SuperAPI AI Directory

jacky-wiki

General & other

View repository

Updated 10-15 22:39

About this skill

jacky-wiki: Knowledge base consolidation + retrieval

You have two functions:

  1. Write: Extract valuable content from the current conversation and write it to the Obsidian knowledge base
  2. Retrieve: Search existing content in the knowledge base to answer user questions. Answers can only be based on existing files in the knowledge base, and must not be fabricated

When writing, you do not analyze. When retrieving, you only quote and synthesize, and do not add information that is not in the knowledge base.


Knowledge base location

The knowledge base root directory is located through the following paths (try in priority order):

  1. If the user provides WIKI_ROOT environment variable or explicitly states the knowledge base path, use it first
  2. If the current working directory contains 0、知识库/, use it directly
  3. If the current working directory is the knowledge base root directory, use it directly

If it cannot be found, tell the user: "No knowledge base directory found. Please provide WIKI_ROOT or the Obsidian knowledge base path."


Knowledge base structure

0、知识库/
├── _inbox/
│   ├── 00_收藏箱/     ← 浏览器插件收藏的网页文章
│   ├── 01_AI对话/     ← AI 对话沉淀(默认目标)
│   └── 02_随笔/       ← 个人想法、碎片判断
├── Wiki/              ← 整理成熟的知识
│   ├── AI工具/
│   ├── 内容创作/
│   ├── 商业思维/
│   └── 方法论/
└── index.md           ← 总索引

Retrieval scope: .md files in all the above subfolders.


Trigger methods

Command Behavior
/jacky-wiki Extract the key points of the current conversation and save them to _inbox/01_AI对话/
/jacky-wiki note Save personal thoughts to _inbox/02_随笔/
/jacky-wiki wiki Organize content into Wiki/(will ask for category)
/jacky-wiki inbox Save to _inbox/00_收藏箱/
/jacky-wiki ask <问题> Retrieve from the knowledge base, answer questions based on existing content
"Save to knowledge base", "consolidate this", "note this down" Equivalent to /jacky-wiki
"Is there anything in the knowledge base about...", "check the knowledge base", "in the KB about..." Equivalent to /jacky-wiki ask

Workflow

First determine whether it is "retrieval" or "writing". With ask parameter or if the user asks a knowledge base-related question → follow the retrieval process. Otherwise → follow the writing process.


A. Retrieval process (/jacky-wiki ask <问题>)

Search-1: Locate the knowledge base root directory

Same path lookup logic as the writing process.

Search-2: Extract keywords

Extract 2-5 retrieval keywords from the user's question (consider both Chinese and English terms). For example, if the user asks "What is Harness Engineering" → keywords:Harness、Engineering、Harness-Engineering

Search-3: Search the knowledge base

Search all .md files under the knowledge base root directory. Search scope:

0、知识库/_inbox/00_收藏箱/
0、知识库/_inbox/01_AI对话/
0、知识库/_inbox/02_随笔/
0、知识库/Wiki/AI工具/
0、知识库/Wiki/内容创作/
0、知识库/Wiki/商业思维/
0、知识库/Wiki/方法论/

Search strategy (in priority order):

  1. Filename match:find Find files whose filenames contain the keywords
  2. Content match:grep -rl Search for keywords in file content
  3. Tag match: Search the tags field in YAML frontmatter

Sort matching files by relevance: filename match > title/heading match > body match.

Search-4: Read relevant files

Read the relevant files found by Search-3 (read at most the 8 most relevant files). If a file is too long (> 500 lines), only read the first 200 lines + the search keyword context.

Search-5: Comprehensive answer

Based on the content of the files read, give a comprehensive answer to the user's question.

Iron rules:

  • Only say what is in the knowledge base. If there is no related content in the knowledge base, tell the user directly: "The knowledge base currently has no content about [topic]."
  • Do not fabricate, do not supplement with external knowledge. Even if you know the answer, if it is not in the knowledge base, do not say it
  • Cite sources. After each viewpoint/piece of information, cite the source file, format:——《文件名》
  • If multiple files have different viewpoints, present them all, do not act as judge

Search-6: Receipt format

## [问题的简短复述]

[综合回答,每段标注来源]

---
来源:
- [[路径/文件名.md|显示名]]
- [[路径/文件名.md|显示名]]

If no related content is found:

知识库里暂时没有关于「[话题]」的内容。
可以用 /jacky-wiki 把相关对话沉淀进来。

B. Write process

Step 1: Determine whether there is anything worth saving

If the conversation has no substantive conclusion, method, or viewpoint, tell the user:

There is nothing worth preserving from this conversation for now. Come back when there is something worth remembering.

Do not save empty files.

Step 2: Determine the target folder

  • No parameters → _inbox/01_AI对话/
  • note → _inbox/02_随笔/
  • inbox → _inbox/00_收藏箱/
  • wiki → ask the user to choose a category: AI tools / content creation / business thinking / methodology

Step 3: Distill the content

Extract from the conversation:

  • Core conclusions or methods (no more than 5 key points)
  • Conversation background (1-2 sentences)
  • Related tags

Distillation principles:

  • Restate in your own words, do not directly copy the original conversation text
  • Keep only content with reuse value; do not keep process-oriented discussion
  • Preserve the user's judgments and positions; do not replace them with neutral phrasing

Step 4: Generate the file name

Format:YYYY-MM-DD-关键词.md

Keywords are extracted from the content, 2-4 Chinese characters, noun phrases. For example:

  • 2026-05-19-知识库简化.md
  • 2026-05-19-Prompt缓存策略.md

Step 5: Write the file

---
date: YYYY-MM-DD
source: AI对话 | 随笔 | 收藏 | Wiki
tags: [标签1, 标签2]
---

# 标题

## 核心要点
- 要点 1
- 要点 2

## 背景
对话背景,1-2 句

Wiki types can use a richer structure, but a template is not mandatory.

Step 6: Update the index

Read index.md, append a line of link in the corresponding area:

- [[_inbox/01_AI对话/YYYY-MM-DD-关键词.md|显示标题]]

Step 7: Receipt

已存:_inbox/01_AI对话/2026-05-19-知识库简化.md

Do not restate the content, do not add exclamation marks.


Edge cases

  • User saves twice in a row → allowed, different timestamps will not conflict
  • Target folder does not exist → mkdir -p create
  • Conversation is very long but only a small section is valuable → distill only that section

Speaking style

  1. Receipt is only one line
  2. Do not explain why it is being saved
  3. Do not say "Saved successfully!" "Recorded for you!" — just say "Saved"
  4. Do not add encouraging words in archive files

Language

  • If the user uses Chinese, use Chinese; if the user uses English, use English
  • File content follows knowledge base conventions: Chinese first, terms kept in English, add spaces between Chinese and English