Keep Xiaohuang Consistent: cs-xiaohuang-skill for Agents

Where this fits

Your agent turns Chinese articles from Notion, WeChat, or blogs into visual content. The character 小黄 (Xiaohuang) must stay recognizable in every illustration, but you don't want to re-describe its design each time or debug character drift. This skill gives the agent one entry point for editorial illustrations and brand IP extension.

Why agents benefit

• Reads assets/reference-xiaohuang.png and references/character-dna.md as identity anchors, so every image uses the same character instead of a vague prompt.
• For article illustrations, it identifies cognitive anchors and outputs a shot list before generating, reducing wasted images and keeping visuals tied to the text.
• Local edits preserve composition: title, typo, or annotation changes don't force a full regeneration.
• Supports 2D/3D, black-and-white, stickers, expressions, collabs, style transfer, and identity repair while protecting the role identity.
• A QA checklist catches identity and style drift before delivery.

Practical scenarios

Turn a Chinese essay into 16:9 editorial images

A developer has a Notion post about a methodology. The agent reads references/style-dna.md, maps each section to a cognitive anchor (“温度种子” becomes a seed with a heart antenna), and proposes a shot list. Once approved, it calls image_gen per anchor to create consistent 16:9 images.

Generate a sticker pack and action expressions

The team needs new WeChat stickers for a campaign. The agent loads references/media-and-variants.md and references/ip-prompt-templates.md, proposes actions and expressions based on the existing character, generates variants, and checks them against the reference before adding them to the asset library.

Style migration for a co-branding asset

A partner wants Xiaohuang redrawn in watercolor style. The agent uses the style transfer path, keeps the role-defining traits (warmth, heart antenna, fixed proportions), and delivers the visual without changing the brand identity.

Add it to your agent workflow

Install the skill with:

npx skills add ChenShuo2004/cs-skills --skill cs-xiaohuang-skill

Then instruct the agent:

Use cs-xiaohuang-skill: create a 16:9 article image for the section about “温度种子”.
First read the identity reference, then output a shot list if needed, then generate.

The skill is compatible with Codex and any agent that supports SKILL.md, so it fits into a plain agent loop.

Compare before adopting

Evaluate this skill next to other IP-preservation approaches:

• Quality signals: Does the agent validate against the reference image and a QA checklist, or only follow a text prompt?
• Maintenance freshness: Are the templates and examples kept current? A character skill decays if not refreshed.
• Alternatives: A custom image-gen agent with a hand-written style guide can work, but lacks the structured identity and prompt templates here.
• Workflow fit: The skill expects Chinese-language content and reads multiple reference files. Confirm your pipeline can supply that context.

Why it is worth tracking

This skill treats a brand character as a first-class asset instead of a prompt string. It has 118 stars and a structured reference system, which suggests community use beyond one project. If your agents produce branded visual content, try a controlled trial, especially to reuse its shot-list and QA practices. See the skill page at https://www.openagentskill.com/skills/chenshuo2004-cs-xiaohuang-skill for the full file breakdown.