跳至正文
OpenAgentSkill
浏览已开放 · 目录记录日期 2026-11-21 · 自动更新和账户服务暂时关闭。
图像与绘图

Poster

User wants a 海报, flyer, campaign visual, event one-pager, hiring poster, webinar poster, comic-as-visual, or a single designed poster they will click-edit in Paw Work Design (tldraw). Do NOT trigger for a multi-slide PPT/deck (slides) or a clickable 官网 (html-site).

Player-YN · 已有 AI 审核记录

安装与使用

运行命令前先阅读来源与依赖说明。复制命令不会自动安装或执行。

npx skills add Player-YN/PawWork_ZhuaZhua --skill Poster

技能说明


name: Poster description: User wants a 海报, flyer, campaign visual, event one-pager, hiring poster, webinar poster, comic-as-visual, or a single designed poster they will click-edit in Paw Work Design (tldraw). Do NOT trigger for a multi-slide PPT/deck (slides) or a clickable 官网 (html-site).

Design poster playbook

Outcome is a Paw Work Design canvas (tldraw pawCanvas), not HTML, not a blog, not a deck of slides, not one fused PNG. Never write a .html poster (USE_CANVAS).

Classify each selection as plate (成片 — attach as-is), reference (style cue; acquire image only for that one no-text plate), or evidence (facts — do not paste as decoration). Clarify once if a reference is ambiguous.

Semantic compile

One poster layoutId + one themeId → one run createScene (kind=poster) with frames: [{id, layoutId, slots}]. No raw box / nodes on the normal path. Runtime owns geometry. One dominant hierarchy and one visual anchor. If content will not fit, cut copy or pick another layout.

Themes: hanbai, ink-rose, midnight-cyan, forest, studio-amber, editorial, cobalt, mono.

Poster layouts: poster-hero, poster-split, poster-event, poster-quote, poster-product, poster-editorial, poster-data, comic-panel.

Visuals, in this order: selected / user asset first → packaged icon → native motif → native chart (real numerical data only; never invent statistics) → acquire action=image no-text. After acquire action=image, the next canvas write must attach the returned path as the visual slot. Never bake lettering into a generated image.

Unknown icon: deck act=read catalog="icons" query="…" limit=8. Motifs / charts: catalog="motifs" / catalog="charts". Generated plate: deck act=read catalog="image-brief" with layoutId, themeId, subject → acquire with the returned prompt / aspect → attach path. Compile does not generate images.

Comic / 自我介绍长图: one panel story per Frame (comic-panel). Paper count and size are your judgment. Raw nodes / box only as last-resort polish after a semantic plate exists.

Other routes

  • Live page / 这页 / this site with focusPage: acquire fetch then fromPage.
  • Selection fragments: fromSelection
  • Flatten screenshot remake: visual-compile (and remake-poster when they asked 变成可编辑海报). Photo/UI flattens use fromRaster scan: "auto".
  • A clickable 官网 is skill html-site. A fused PNG is compose-image.

fromPage extracts content into engine nodes — it is not CSS-faithful HTML.

QA and mutate

On CANVAS_QA_FAILED, repair slots / change layout / split content and resubmit to the same artifact. Never a second poster file.

After the poster exists, drive the canvas with the deck tool (act=read|write|export). Omit nodeId — host pins selection. After run writes slots JSON, deck write with that path — do not retype.

  • setSlotText / setSlotSrc — clicked node
  • Semantic replacePlate {plateId|frameId, layoutId, themeId?, slots} — rewrite one frame, keep identity

Export PNG / SVG / PDF from the Design tab (current Frame), not by screenshotting chrome.