SuperAPI AI Directory
Partner with us中文AI newsFavorites
Skillscontent-goldmine-gemini

content-goldmine-gemini

General & other

View repository

Updated 10-16 09:35

About this skill

Content Goldmine (Gemini CLI version)

Break collected articles, especially English articles, into long-term reusable Chinese material assets, and stably store them into the library according to the existing material library template.

Trigger method

Triggered when the user says these things:

  • Break this collected article into material fragments
  • Analyze this batch of clippings and deposit them into the material library
  • Process these articles according to content-goldmine
  • Help me break these collections into titles/quotes/structures
  • Break this English article into Chinese material library assets
  • Batch process Clippings recently collected articles in the directory
  • Process this article with gold mining

Core capabilities

  1. 4D underlying breakdown: philosophical layer, psychological layer, communication layer, social layer
  2. Structural analysis: narrative methods of the beginning, middle, and ending
  3. Psychological pattern recognition: paradox/transformation arc/pain point resonance, etc.
  4. Fragment material extraction：
    • Titles (pass/fail screening)
    • Openings (pass/fail screening)
    • Quotes (pass/fail screening)
    • Structural formulas (pass/fail screening)
  5. Chinese output: All final materials stored in the library must be in Chinese, suitable for direct reuse in subsequent creation

Input modes

Mode A: single file

把这篇 clipping 拆进素材库：00_收件箱/Clippings/Dan Koe/xxx.md

Mode B: batch directory

把 00_收件箱/Clippings/Dan Koe/ 下面最近 20 篇都拆了

Mode C: specified file list

只处理这 3 篇：A.md、B.md、C.md

Output directory

Strictly inherit the original command's path rules:

  • 02_素材库/灵感素材/标题/ - Titles that passed screening
  • 02_素材库/灵感素材/开头/ - Openings that passed screening
  • 02_素材库/灵感素材/金句/ - Quotes that passed screening
  • 02_素材库/灵感素材/结构/ - Structural formulas that passed screening
  • 02_素材库/高表现内容/长文/ - Complete deconstruction (>800 words)
  • 02_素材库/高表现内容/短图文/ - Complete deconstruction (≤800 words)

Technical architecture

Analysis engine

  • Default model：gemini-3-flash-preview
  • Invocation method：Gemini CLI one-shot
  • Output format: structured JSON

Local script

  • Unified entry point：scripts/run_content_goldmine.sh
  • bootstrap：scripts/bootstrap.sh
  • Core script：scripts/process_goldmine.py
  • Responsibilities：
    1. Read article Markdown
    2. Call Gemini CLI to output structured JSON
    3. Keep only fragments that pass
    4. Stably write to disk according to the existing material library template
    5. For English content, keep the original text, and additionally supplement ## 中文转写

First-time use instructions

This version has already been adapted to Claude Code.

If this is your first time getting this skill:

  1. Do not directly run process_goldmine.py
  2. Run uniformly scripts/run_content_goldmine.sh
  3. It will first check Python 3 and Gemini CLI
  4. If Gemini CLI is not installed, it will prioritize trying to install it automatically
  5. If Gemini CLI is not logged in yet, it will clearly guide you to first run gemini to complete login authorization

For more complete instructions, see:

  • references/config/first-time-setup.md
  • README-学员分发包.md

Design Principles

  • Gemini is responsible for extraction, not for the final template output to disk
  • The script is responsible for consistency in paths, naming, frontmatter, and heading hierarchy
  • The reason for this is very practical: when running in batches over the long term, the material library cannot change format every day

Fragment Quality Standards (Cleanliness Method)

Only store "reusable assets", not noise.

Title (must satisfy all of the following)

  1. Can restate the viewpoint or promise at a glance (not a "generic topic")
  2. Has at least 1 hook: counterintuitive / contrastive tension / time or number anchoring / identity callout / strong question
  3. Do not "pile up soft words" (such as: improve / empower / share / think / brief discussion / how to...)

