paper-operators
Generate or plan publish-grade 16:9 article illustrations using folded-paper action characters, with Chinese-first article/body-image workflows. Use when a user wants Chinese or English article illustrations, blog figures, explainer visuals, social/article images, or concept images where a small paper person should physically perform the core idea. Especially useful for turning abstract claims, emotions, cultural ideas, art criticism, product thinking, systems, workflows, tensions, boundaries, transformations, relationships, and evidence into clear high-angle paper-model scenes, as single figures or chained multi-image series. Do not use for generic mascots, cute stickers, Xiaohei-style black figures, stock PPT infographics, or image requests where no action character helps comprehension.
Alexsun1one · 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 Alexsun1one/paper-operators --skill paper-operators技能说明
name: paper-operators description: Generate or plan publish-grade 16:9 article illustrations using folded-paper action characters, with Chinese-first article/body-image workflows. Use when a user wants Chinese or English article illustrations, blog figures, explainer visuals, social/article images, or concept images where a small paper person should physically perform the core idea. Especially useful for turning abstract claims, emotions, cultural ideas, art criticism, product thinking, systems, workflows, tensions, boundaries, transformations, relationships, and evidence into clear high-angle paper-model scenes, as single figures or chained multi-image series. Do not use for generic mascots, cute stickers, Xiaohei-style black figures, stock PPT infographics, or image requests where no action character helps comprehension.
Paper Operators / 纸片人
Paper Operators turns an article idea into a tactile paper-stage illustration where a small folded-paper person performs the core conceptual action.
默认面向中文文章配图;用户使用英文时再切换英文输出。Use this skill to create a consistent article-illustration style: paper people, physical metaphors, readable labels, clear reader paths, and domain-aware tools. The character is never decoration. If removing the paper operator does not weaken the image, do not use one.
Workflow
- Read the article, excerpt, topic, or image-edit request.
- Identify one source anchor: the sentence, paragraph, claim, tension, turn, or feeling that deserves a visual.
- Read the user's real intent with
references/intent-reading.md: who reads it, what they should feel, the author's stance, and the content-specific particulars that must appear. The image must fit THIS document, not the topic in general. - Route the asset with
references/asset-routing-and-truth.md: decide whether this is an article figure, story/book card, data story scene, center illustration for a card/slide, or reference-informed explainer. Record final container, display ratio, text ownership, exact truth constraints, and whether another tool should own the job. - For stories, books, biographies, cases, or narrative explainers, read
references/story-card-grammar.mdand identify protagonist, conflict, turning point, choice, consequence, and transformation. - For metrics, charts, rankings, benchmark results, project reports, or evidence summaries, read
references/data-story-scenes.mdand write the exact data contract before prompting. - For scientific, historical, cultural, geographic, brand, model, artifact, species, or apparatus topics, read
references/reference-informed-explainers.mdand gather stable visual/factual cues before prompting when accuracy matters. - Name the core relationship the image must show, using
references/relationship-grammar.md: connection, sequence, dependency, causality, feedback, contrast, tradeoff, hierarchy, transformation, boundary, divergence, or tension. Choose the relationship before the operator; precision comes from rendering the exact relation, direction, condition, and state. - Choose the domain and mood: art, culture, personal essay, business, product, engineering, education, psychology, finance, history, lifestyle, or another field.
- Choose the core action: pull, frame, reveal, block, tune, fold, sort, mend, weigh, arrange, carry, shelter, distill, balance, light, archive, or transform.
- Decide whether a Paper Operator is necessary using the inclusion test.
- Before locking the default metaphor, diverge with
references/creative-divergence.md: sketch a few candidate worlds and converge on the one that fits this content most precisely — do not reflexively reuse route-table, gallery, or room. - Select one operator family from
references/operator-library.mdthat can physically perform the relationship; use the Operator × Relationship map when unsure. - Choose the composition mode with
references/composition-modes.md: one dense single-image-multi-beat composition (e.g. four river crossings on one terrain) vs a multi-image series — pick what the content's own structure demands, not a default. - Assemble the scene from the kit in
references/primitives.md, and encode status, degree, and state precisely withreferences/state-coding.md. Reuse the same primitive for the same meaning so the image stays consistent. - Adapt the world, props, palette, texture, and typography using
references/domain-adaptation.md. - Write readable labels with
references/text-strategy.md. For simple figures, use 3-6 short labels; for story/book cards use 4-8 beat labels; for data scenes, include exact values only when readable. Prefer short labels, and move headline/explanation/citations outside the image when the outer layout should own them. - For exact labels, data, chart values, scientific parts, historical cues, brand cues, or proper names, put exact truth constraints into the final prompt. Do not let the image model invent numbers, categories, dates, axes, or reference-specific details.
- For final artifacts that will enter README, social cards, slides, docs, or article packages, read
references/layout-handoff.mdand return caption, crop guidance, safe margin, and handoff owner. - Build the final image prompt using
references/prompt-template.md. - Generate or edit one image at a time.
- When the article needs more than one image, plan the set with
references/series-and-chaining.mdso the figures share a world, operator, palette, and a progressing throughline instead of reading as N unrelated pictures. - For public examples or reusable packages, record the source anchor, asset role, truth constraints, reference cues, final prompt, output path, rejected attempt reason, and QA result using
references/prompt-records.md. - Run QA with
references/qa-checklist.md, include the asset-routing checks fromreferences/asset-routing-and-truth.md, the failure scan fromreferences/failure-patterns.md, and the Swap Test from `references