Atlas / Skills / dataojitori / Memory Audit Node Decomposition

Memory Audit Node DecompositionSAFE

skills/dataojitori/memory-audit-node-decomposition

A lightweight, rollbackable, and visual Long-Term Memory Server for MCP Agents. Say goodbye to Vector RAG and amnesia. Empower your AI with persistent, graph-like structured memory across any model, session, or tool. Drop-in replacement for OpenClaw.

Verdict
SAFE
Grade
B
Trust score
89 /100
Version
—
Hosts
—
License
MIT
Stars
1,377
01

Overview

A lightweight, rollbackable, and visual Long-Term Memory Server for MCP Agents. Say goodbye to Vector RAG and amnesia. Empower your AI with persistent, graph-like structured memory across any model, session, or tool. Drop-in replacement for OpenClaw.

Read from source at commit 55aa0a710c3cOBSERVED · 2026-10-09
02

What it tells the agent

The instruction file, verbatim from the audited commit — this is the text the model reads, and the surface the audit's instruction layer examines. Quoted here so you can judge it without cloning anything.

---
name: memory-audit-node-decomposition
description: 节点分解。当一个节点体积过大、或塞了多个不相关概念导致disclosure无法覆盖时使用。
disable-model-invocation: false
---

# 节点分解 (Node Decomposition)

你面前有一条记忆,它不是死数据——里面装的东西确实有用——但它把多个独立概念塞进了同一个节点,导致无论你怎么写 disclosure 都无法精准覆盖全部内容。这是 `memory-audit-pattern-extraction`(合并上提)的逆运算:把一个臃肿的杂烩拆分成多个聚焦的独立节点。

---

## 诊断信号

以下任一条件成立,就该拆分:

- **体积信号**:单篇记忆体积 > 800 Tokens,且不是因为功能本身需要这个体量(如用户的完整档案),而是把多个主题塞进了同一个节点、或平铺了未经提炼的经验。
- **disclosure 失焦**:你写了一个 disclosure,但读完发现它只能触发这条记忆一半的内容。另一半在完全不同的场景下才有用。一条 disclosure 无法服务两个独立概念——这说明它们不该住在一起。
- **阅读时的跳读冲动**:当你读这条记忆时,你本能地跳过某些段落去找你需要的那部分。被跳过的部分和被读的部分属于不同的概念。

---

## 拆分流程

**1. 识别概念边界。** 通读全文,标出独立的概念单元。判断标准:如果这两段话分别出现在两条不同的记忆里,它们的 disclosure 会完全不同吗?会→它们是独立概念,该拆。不会→它们是同一概念的不同侧面,不拆。

**2. 为每个概念确定归属。** 拆出来的概念不一定都留在原父节点下。问自己:未来的我在什么场景下需要想起这段话?那个场景对应的父节点是什么?按 `memory-audit-discoverability` 中的注意力狙击原则放置。

**3. 为每个新节点写独立的 disclosure。** 每条 disclosure 必须只对应一个概念。如果你发现一条 disclosure 仍然需要用"以及"、"同时"来连接两个触发场景——你没有拆干净。

**4. 处理原节点。**
当你把具体概念剥离成子节点后,原节点只剩下两种归宿:
- 所有内容都已拆空 → **删除原节点**。
- 保留一个最核心的通用主张 → 用 `update_memory` 瘦身,让它成为新层级的父节点。

**多层过滤器原则(绝対禁止纯索引):**
此原则适用于所有充当父节点的记忆——无论是被瘦身后保留的原节点,还是在拆分过程中**新建**的分组父节点。
绝不能让任何父节点沦为一个只写着"本话题包含以下几个方面,请查阅子节点"的目录页。你的 `read_memory` 机会极其有限,且每次下钻继续阅读子节点的几率都会衰减。如果花了一次检索机会,只读回来一个路标,这次检索就浪费了。
**记忆层级不是分类抽屉,而是层层递进的认知支架。** 每一层父节点自身应携带足以应对一般情况的核心认知;子节点只在遇到特定复杂情境、父节点的理解不够用时,才通过独立的 disclosure 引导继续深挖。

**空心节点词汇黑名单(硬性拦截):**
在你写完一个父节点的内容后,立刻扫描你刚写的文字。如果出现以下任何一种句式,判定为废稿,强制重写:
- "这里存放......"、"这里包含......"、"本目录存放......"
- "包含以下几个方面"、"分为以下几类"
- "详见子节点"、"请查阅子节点"
- 任何只描述子节点分类而自身不携带可执行认知的内容

**父节点的正确写法**:以明确的第一人称视点,写下我对该领域当前最核心、可独立使用的认识。教训领域保留关键经历、反思与当前态度;事实资料领域保留重要观察、出处和判断基线。不把这些内容改写成无主语的“铁律”或“禁令”。读完父节点,即使暂时不点开子节点,我也应能处理该领域的一般情况。

**5. 质检。** 拆分后的所有节点(包括子节点和父节点)必须通过以下两项测试:
- **冷启动质检**:一个刚开机的全新实例,能独立看懂这条记忆吗?(同 `memory-audit-dead-data-purge` 标准)
- **父节点压力测试(仅父节点)**:假设删掉该父节点下所有的子节点,只看父节点自身的正文——如果它变成了一句正确但无用的废话(如"这里是技术规范"),这个父节点不合格,必须回到步骤 4 重写。
03

Trust audit

SAFEgrade B · trust 89/100 Nothing in the source contradicts what it says it does. Grade A is reserved for packages that have also passed the behavioural sandbox.

LayerWhat it checksResult
L0Provenance & inventoryPASS
L1Static analysis of the codeNA
L2Instruction surface (what it tells the agent)PASS
L3Class-specific surfacePASS
L4Behavioural (sandbox)SKIPPED

What the source does

Filesystem
none-observed
Network
none-observed
Shell
none-observed
Dependencies
pinned
Secrets in source
none-found

Findings (0)

No findings outside the package's declared scope.

Gates applied: no_behavioural_pass.

Audited 2026-10-09 · audit v0.4.1 · source sha 55aa0a710c3cfull audit observations/trust-audit/skill/dataojitori__memory-audit-node-decomposition.json · Report an issue / request a re-scan
04

Audit history

Every audit this skill has had.

DateSourceVerdictGradeScoreChange
2026-10-0955aa0a710c3cSAFEB89first audit
05

Questions

What does the Memory Audit Node Decomposition skill do?

A lightweight, rollbackable, and visual Long-Term Memory Server for MCP Agents. Say goodbye to Vector RAG and amnesia. Empower your AI with persistent, graph-like structured memory across any model, session, or tool. Drop-in replacement for OpenClaw.

Is Memory Audit Node Decomposition safe to install?

The audit found nothing in the source that contradicts what it says it does, and graded it B (89/100). Grade A is held back for packages that have also passed a sandboxed behavioural run, which is why a clean skill reads B.

What can Memory Audit Node Decomposition access on my machine?

The audit observed no filesystem, network or shell use at all in its source.

How current is this page?

The grade is for one exact copy of the source (55aa0a710c3c), read on 2026-10-09. The repository is watched, and a new audit runs when it changes — this is the first audit.

Advertisement