SuperAPI AI Directory
Partner with us中文AI newsFavorites
Skillsnote-meta-skill

note-meta-skill

Writing & research

View repository

Updated 10-15 19:55

About this skill

Note Meta Skill (Knowledge Extractor)

Extract high-quality knowledge from the public internet or user-provided documents, refine core content such as workflows, methodologies, and principles, and automatically encapsulate it into a standardized Skill.

Prerequisites

  1. NotebookLM Skill is installed: This skill depends on the notebooklm skill for knowledge collection
  2. Skill Creator is available: Used for final skill encapsulation
  3. NotebookLM login has been completed：notebooklm login

Proxy Configuration (Required in Mainland China)

Set the proxy before all notebooklm commands:

$env:HTTP_PROXY="http://127.0.0.1:7890"
$env:HTTPS_PROXY="http://127.0.0.1:7890"

Login Notes

⚠️ notebooklm login After completing,you must return to the terminal and press ENTERto save the authentication state.

Common Issues Quick Reference

Issue Solution
All connection attempts failed Set HTTP_PROXY/HTTPS_PROXY
Auth expired still reported after login After login, you must press ENTER
TargetClosedError Set $env:NOTEBOOKLM_HOME="~/.notebooklm_fix" to switch to a new directory

Workflow

用户输入 (主题/URL/文档)
        ↓
  阶段1: 知识收集
        ↓
  阶段2: 知识萃取
        ↓
   ★ 确认点 ★
        ↓
  阶段3: Skill 封装
        ↓
      交付 Skill

Phase 1: Knowledge Collection

Step 1.1: Clarify the Goal

Ask the user:

  • Topic: "From which field/topic do you want to extract knowledge?"
  • Source: "Do you have ready-made documents/URLs? Or do you need AI to search for you?"
  • Expected Output: "What type of Skill do you want to generate? (Workflow guidance / Methodology reference / Automation scripts)"

Step 1.2: Create a NotebookLM Notebook

notebooklm create "知识萃取: [主题名称]"

Step 1.3: Collect Knowledge Sources

Method A - AI Autonomous Search (Deep Research):

notebooklm source add-research "[搜索查询]" --mode deep --no-wait
notebooklm research wait --import-all

Method B - User-Provided Sources:

# 添加 URL
notebooklm source add "https://..."

# 添加本地文件
notebooklm source add ./document.pdf

Method C - Hybrid Mode (Recommended): First add user documents, then use Deep Research for supplementary search.

Step 1.4: Verify Sources Are Ready

notebooklm source list --json

After confirming that all sources have status = READY, continue.

Phase 2: Knowledge Extraction

Use structured questioning to extract core knowledge from NotebookLM.

Extraction Dimensions

Execute the following questions in order (for the complete Prompt, see extraction_prompts.md）：

Dimension Purpose Core Question
Workflow Extract step-by-step process "What is the complete workflow of this method? What specifically does each step do?"
Principles Extract core principles/methodology "What are the core principles behind it? Why do it this way?"
Templates Extract reusable templates "Are there templates, frameworks, or checklists that can be reused directly?"
Scripts Detect whether there are scripts "Does it involve code, scripts, or automation tools?"

Execute Extraction

# 依次执行萃取提问
notebooklm ask "[Workflow 萃取 Prompt]" --json
notebooklm ask "[Principles 萃取 Prompt]" --json
notebooklm ask "[Templates 萃取 Prompt]" --json
notebooklm ask "[Scripts 萃取 Prompt]" --json

Organize Extraction Results

Organize all extraction results into a structured document in the following format:

# 知识萃取报告: [主题名称]

## Workflow (工作流)
1. 步骤1: ...
2. 步骤2: ...

## Principles (核心原则)
- 原则1: ...（解释 Why）
- 原则2: ...

## Templates (模板)
[如有]

## Scripts (脚本)
[如有，提取代码；如无，标注"无"]

★ Confirmation Point ★

Pause and show the extraction results to the user, to confirm:

  1. Is the extracted content accurate and complete?
  2. Is there anything that needs to be supplemented or modified?
  3. After confirmation, proceed with Skill packaging.

Phase 3: Skill Packaging

Step 3.1: Determine Skill metadata

Determine based on the extraction results:

  • name: lowercase + hyphens, e.g. writing-master
  • description: [action] + [scenario] + [trigger word]
  • Type: workflow guidance / methodology reference / automation script

Step 3.2: Create the directory structure

mkdir -p ~/.gemini/antigravity/Skills/[skill-name]/references

Step 3.3: Generate SKILL.md

Use skill_template.md as the base template, and fill in:

  • Frontmatter：name, description
  • Overview: description of the Skill's purpose
  • Workflow: generated from the Workflow extraction results
  • Core Principles: generated from the Principles extraction results
  • Templates/References: link to the references/ directory
  • Examples: provide 1-2 usage scenarios

Step 3.4: Populate references/

Create based on the extracted content:

  • principles.md: detailed methodology
  • templates.md: reusable templates
  • [其他].md: add as needed

Step 3.5: Handle Scripts (if any)

If the extraction results include scripts:

mkdir -p ~/.gemini/antigravity/Skills/[skill-name]/scripts

Save the script code to the scripts/ directory.

Output Checklist

Final deliverables:

[skill-name]/
├── SKILL.md              # 主技能文件
├── references/           # 参考资料
│   ├── principles.md     # 核心原则（如有）
│   └── templates.md      # 模板（如有）
└── scripts/              # 脚本（如有）

Example

User request: "I want to learn this writing blogger's methods; help me turn them into a Skill"

Execution flow：

  1. Create notebook notebooklm create "知识萃取: 写作方法论"
  2. Add the blogger's article URL
  3. Deep Research supplementary search "writing techniques methodology"
  4. Extract Workflow, Principles, Templates
  5. User confirms the extraction results
  6. Generate writing-master/SKILL.md
  7. Deliver the skill

Notes

  • Quality over speed: better to run multiple rounds of extraction than to miss key content
  • Dynamic Scripts: scripts are not mandatory; include them if present, omit them if not
  • Language style: the generated Skill should be in Chinese, consistent with user habits