| name | baoyu-creative |
| description | Baoyu creative suite: article illustrations, knowledge comics, and infographics with consistent type × style × palette workflows. |
| version | 1.0.0 |
| author | Hermes Agent (adapted from 宝玉/JimLiu baoyu-skills) |
| license | MIT |
| platforms | ["linux","macos","windows"] |
| metadata | {"hermes":{"tags":["creative","image-generation","article-illustration","comic","infographic"],"category":"creative"}} |
Baoyu Creative Suite
Unified skill for all Baoyu (宝玉/JimLiu) creative generation workflows. All three modes share the same core conventions: image_generate for rendering, clarify for user confirmation, prompt-file-as-reproducibility-record, and style/type/palette dimension systems.
Shared Conventions
- Image generation: Use Hermes'
image_generate tool (prompt + aspect_ratio → URL). Download returned URLs to local PNGs via curl -fsSL "<url>" -o <absolute-path>. Always use absolute paths for -o.
- Aspect ratio mapping:
16:9 → landscape, 9:16 → portrait, 1:1 → square. Custom ratios → nearest named aspect.
- Prompt files: Write the full final prompt to a file under
prompts/ BEFORE calling image_generate. The prompt file is the reproducibility record.
- Clarify workflow: Use
clarify tool for confirmations (one question at a time). Ask the most important question first. If clarify times out, treat as default for THAT question only and continue asking remaining questions — surface defaults visibly so user can correct.
- Data integrity: Never summarize, paraphrase, or alter source statistics. Strip all credentials/secrets from source content before including in output.
- Backup rule: When regenerating, rename existing files with
-backup-YYYYMMDD-HHMMSS suffix.
- Output structure:
{type}/{topic-slug}/ with source-{slug}, analysis.md, prompts/, and generated PNGs.
Section A: Article Illustrations
Trigger: User asks to illustrate an article, add images to content, "为文章配图".
Three dimensions: Type × Style × Palette.
| Type | Best For |
|---|
infographic | Data, metrics, technical |
scene | Narratives, emotional |
flowchart | Processes, workflows |
comparison | Side-by-side, options |
framework | Models, architecture |
timeline | History, evolution |
Output: {article-dir}/imgs/ or illustrations/{topic-slug}/ with outline.md, prompts/NN-{type}-{slug}.md, and NN-{type}-{slug}.png.
Workflow: Detect refs → Analyze content → Confirm settings (clarify) → Generate outline → Generate prompts → Generate images → Finalize (insert markdown image links).
Detailed procedures: references/article-illustrator-workflow.md
Style definitions: references/article-illustrator-styles.md
Section B: Knowledge Comics
Trigger: User asks for knowledge/educational comic, biography comic, tutorial comic, "知识漫画".
Visual dimensions: Art × Tone × Layout × Aspect.
| Art Style | Tone | Layout |
|---|
| ligne-claire, manga, realistic, ink-brush, chalk, minimalist | neutral, warm, dramatic, romantic, energetic, vintage, action | standard, cinematic, dense, splash, mixed, webtoon, four-panel |
Presets: ohmsha, wuxia, shoujo, concept-story, four-panel — each with special rules beyond plain art+tone.
Output: comic/{topic-slug}/ with storyboard.md, characters/characters.md, prompts/, and page PNGs.
Key: Character consistency via text descriptions in characters/characters.md embedded inline in every page prompt. Character sheet PNG is human review artifact only.
Detailed procedures: references/comic-workflow.md
Character template: references/character-template.md
Storyboard template: references/storyboard-template.md
Section C: Infographics
Trigger: User asks for infographic, visual summary, "信息图", "可视化", "高密度信息大图".
Two dimensions: Layout (21 options) × Style (21 options). Freely combine any layout with any style.
Keyword shortcuts: Auto-select layout when user input contains 高密度信息大图 → dense-modules, 信息图 → bento-grid.
Output: infographic/{topic-slug}/ with structured-content.md, prompts/infographic.md, and infographic.png.
Workflow: Analyze → Generate structured content → Recommend combinations → Confirm (clarify) → Generate prompt → Generate image → Summary.
Detailed procedures: references/infographic-workflow.md
Layout definitions: references/layouts/
Style definitions: references/styles/
Shared References
| File | Content |
|---|
references/base-prompt.md | Base prompt template for image generation |
references/analysis-framework.md | Content analysis methodology |
references/article-illustrator-workflow.md | Article illustration detailed procedures |
references/article-illustrator-styles.md | Article illustration style gallery |
references/comic-workflow.md | Knowledge comic detailed procedures |
references/character-template.md | Character definition format |
references/storyboard-template.md | Storyboard structure |
references/infographic-workflow.md | Infographic detailed procedures |
references/layouts/ | 21 infographic layout definitions |
references/styles/ | Infographic style definitions |
Pitfalls (All Modes)
image_generate returns a URL, not a local file — always download via terminal (curl) before using.- Use absolute paths for
curl -o — persistent-shell CWD can drift between batches.
- No backend selection —
image_generate uses user-configured backend; don't write model names expecting routing.
- Prompt files are mandatory — no image generation without a saved prompt file.
- Strip secrets — scan source for API keys/tokens/credentials before any output.
- Never alter source data — "73% increase" stays "73% increase".
