// AI-generated visual presentations with full-bleed Gemini images. Triggers: mofa, mofa ppt, mofa deck, slides, 幻灯片, generative slides, build a mofa ppt, 用mofa做PPT, AI deck, 做个PPT, make slides.
AI-generated greeting cards as PNG images. Triggers: greeting card, 贺卡, mofa card, mofa 贺卡, make a card, CNY card, New Year card, 新年贺卡, ink-wash card. Generates full-bleed AI artwork via Gemini in various Chinese art styles.
Shared CLI binary for all mofa skills (slides, cards, comic, infographic, video). Not a user-facing skill.
AI-generated comic strips and illustrations. Triggers: comic, manga, xkcd, 漫画, comic strip, 四格漫画, panel comic, illustration strip, snoopy. Generates multi-panel comics via Gemini with optional Qwen-Edit refinement, stitched into a single image.
AI-generated infographics and visual posters. Triggers: infographic, poster, 信息图, 海报, data poster, visual summary, mofa infographic. Generates multi-section infographic via Gemini with optional Qwen-Edit refinement, stitched into a single tall image.
Deploy static sites to GitHub Pages or Mac Mini hosting. Triggers: mofa publish, deploy site, publish website, 发布网站, 部署网页, github pages, deploy to mini, host website, 上线, push to pages, mofa deploy, 发布到GitHub.
Multi-speaker podcast and dialogue generation with TTS voice cloning. Triggers: podcast, 播客, multi-speaker audio, 多人对话, 多人语音, radio show, talk show, 对话生成, 锵锵三人行, 两人对话, 三人对话, voice dialogue, 用XX的声音, 用XX和XX的风格, 声音模仿, 角色对话, 配音.
| name | mofa-slides |
| version | 0.7.0 |
| description | AI-generated visual presentations with full-bleed Gemini images. Triggers: mofa, mofa ppt, mofa deck, slides, 幻灯片, generative slides, build a mofa ppt, 用mofa做PPT, AI deck, 做个PPT, make slides. |
| always | true |
| requires_bins | mofa |
| requires_env | GEMINI_API_KEY |
AI presentations via mofa slides. Styles live in mofa-slides/styles/*.toml. Reference docs in docs/ are loaded on demand via the manifest's discovery.hints — read the matching doc BEFORE you author a deck or call mofa_slides.
Use RELATIVE paths. Never prefix skill-output/ yourself — the Octos host rebinds plugin output paths to <workspace>/skill-output/ automatically; a manual prefix double-prefixes and breaks delivery. Never use absolute paths like /tmp/....
Inside an Octos slides workspace (the normal case via /new slides <name>):
"out": "slides/<slug>/output/deck.pptx""slide_dir": "slides/<slug>/output/imgs"Standalone: use a unique per-request subdir like "deck-<YYYYMMDD-HHMMSS>/deck.pptx".
Inside an Octos slides workspace, do NOT answer from chat history. Two tools are authoritative:
check_background_tasks — execution state (running / verifying / delivering / completed / failed).check_workspace_contract — deliverable truth (PPTX present, manifest present, preview images present).If task state says completed but the workspace contract isn't ready, report the deck as incomplete and name the missing artifacts.
Before generating, gather: topic, style, slide count (5-8 pitch, 10-15 deck), resolution (default 2K), GEMINI_API_KEY presence. Present a slide plan for confirmation. On Telegram, use inline keyboard buttons for style picks.
mofa_list_styles before picking a style — the deployed copy may have more, fewer, or differently-named styles than any doc on disk. The binary errors out (no silent fallback) if you pick a name that isn't installed. If the user asks "有哪些模板？" / "list styles", call the tool — never recite. See docs/style-prompts.md only for inline-prompt overrides, NOT as a substitute for the live tool.prompt: (a) the instructional words that structure the brief for Gemini — 主标题, 副标题, 卡片, 底部洞察条, 概念插图 in Chinese decks vs Title, Subtitle, Card, Bottom insight bar, Concept illustration in English decks — and (b) the quoted text Gemini will render onto the image — titles, body, captions. Rule (a) is strict — DO NOT mix English structural templates (VISUAL: / TITLE: / SUBTITLE: / Elements:) with Chinese instructional voice in the same prompt. That's the #1 cause of 乱码 (see docs/cc-ppt-authoring.md anti-pattern). Rule (b) is loose — the rendered text can mix as needed: a Chinese deck's title may legitimately be "普洱茶 · Pu'er Tea", a body line may name Linux or Gemini or any proper noun, a slide may quote an English line for emphasis. Per-deck VQA should explicitly list which English proper nouns are allowed (like cc-ppt's STRICT ⑧ clause).texts) is the default for "editable PPT". Mode 3 (auto_layout VQA) is a fallback — slow, expensive, output usually needs heavy human cleanup. When the user says "可编辑PPT" / "editable slides", you want Mode 2, not Mode 3. Only enable auto_layout when the user explicitly says "auto layout" / "VQA extract" / "let the model decide", or for PDF-to-PPTX (Mode 4). See docs/modes.md.const VQA block at the top in the deck's primary language, spliced into every slides[i].prompt via ${VQA}. Write the script to disk with write_file first, then call mofa_slides with input: "<path>" — do NOT pass inline slides: [...]. The runtime no longer auto-appends a "do not render text" clamp; write that into the per-deck VQA yourself. See docs/cc-ppt-authoring.md.Manifest discovery.hints routes you to the right doc on-demand. The full set:
docs/modes.md — Mode 1/2/3/4 decision tree, when NOT to use auto_layout, costs of enabling it.docs/cc-ppt-authoring.md — required deck-script structure (top constants, const VQA, module.exports), mandatory rules, Chinese + English minimal worked examples.docs/custom-styles.md — inline prompt overrides AND full-TOML custom style authoring (where to save, schema, workflow, "style not found" failure modes).docs/worked-examples.md — full 5-slide editable business deck (Mode 2 with texts, runs, fill, margin), Mode 3 / Mode 4 / reference-image shapes, slide-canvas layout reference, TextOverlay + TextRun schemas.docs/style-prompts.md — inline style-prompt quick-reference (Art Deco, Bauhaus, Glassmorphism, 国潮, 水墨, 敦煌, 青花瓷, etc.), built-in style category cheatsheet (verify with mofa_list_styles), API modes, timing/timeout budget, CLI flags, config, PPTX editing utility scripts (ooxml/, pptx-scripts/).