Opening (must satisfy all of the following)

  1. 1-3 short paragraphs (total length should be <= 120 characters as much as possible)
  2. The first three sentences must accomplish at least 1 thing: conflict / suspense / data / scenario / rhetorical question / counterintuitive point
  3. Prohibit "academic-style setup"

Punchline (must satisfy all of the following)

  1. A single sentence can spread independently (it still holds outside of context)
  2. As short as possible (for Chinese, recommended <= 40 characters)
  3. Must contain one of judgment / priority / contrast / causality

Structural Formula (must satisfy all of the following)

  1. Is a "template that can be applied", containing placeholders or clear step relationships
  2. Must not be a long retrospective that "only holds for this one article"
  3. After reading it, it can be immediately used on another topic

Execution Flow (Single Article)

  1. Read the article Markdown
  2. Extract title, source, and body
  3. Call Gemini CLI (model:gemini-3-flash-preview）
  4. Require Gemini to output JSON according to the 4D framework
  5. Script validates JSON
  6. Keep only the titles / openings / punchlines / structural formulas that pass
  7. Write to the corresponding directory according to the material library template
  8. Return the write manifest

Current Storage Template

Uniformly stay close to your existing material library files:

---
source: ...
url: ...
collected: ...
from_article: ...
---

## 内容

[原文片段]

## 中文转写

[中文可复用版]

## 为什么好

- ...

## 4D标签

- 哲学层：...
- 心理层：...
- 传播层：...
- 社会层：...

## 可变体方向

- ...

Hard Rules for Chinese Output

Even if the original text is in English, the final stored material must also be in Chinese:

  1. All final stored content uses Chinese
  2. The English original text is only used as a reference field, optionally retained, and does not enter the default material body
  3. Punchlines, titles, and structural formulas must all be done as Chinese rewritten versions, not word-for-word literal translations
  4. Chinese expression should prioritize:
    • Reusable
    • Spreadable
    • Like Chinese content creation material
    • Does not retain an obvious English tone

Example Usage

Run help first the first time (it will also do bootstrap along the way)

CONTENT_GOLDMINE_ROOT="<你的 content-goldmine-gemini 目录>"
bash "$CONTENT_GOLDMINE_ROOT/scripts/run_content_goldmine.sh" --help

Single Article

CONTENT_GOLDMINE_ROOT="<你的 content-goldmine-gemini 目录>"
bash "$CONTENT_GOLDMINE_ROOT/scripts/run_content_goldmine.sh" \
  --input "00_收件箱/Clippings/Dan Koe/2025-11-20 How to articulate yourself intelligently.md"

Batch Directory (most recent 10 articles)

CONTENT_GOLDMINE_ROOT="<你的 content-goldmine-gemini 目录>"
bash "$CONTENT_GOLDMINE_ROOT/scripts/run_content_goldmine.sh" \
  --dir "00_收件箱/Clippings/Dan Koe" \
  --count 10

Specified File List

CONTENT_GOLDMINE_ROOT="<你的 content-goldmine-gemini 目录>"
bash "$CONTENT_GOLDMINE_ROOT/scripts/run_content_goldmine.sh" \
  --url-file "00_收件箱/articles.txt"

Save the complete deconstruction draft

CONTENT_GOLDMINE_ROOT="<你的 content-goldmine-gemini 目录>"
bash "$CONTENT_GOLDMINE_ROOT/scripts/run_content_goldmine.sh" \
  --input "00_收件箱/Clippings/Dan Koe/example.md" \
  --save-full-analysis

Success Criteria

  1. Can process a single Markdown article
  2. Can correctly write to disk according to the established directory
  3. The final output is in Chinese
  4. Only write passing materials
  5. File naming is stable and does not overwrite existing files

This skill turns "feels good" into "knows why it is good", and then into "Chinese material assets that can be reused at any time".