A set of images in a unified style + a caption for each — ideation, image order, image generation, image review.
Suite description · This skill is part of
aws-wechat-article-*a one-stop suite (9 slugs total, entry pointaws-wechat-article-main). Cross-skill relative references depend on the sameskills/directory; it is recommended to install the wholeclawhub installsuite together. Source code:https://github.com/aiworkskills/wechat-article-skills
Capabilities disclosure
This skill orchestrates the image post workflow and calls the same suite's image_create.py(images) for image generation; it may optionally call publish.py to publish.It will send image prompts to an external image API; if the publishing path is used, it will upload the images as the POST body to the WeChat API.
- Credential reading:
aws.envofIMAGE_MODEL_API_KEY; when publishing, it additionally readsWECHAT_{N}_APPID/WECHAT_{N}_APPSECRET - Credential egress:
IMAGE_MODEL_API_KEYtoAuthorization: Bearersent toimage_model.base_url; WeChat credentials are used to exchangeaccess_token - Content egress: image prompts are sent to the image API; image files are sent to WeChat
material/add_material/draft/add - File read:
.aws-article/config.yaml, this piece'sarticle.yaml、imgs/prompts/*.md - File write: this piece's
imgs/*.{png,webp}、imgs/outline.md、article.yaml - shell:
{python} {baseDir}/../aws-wechat-article-images/scripts/image_create.py; optional../aws-wechat-article-publish/scripts/publish.py
At minimum, install both aws-wechat-article-images(image generation capability is there); to use publishing, also install aws-wechat-article-publish。
Routing
Long-form article (title + body + illustrations + backend publishing) → aws-wechat-article-main; single illustration within a long-form article → aws-wechat-article-images. This skill doesimage-centriccontent: multiple images + a caption for each, with a unified style for the whole set.
Output goals (read this first)
- Input: topic / topic card / user material images (choose any one)
- Output:
imgs/(outline.md+prompts/+ images), unified style for the whole set, no drift - Publishing: ⛔ WeChat native "image messages" arenot implementedin this suite. Either assemble the whole set of images into an article and go through the regular pipeline, or deliver only the images for the user to send manually in the backend — see branches.md "One"
Configuration check ⛔
Before any operation, first follow First-time guidance · Detection order and continue only after it passes (or the user has explicitly confirmed in writing "no check this time").
Image model:image_model(provider、base_url、model) is in .aws-article/config.yaml;IMAGE_MODEL_API_KEY at the repository root aws.env. Key name mapping env.example.yaml。
Must comply with main'sagent behavior constraints: when environment validation has not passed and the user has not explicitly granted a "one-time exception",must not pretend to have used the dedicated image generation API。
Workflow
贴图进度:
- [ ] 第1步:⛔ 环境检查 + 合并本篇约束
- [ ] 第2步:确定选题
- [ ] 第3步:确定风格(全组统一)
- [ ] 第4步:规划图序 → imgs/outline.md
- [ ] 第5步:⛔ 展示方案并等用户确认
- [ ] 第6步:生成图片
- [ ] 第7步:审图
- [ ] 第8步:交付(拼图文发 或 交给用户后台手发)
Step 1: Merge this piece's constraints
When this piece is under drafts/…/ , read and merge according to .aws-article/config.yaml → this piece's article.yaml (for the same key, this piece takes precedence). Key fields:multi_image_count、tone、target_reader、custom_sticker_style > default_sticker_style。
When there is no valid YAML, use the user's spoken topic + the skill defaults.
Step 2: Determine the topic
Image cards produced by topics / the user directly provides a theme / the user provides source images.
Step 3: Determine the style
For the loading priority and the classification of the nine forms (5 information slots / 3 rhythm slots), see branches.md "III"。
The whole set of images uses one unified form — the value of a set of images lies in its consistency; changing one for each image is the same as having no style.
Step 4: Plan the image sequence
Produce imgs/outline.md: the purpose, caption, and filename of each image. For the template, see references/workflow.md; for sticker types and narrative lines, see branches.md "IV"。
Number of images ≤ 9(WeChat limit).
Step 5: Present the plan and wait for user confirmation ⛔
Present the entire outline.md table to the user,wait for confirmation before generating images— generating images costs money; if the plan is wrong, every image must be redone.
Step 6: Generate images
Prefer calling image_create.py(depends on image_model + IMAGE_MODEL_API_KEY); only when the environment is not ready or the user accepts main's "exception for this time" should it be downgraded to Agent image generation / outputting prompts only.
You must tell the user which one is currently being used:
ℹ️ 使用 image_create.py 调用专用生图模型({model})ℹ️ 本次未走 image_create.py(原因:…)
Inrepository rootexecute (adjust the path according to this imgs/ ):
# 整批
{python} {baseDir}/../aws-wechat-article-images/scripts/image_create.py batch drafts/YYYYMMDD-slug/imgs/prompts/ -o drafts/YYYYMMDD-slug/imgs/
# 单张
{python} {baseDir}/../aws-wechat-article-images/scripts/image_create.py generate imgs/prompts/01.md -o imgs/01.png
# 连通性自检
{python} {baseDir}/../aws-wechat-article-images/scripts/image_create.py test
The rules for text within images and prompt construction are completely identical to those for long-form accompanying images, see images SKILL and prompt-construction.md. For the four classifications of call failure handling, see branches.md "II"。
Step 7: Review images
Checklist dedicated to stickers:references/checklist.md. The key point isviewing the whole set: whether the style is unified, whether it is still distinguishable under the first thumbnail, and whether the image sequence reads smoothly.
Step 8: Deliver
⛔ Do not run publish.py full directly on the sticker directory — it hard-requires article.yaml + article.html + cover.*, and the sticker directory does not have these three things; running it will inevitably report an error.
For the two implementation paths (assemble into an image-text post and publish automatically / deliver only the images for the user to publish manually in the backend), see branches.md "I". When taking the second path,do not claim that the publishing loop is complete。
Branches and detailed rules
For the real boundaries of publishing paths, the four classifications of image generation failure, style and form priority, sticker types and narrative lines, what "nine-grid" refers to, and standalone installation → references/branches.md。
Process files
| read | produce |
|---|---|
topic-card.md(optional),.aws-article/config.yaml + this article.yaml |
imgs/outline.md、imgs/prompts/*.md、imgs/*.png |