yetong-cover

Yetong Original Xiaohongshu Covers | Public Package Access

This is a local fitness business access layer, not another cover method. The user has explicitly chosen to put the personal version yetong-xhs-cover into YETONG as-is: the full original text is at Original Cover Workflow, the visual specifications are at Original Visual Language, and the compositions are at Original Composition Templates. The three original texts are kept as-is, and the public call short name remains yetong-cover. Only the method text is distributed; no photos, appearance, venues, personal profiles, or private materials of Yetong or any trainer are built in.

Pre-entry Check

1. Use the user's own person-dna.md path explicitly handed over for this session, first run from this Skill's directory python3 scripts/check_profile.py <person-dna.md>. If only the path is missing, ask only for the path; do not scan private directories. The profile must be structurally valid and be active, and the identity must match; the pre-check does not prove any single fact, image authorization, or result. An activated schema_version: 2 profile from an older version can still be used; do not revisit it for missing fields due to new issues.
2. This module serves offline potential members in the city where this trainer is located. There must be a specific topic, body copy, or content summary that can clearly state the body copy's promise; if it is only "make a fitness cover," send it back for a topic. If there is only a title but the specific content is unclear, follow the original Step 0 to ask only for the content, and do not invent a story based on the title. Numbers, cities, course standards, and service frequency may only use confirmed and publicly shareable facts.
3. The six DNA groups are the basis for sourcing material per post, not another visual template: take facts from the venue features, main customer group, customer quotes, identity and delivery, publicly shareable highlights, and the person's own reasons for choosing relevant to this post; do not treat questionnaire examples as facts, and do not turn self-descriptions into member reviews. Visuals and Q&A follow the original; do not replace it with the old fitness five modes, neutral color schemes, or person-free posters.
4. A title explicitly given by the user is by default used as-is as this cover's text, and no additional "cover" label is required; only when the user explicitly states that it is only for the note publish title and separately requests cover candidates should it be treated as no cover text provided. Line breaks and top/bottom sections may be arranged, but words may not be deleted or changed. When no title is given, generate three numbered candidates per the original, handle only this cover, and do not arbitrarily change the note publish title. If there are unsupported result guarantees, false prices, medical claims, or overreaching privacy, only explain the specific issues and the evidence that needs to be verified, and stop formal production; do not draft an alternative title in the cover module or ask whether to use an alternative sentence. When the user separately requests a topic change, hand it over to the title/copy module; do not silently change the title, switch to a personal story, or use small text to debunk and bypass it.

When the profile does not exist, is invalid, or is not activated, stop personalization and prompt exactly as follows:

Person DNA is not yet activated. Currently only general methods can be called, and personalized topic selection, real-experience mapping, and personal expression calibration cannot be enabled. Complete the "Person Background Interview" and confirm the profile to unlock the full Yetong content method engine.

Only when the user explicitly skips the interview may the "general version" original process be described without personal facts; do not use another person's city, appearance, or story for their own cover, and do not activate it on the user's behalf.

Run the original version; do not run the old method in parallel

After the check passes,fully read the original cover workflow, and actually execute steps 0–6 in it. Keep the tech/product examples in the original text and do not rewrite them into fitness templates; in the local business, only fill the content slots with the user's real training topic, equipment, planner, or service scenario. When there is no physical product, the original "single-person talking-head teaching style" may be chosen; do not force-create a product.

• Ask only one current question at a time, wait for the answer before continuing, and keep the order as in the original steps 0→1→2→3→4→5→6. Even if the user first says there is no image 1, when the composition has not yet been confirmed, still give the step 1 composition choices first; only after the composition is chosen should you stop at step 2 to upload image 1. You cannot first ask for the photo and then go back to choose the composition. Choices for composition, auxiliary materials, action, background, and when no title is given all follow the original numbering. Information already provided should be recorded directly and skipped, including the step the user has already explicitly authorized the system to recommend; you cannot disguise no choice as already chosen. Once the title is given, do not ask about the title again, and do not ask further about font or title effects.
• Record the current step, chosen composition, material ledger, action, background, title as-is, and output status. When the user replies with a number or uploads an image, return to the same cover stage and do not restart topic selection or the questionnaire; when the context cannot determine what the number means, only verify which step that number corresponds to.
• Image 1 is the person reference material group of the person themselves; multiple images are recorded as image 1a/image 1b/image 1c; from image 2 onward are auxiliary materials. Map this logical numbering item by item to the tool's real attachments/paths, and do not treat the second person photo as a product. If no readable image has actually been received, do not claim to have looked at the image.
• If the person reference image is missing, stop at the original step of uploading image 1. Even if the face is not shown, the person's own hand, back view, side profile, or partial reference is still required; you cannot generate a stranger's face from archival text, nor substitute a pure information poster and pass it off as complete. Do not scan the photo album to find photos for the user.
• In a fitness context, product/auxiliary materials may be understood as actual equipment, training plans, processes, or venue materials, but keep the original options. When precise references are missing, only the generic unbranded objects already present in the original version are allowed; do not fabricate equipment models, features, logos, certificates, ratings, chats, or service evidence.
• Check venue or membership images item by item for source and use in this task. Anonymity does not equal public authorization; for body comparisons, only authorized real original images may be used, and body changes must not be generated or altered. Creative backgrounds may be chosen according to the original version, but it must be made clear that they are creative composites, and generated backgrounds must not be treated as real stores, actual past experiences, or evidence of results.
• After collection is complete, fully read the original visual language, and take only the already selected template from the original composition templates . The original version fixes 3:4, real-photo human texture, huge top and bottom text, white #FFFFFF/ warm orange-yellow #F5A623, soft shadows, and a small amount of handwritten annotation; this is the design specification of the selected template and is not claimed to be each coach's original venue brand color. Do not call the old cover-method.md or layout-patterns.md。

Prompt and image continuation

Follow the original delivery order:plan summary → complete image generation prompt → text layout checklist → generation constraints. By default, text is written into the prompt; only when the user explicitly wants a text-free base image should a safe area be left, with the exact text kept in the layout checklist. Do not call a plan with images still to be added a complete finished product, and do not say that copying text across windows has passed along reference images.

This is the only additional follow-up action in the original version, and it does not change the preceding question-and-answer: after delivering an executable complete prompt, when the current tools, reference materials, and authorization are all in place and the user has not yet explicitly requested generation, ask:

Do you need to complete the cover based on this prompt? Reply "generate directly," or choose "continue optimizing the prompt."

When the user explicitly wants only the prompt / does not want generation, do not repeatedly press the question; when there is currently no image generation capability or it is impossible to pass in the necessary reference images, explain the actual gap, retain only executable prompt optimization options, and do not promise that generation can currently be done. When the user has explicitly requested the final image in this round and the conditions are complete, follow the original version and create it directly, without repeatedly asking for the same action.

"Generate directly" authorizes only image creation for this round; it does not authorize public use of character materials or publishing notes. After rechecking this prompt and the materials, use the currently available image generation capability, and check character/equipment consistency, fingers, 3:4, Chinese characters, numbers, place names, and that the title does not obscure key parts. When the text is inaccurate, correct it with the available editing / text overlay capability; if it cannot be completed, state that it is an unfinished draft and must not be called directly publishable. "Continue optimizing the prompt" only modifies the prompt for this piece and does not generate images without authorization. When the user needs to choose a direction for adjustment, follow the original version and use one numbered question.

Do not promise plays, clicks, consultations, or transactions. Do not install, update, publish, or modify Skills in the cover task; these actions require a separate explicit request.