SuperAPI AI Directory
Partner with us中文AI newsFavorites
Skillsdesignkit-ecommerce-product-kit

designkit-ecommerce-product-kit

Design & creation

View repository

Updated 10-16 00:23

About this skill

DesignKit Ecommerce Product Kit — Ecommerce Product Kit

Intended forconversation: guide the user through information collection and deliver the kit results.

Step-by-step output (mandatory, merging the two steps is prohibited)

  • In the same round of replies, only one collection step may be advanced: either ask only about selling points + style preference, or ask only about listing configuration,prohibitedfor both "core selling points/style preference" and "platform/country/language/ratio" to appear in the same message at the same time.
  • Correct rhythm: first sendthe question about selling points + style preference→ stop, wait for the user's reply → after the user replies, then sendlisting configuration onlyexplanation → stop, wait for the user's reply → then proceed toviral style(prompt for selection or allow skipping) → stop, wait for the user's reply → then generate images / call the API.
  • If the user proactively writes all of the selling points, style preference and configuration in one message, the parameters may be adopted all at once, butyour reply should still confirm them item by item。
  • Fast track: if the user has already attached listing configuration information when confirming the selling points (e.g. "OK, Amazon US English 1:1"), it may be adopted directly, skipping the second round of configuration questions, confirming in one sentence and then going straight to style generation.

Workflow (in order)

  1. Product image: if there is no image, prompt to upload or provide a URL/path.
  2. Core selling points + style preference (round 1, occupying one message on its own): after obtaining the product image, based on the product imageautomatically generatea set of core selling point suggestions to show the user, and then:
    • Ask: "You can adopt it directly, modify it, or enter your own selling point description"
    • At the same time ask about style preference: "Do you have a particular visual style direction in mind? If not, I will recommend one intelligently for you"
    • This message must not mention listing configuration information such as platform, country, language, ratio, etc.
    • After the user confirms adoption, gives modification suggestions, or enters their own, use theversion finally confirmed by the useras the product_info
    • style preference mentioned by the user, record it, and when showing the style options returned by the API, matching items may be recommended first
    • Only when the AI cannot infer meaningful selling points from the image should it directly ask the user to enter them
  3. Listing configuration (round 2, occupying one message on its own): only after the user has replied about the selling points (or confirmed skipping), then sendonly configuration-relatedfriendly explanation (a short paragraph + brief list is fine, no need to interrogate item by item like a form):
    • Platform: Amazon, Taobao, Temu, Pinduoduo, TikTok Shop...
    • Country/Region: United States, China, Russia, Japan, Spain…
    • Language: English, Chinese, Russian, Japanese, Spanish…
    • Aspect ratio：1:1、3:4、9:16、16:9 Default: If the user does not specify, generate according to Amazon / United States / English / 1:1 . When asking,you must clearly inform them of this set of default values, so the user knows that replying directly with "use defaults" or "okay" will skip this step. Parameter mapping: The user selects in Chinese; before calling the API, it must be converted to API parameter values: | Platform | API value | Country | API value | Language | API value | |------|--------|------|--------|------|--------| | Amazon | amazon | United States | US | English | English | | Taobao | taobao | China | CN | Chinese | Chinese | | Temu | temu | Russia | RU | Russian | Russian | | Pinduoduo | pinduoduo | Japan | JP | Japanese | Japanese | | TikTok Shop | tiktok_shop | Spain | ES | Spanish | Spanish | Rules: The usernot filling in, skipping, or filling in only partis all acceptable,do not repeatedly ask follow-up questions to complete the configuration; fill in missing values with the above defaults.Do not give another lengthy explanation of selling points in this message. Before executing the API call, useone sentenceto restate the actual configuration used (user-provided vs default). Fast track: If the user already included configuration information in the previous step, skip this step.
  4. Viral style (triggered as needed, not a mandatory step)
    • Trigger condition: Only when the user in round 1 (selling points + style preference)explicitly specifies a specific style direction(such as "tech feel", "minimalist", "I want to see style options"), enter this step to display style options for the user to choose from.
    • Skip by default: If the user does not mention style, replies "none", ignores the style question, or gives a vague reply (such as "as long as it looks good", "whatever", "your recommendation"), allskip this step and go directly to image generation, and the server will automatically match the style.
    • When entering this step, call the style API and display the options:
      1. bash __SKILL_DIR__/scripts/run_ecommerce_kit.sh style_create --input-json '<含 image、product_info、platform、market 等>'
      2. Use the returned task_id polling:bash __SKILL_DIR__/scripts/run_ecommerce_kit.sh style_poll --input-json '{"task_id":"..."}'(can include as needed max_wait_sec / interval_sec）
    • Prohibitedto make up styles on your own; only display the options returned by the API. Ask the userto select one of them, and after selection you must usethe complete structure of that set returned by the API(such as the brand_style object), and must not hand-write a replacement.
    • After the user sees the options and chooses "skip", "automatic", "you decide", etc.:render_submit do not pass brand_style, and the server will automatically match the style.
  5. Image generation + automatic download：
    • Call render_submit / render_poll to generate the finished image (without brand_style omit this field, the server automatically selects the style)
    • render_submit ratio parameter, priority is given to reading aspect_ratio, while also compatible with ratio
    • During polling, based on the stderr output of [PROGRESS] report progress information to the user (such as "completed 3/7 images")
    • After completionautomatically downloadall finished images to the working directory, naming rule:{商品名}_{序号}_{label}.jpg(label is taken from the items[].label returned by the API)
    • Display the results in Markdown image format, and inform the file save location
    • Rerun with a different style: if the user wants to switch to another style after viewing the finished images, then call style_create to get style options for the user to choose from, and after selection jump to the render_submit step to regenerate the images.

Output directory rules

render_poll After completion, the generated results are automatically downloaded, and the output directory is resolved in the following priority order:

  1. input-json explicitly passed in output_dir
  2. environment variable DESIGNKIT_OUTPUT_DIR
  3. when the current working directory contains openclaw.yaml , use ./output/
  4. if {OPENCLAW_HOME}/workspace/visual(or ~/.openclaw/workspace/visual) exists, use the output/designkit-ecommerce-product-kit/
  5. otherwise fall back to ~/Downloads/

Constraints:

  • output directorymust not be located inside the current skill repository
  • The script will automatically mkdir -p
  • render_poll the JSON output will additionally include output_dir and local_paths

Hard ordering constraints: product images → selling points + style preferences (one round) → user reply → listing configuration (one round, or skipped via fast track) → user reply → viral style (entered only when the user expresses interest in a style, otherwise skipped) → finished images + automatic download。