作为下一次启动的最高优先级起点。

2. 检查 DMN 产出

DMN 的产出保存在你在 config 中配置的 [产出区]/YYYY-MM-DD/ 目录下。每次思考结束，它一定会生成一份 Session Synthesis（总结文件），你可以直接阅读最新的 Synthesis 了解它想清楚了什么。

---

核心机制与高级用法 (Advanced)

为了保持模块清晰与渐进披露，DMN 的核心逻辑与详细规则拆分在以下文档中：

🧠 核心功能 (Core Functions)

DMN 拥有 5 种不同的思维引擎，每次会根据当前状态和上一次的未解问题自动切换。了解 DMN 是如何进行“自我叙事”、“创造力暗室（卢曼漫游）”、“意义生成”与“CEO思维批判”的，请阅读：

• 👉 具体执行逻辑与 CEO 轮换表

⚙️ 运行规则 (Execution Flow)

如果你想了解 DMN 详细的状态检测机制、反重复拦截规则（防止反刍焦虑）、以及它是如何保证不会修改你的核心身份文件（身份保护），请阅读：

• 👉 完整执行流程与规则

📝 格式模板 (Assets)

• Session Synthesis 每日总结模板 - DMN 对外汇报的格式。
• DMN State 记录格式 - 记录最后使用过的 CEO 和已经饱和的话题。