WeChat Cover Design

Create one editorial WeChat cover from the article's actual claim. This skill is optimized for the biopharma workflow: mechanism explanation, delivery technology, clinical evidence, platform analysis, molecular design, industry landscape, and CMC/manufacturing.

Execution rule

Before drafting the final answer, explicitly ask the user to choose an execution path. Do not infer the path from host capability.

If the user has not selected a path, ask only which they want: direct image generation or prompt-only. Never claim an image was generated unless the host returned an image artifact.

WorkBuddy runtime

• Prefer WorkBuddy's built-in image-generation tool for the direct-image-generation path. Do not request an external image API key or send the article, prompt, or generated assets to an independently configured image provider.
• Resolve every bundled path relative to this skill directory. Do not assume the user's workspace contains a clone of the source repository.
• Before running the Python compositor or image audit, verify that Python and Pillow are available. Do not install packages without the user's approval.
• If Pillow or local script execution is unavailable, use a deterministic host image editor when it can preserve the exact title and coordinates. Otherwise return the text-free background and overlay specification as an explicit handoff; do not present it as a finished cover.

Input and visual brief

Extract only what is needed to make a cover specific to the article:

If the article does not provide enough information for a factual visual, ask up to three focused questions. Do not invent experimental results, clinical claims, company facts, or medical efficacy.

Select one of the 13 themes

Read references/theme_registry.md, references/typography_layout.md, and the matching section in references/theme_specs.md before composing the prompt. Theme IDs are fixed at 1-13; do not refer to or restore any retired themes.

Honor an explicit theme choice unless it contradicts a factual constraint. When two themes are genuinely equally suitable, present two one-sentence options and ask the user to choose.

Fixed title composition

All covers use a 1584x672 canvas (about 2.35:1). The title field is always on the left: x=76..620, y=88..585 (roughly the left 39% of the canvas). Keep the right 60% for the focal biomedical scene.

• Use a natural quiet field, not a translucent card, underline, divider, shadow, or gradient panel.
• Reserve the left field before generating the background. Keep it low-detail, clear of focal objects, high-contrast edges, labels, and data marks.
• Use the selected theme's palette. Prefer a host-available Chinese serif or editorial display face and a compatible Latin display face. When using scripts/compose_cover.py, the user must provide local font paths explicitly.
• Keep the hierarchy to eyebrow, main title, optional subtitle, and one short footer at most.
• Use the fixed title group: one-line eyebrow; optional title prefix; one or two semantic main-title lines; up to two subtitle lines; one-line footer. A title prefix such as Ab-mRNA-LNP stays in the title group, never in the eyebrow or footer.
• Main title default is 62 px, with a 52 px minimum. The compositor measures each line against the 500 px effective width and rejects overflow; it never moves title text toward the scientific subject.
• If the source artwork has no usable left quiet field, regenerate the background. Do not move the title toward the scientific subject.

Direct image-generation path

Use this path only after the user selects direct image generation. Write a text-free image prompt that explicitly reserves the fixed left title field. Generate the artwork with the host's native image tool. If the provider only supports 16:9, generate at 16:9, center-crop to 2.35:1, then audit the cropped result. Run scripts/validate_cover.py --input <background.png> --theme <1-13> --json before any type is applied. A failed audit means regenerate the background. The audit rejects high-contrast long lines, charts, process paths, labels, or any large structured mark in the title field. Do not describe the image as audited unless the command returned "status": "PASS"; phrases such as “审计一次过” are not validation evidence. Only then render the supplied exact title in the field using host layout/editing or scripts/compose_cover.py.

Handoff path

Use this path only after the user selects prompt-only. Provide one self-contained English prompt that contains all of the following:

1. Target dimensions: 1584x672 or 2.35:1.
2. The selected theme's visual scene, palette, materials, focal object, and right-side placement.
3. A left-side low-detail title field occupying 39% of the width.
4. An explicit text-free background instruction, with no readable text, labels, logo, or watermark anywhere in the generated image.
5. A negative prompt preventing objects, high-contrast edges, labels, and data marks from entering the title field.

Then append a separate Overlay specification containing the exact title, subtitle, coordinates, fonts, and colors. Never ask the image model to render these strings.

Deliverables

When an image was generated

Return:

1. The image artifact.
2. Theme and one-sentence rationale.
3. A concise Chinese production note: title treatment, crop safety, and any text-rendering limitation.
4. The validation result.
5. The raw JSON result from scripts/validate_cover.py --json.

When a handoff is required

Return:

1. Theme and one-sentence rationale.
2. A complete, copy-ready English prompt for a text-free background.
3. Overlay specification: title, subtitle, x=76..620, y=88..585, Chinese font, English font, palette, and left alignment.
4. A concise Chinese production note and the validation result.

Prompt package format

Use this package structure for both paths. It is also the required input to scripts/validate_prompt.py --all; replace every quoted value with article-specific content.

Cover plan:
Theme: Theme 6 - Swiss poster
Generation path: user-confirmed direct-image-generation / prompt-only

Alignment Record:
Article thesis: "..."
Reader takeaway: "..."
Content class: mechanism / evidence / transaction / industry_landscape / other registered class
Theme rationale: "..."
Visual metaphor: "..."
Element provenance: "focal object: article section ...; supporting object: reader takeaway ..."
Forbidden substitutions: "article-irrelevant visual shortcuts only; do not repeat image-quality negative terms"

English image prompt:
Create a text-free 16:9 or 1584x672 (2.35:1) WeChat cover in [selected theme] style. Place [focal scientific scene] in the right 60% of the canvas. Reserve left x=4.8%-39.1%, y=13.1%-87.1% as a low-detail, theme-native title field. Do not render readable text, labels, logos, or watermarks anywhere in the image.
Text-rendering strategy: audit the background, then overlay exact supplied text with deterministic layout.
Negative prompt: watermark, logo, readable text, gibberish, crowded composition, focal objects, high-contrast edges, labels, or data marks in the title field.

Chinese production note:
...

Validation

Before delivery, confirm that the final prompt has no unresolved placeholders, states the 2.35:1 format, includes a negative prompt, names the theme's visual language, and preserves the left title field.

• Run scripts/validate_prompt.py --all for every text-free background prompt.
• Run scripts/validate_cover.py --input <background.png> --theme <1-13> --json after generation and before title overlay. If it fails, regenerate the background. There is no bypass flag.
• If scripts cannot run, apply the equivalent checklist manually and state that fact.

References

• references/theme_registry.md: theme names and article-signal mapping.
• references/theme_specs.md: detailed visual prompt skeletons.
• references/typography_layout.md: fixed coordinates, palette, and font-selection rules.
• references/title_overlay_policy.md: native-image and prompt-only title rules.
• scripts/compose_cover.py: deterministic title compositor.
• scripts/validate_cover.py: image-level quiet-field and contrast quality gate.
• scripts/validate_prompt.py: prompt validator.