Open source public account title Skill, https://github.com/liucongg/liucong-skills/tree/main/skills/gzh-title-strategist … Distilled from 645 articles by me, Canghe @canghe, Daishudi @aikangarooking, Jiamu @jiamu_future, Xiaoli @li_mo60607. The results are amazing so far. Just look at the image. If you also write for a public account, you can directly try it with one of your own articles. Feel free to share feedback in the comments. Hope this Skill helps you all.

liucongg/liucong-skills

Source: https://github.com/liucongg/liucong-skills

liucong-skills

Liu Cong’s NLP Skill repository. Each Skill is placed under skills//, and can be installed and used as needed.

Included Skills

Installation of painterly-3d2-cinema

First, confirm the Skills directory of the Agent or video platform you are using and set it as SKILLS_HOME. A symbolic link is recommended so that updates in the repository take effect immediately.

git clone https://github.com/liucongg/liucong-skills.git
cd liucong-skills
export SKILLS_HOME="/path/to/your-agent/skills"
mkdir -p "$SKILLS_HOME"
ln -s "$(pwd)/skills/painterly-3d2-cinema" "$SKILLS_HOME/painterly-3d2-cinema"

If the target path already exists, confirm whether it is an old link or old copy, then remove or back it up yourself; do not directly overwrite a Skill that is still in use.

You can also install by copying:

export SKILLS_HOME="/path/to/your-agent/skills"
mkdir -p "$SKILLS_HOME"
cp -R skills/painterly-3d2-cinema "$SKILLS_HOME/"

After copying, repository updates will not sync automatically—you need to copy again. Once installed, reload the platform’s Skills list or restart the corresponding Agent so the Skill gets rediscovered.

Usage of painterly-3d2-cinema

Explicitly invoke the Skill in Agent conversations:

$painterly-3d2-cinema Create a 15-second 3D-to-2D action short: a white-haired lightsaber warrior fights mechanical enemies on a rainy lake shore.

You can also describe the requirement directly; when the task involves 3D-to-2D characters, scenes, action storyboards, or short prompts, an Agent that supports automatic Skill recognition can trigger it automatically.

The default workflow is:

1. Provide 3 story directions and wait for selection.
2. Lock in the story and a rhythm of one segment every 15 seconds.
3. Create prompts for characters, character reference boards, scenes, and opponents.
4. Generate a 15-cell action storyboard prompt for Midjourney V8.1 for each segment.
5. Output corresponding Seedance video prompts based on the storyboard.

This Skill delivers only prompts. Images must be generated externally in Midjourney, videos externally in Seedance, then upload results back to the Agent for review and iteration.

Installation of gzh-title-strategist

A symbolic link is recommended so that changes in the repository sync immediately to the Agent’s Skills directory:

export SKILLS_HOME="/path/to/your-agent/skills"
mkdir -p "$SKILLS_HOME"
ln -s "$(pwd)/skills/gzh-title-strategist" "$SKILLS_HOME/gzh-title-strategist"

You can also install by copying:

export SKILLS_HOME="/path/to/your-agent/skills"
mkdir -p "$SKILLS_HOME"
cp -R skills/gzh-title-strategist "$SKILLS_HOME/"

Usage of gzh-title-strategist

Provide the full article, draft, or existing title to the Agent and explicitly invoke the Skill:

$gzh-title-strategist Generate a set of public account titles based on this article, and select the top 3 most publishable ones.

You can also ask it to review existing titles:

$gzh-title-strategist Based on the reading and interaction data of these articles, analyze which title styles are effective and which may be influenced only by topic or publish time.

This Skill first diagnoses the article type, target reader, core conclusion, factual evidence, and deliverable content, then generates titles in categories: robust and accurate, trending and clickable, professional and authoritative, data-keyword, and long-term. By default, it scores the top 5 and gives 1 main title, 2 backup titles, plus risk warnings.

When involving trending keywords or historical data, it uses only the Skill’s built-in historical samples or user-provided evidence. It will not describe keyword relevance as WeChat recommendation algorithm rules, nor add companies, products, numbers, or exaggerated conclusions not present in the article body just for click-through rate.

Updates

With symbolic link installation, simply update the repository:

cd liucong-skills
git pull

With copy installation, after updating the repository, you need to re-copy the corresponding Skill.

Adding a New Skill to the Repository

Each Skill uses an independent directory:

skills/
└── my-skill/
    ├── SKILL.md
    ├── agents/
    │   └── openai.yaml
    ├── references/
    ├── scripts/
    └── assets/

Only SKILL.md is required; other directories can be added as needed. The Skill directory name should match the name in the SKILL.md frontmatter, using lowercase letters, numbers, and hyphens.

After adding, ensure SKILL.md contains valid YAML frontmatter with at least name and description; the directory name must exactly match name. If the platform you are using provides a Skill validation tool, perform structural validation according to the platform’s instructions.

Finally, add an entry and purpose description in the “Included Skills” table of this README.

License

Apache License 2.0