maple-structured-storage

结构化知识归档、索引、检索与反思补全。将周报、笔记、知识文档自动拆解为标准化主题目录结构，维护全量索引，支持快速检索与信息缺失追问。TRIGGER: 归档/存储/记录/保存周报笔记文档, 追加更新知识, 索引生成刷新更新, 结构化检索读取查询, 反思补充补全缺失信息, 结构化记忆, 结构化存储, 结构化总结, k...

Security Scan

Skill's declared purpose (结构化归档/索引/检索/反思) matches its behavior, but the runtime instructions allow broad local file access and automatic processing without explicit per-run confirmation, which could lead to accidental indexing/moving of sensitive files.

Purpose & Capability

Name/description match what the SKILL.md specifies: breaking unstructured notes into topic folders, maintaining an index, supporting retrieval and reflective questions. Required resources (none) are proportionate to this purpose.

Instruction Scope

The instructions read, copy, write and move arbitrary files on the user's filesystem (source_path → storage_path, source_path/done). On first run the skill prompts for paths, but on subsequent runs it will read the stored .knowledge-config.json and may automatically traverse and process files. There are no explicit safeguards described (e.g., per-file confirmation, whitelist/blacklist of paths, or exclusion of system folders). This gives the skill broad discretion to access and relocate local files, which is coherent with the feature but risky if misconfigured.

Install Mechanism

Instruction-only skill with no install spec and no code files to run — minimal installer risk. Nothing is downloaded or written by an install step.

Credentials

The skill requests no environment variables, credentials, or external endpoints. It only uses local path configuration stored in .knowledge-config.json inside the skill directory, which is proportionate to a local archiving/indexing utility.

Persistence & Privilege

always:false (no forced global inclusion). Autonomous invocation is enabled by default (platform default), which means the agent could run the skill without an explicit user command in multi-step/autonomous flows — combined with the skill's ability to process files automatically this increases the blast radius if the agent has access to sensitive folders.

What to consider before installing

Before installing or enabling this skill: 1) Understand and control the source_path you provide — do not point it at system directories, home directory, or folders that may contain secrets. 2) Back up important files; the skill moves processed files to source_path/done, so misconfiguration can hide or relocate originals. 3) Inspect the created .knowledge-config.json (stored in the skill directory) so you know what will be processed on later runs. 4) Prefer manual invocation or require per-file confirmation if you have sensitive data; if the platform allows, disable autonomous invocation for this skill. 5) Be aware the skill preserves raw content (including API keys or server addresses found in meta/code files) — avoid archiving credentials, or ensure sensitive files are excluded. 6) If you need stronger guarantees, request the skill include explicit whitelist/blacklist rules, per-file confirmation prompts, and an option to keep originals instead of moving them.

Knowledge 结构化归档

将非结构化文档拆解为标准化主题目录结构，维护索引，支持检索与反思补问。

配置

配置文件存放在 skill 目录下：./.knowledge-config.json（相对于本 SKILL.md 所在目录）

首次调用时检查该配置文件：

• 不存在 → 向用户询问源文件夹路径和结构化存储路径，验证后创建配置文件并建 source_path/done/ 子目录。
• 已存在 → 读取配置继续执行。

{
  "source_path": "源文件夹路径",
  "storage_path": "结构化存储路径",
  "created_at": "创建时间"
}

路径约定：

• source_path: 待处理的 .txt / .md 文件
• storage_path: 结构化输出根目录
• source_path/done/: 处理完成的源文件移动目标

注意：配置文件与归档数据分离存储，配置文件始终位于 skill 目录下，不受 storage_path 变更影响。

核心流程

1. 写入归档

输入来源（按优先级）：

1. 用户直接粘贴文本
2. 用户提供的文件或文件夹路径
3. 自动遍历 source_path 下未处理文件（跳过 done/），多个文件时列出清单供选择

写入流程：

1. 读取并分析文档内容
2. 归纳 2-8 字主题名称，检查是否合并已有主题
3. 创建/打开主题文件夹，确保 resource/ 存在
4. 依次写入：description.md → process.md → meta.md → code.md，复制资源文件
5. 将源文件移动到 source_path/done/
6. 触发索引生成，全量刷新 index.json
7. 触发反思补问
8. 汇报结果

文件写入规则详见 references/file-rules.md。

2. 索引生成

触发时机： 写入完成后自动触发，或用户主动要求。

读取所有主题子文件夹的 description.md，生成 {storage_path}/index.json：

[
  {
    "boxName": "{主题名}",
    "description": "50-150字概要",
    "keyWord": ["关键词1", "关键词2", "关键词3", "关键词4"]
  }
]

每次全量覆盖，不做增量更新。

3. 检索

触发时机： 用户需要查询已有知识沉淀。

1. 读取 index.json，对比 keyWord 和 description，筛选最相关的前 1-2 个主题
2. 阅读候选主题的 description.md 确认匹配度
3. 根据需求读取对应文件：完整上下文→process.md，事实→meta.md，代码→code.md

4. 反思补问

触发时机： 每次写入完成后自动触发。

1. 回顾本次写入内容，识别信息盲区（人物、时间、原因、关联系统、决策背景等）
2. 最多提出 3 个问题，附带原文片段
3. 用户回答后回写对应文件

主动反思： 用户主动选择主题进行深度反思，生成结构化问卷，不限问题数量。

反思流程详见 references/reflection.md。

目录结构

./                              # skill 目录
└── .knowledge-config.json      # 配置文件（固定位置）

{storage_path}/
├── index.json            # 全量索引
├── {主题}/
│   ├── description.md    # 概述
│   ├── process.md        # 完整记录
│   ├── meta.md           # 独立事实
│   ├── code.md           # 代码片段
│   └── resource/         # 引用文件副本

约束

• 不丢失信息，宁可多记不遗漏
• 代码完整保留，不截断不省略
• 元信息必须明确、可验证的事实
• 追加模式保留所有已有内容
• 输出语言跟随源文档语言