agentleFS
wangge-aiwangge-ai/wangge-skills/skills/ecom-one-photo-page/SKILL.md

ecom-one-photo-page

Skill4 starsChanged 39 days agoMITView on GitHub

Turn one clear ecommerce product photo plus a short description into

The file

We split this skill into the rules it gives an agent

12 rules. About 0 are common, 8 are distinctive, and 4 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈8

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 4

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Create a useful first version from minimal input without pretending that one photo proves unseen structure or performance.

    line 13
  2. Very specific:

    If the user asks for planning, copy, page structure, or prompts, use plan mode and do not generate images.

    line 17
  3. Very specific:

    If the user explicitly asks to generate, produce, render, or make the images, use produce mode after the plan is internally coherent.

    line 18
  4. Distinctive:

    If the user provides an existing page for review, diagnose it only when the request is about the one-photo workflow.

    line 19
  5. Distinctive:

    Read the references needed for the selected work:

    line 21
  6. Distinctive:

    When the user asks where this method came from or the Skill is being revised, read references/method-provenance.md.

    line 30
  7. Distinctive:

    Do not interrogate the user for optional fields.

    line 36
  8. Distinctive:

    Run the product-consistency preflight and select a route: close-angle generation, original-pixel hybrid composition, or …

    line 50
  9. Distinctive:

    Never invent a back view, internal layer, exploded structure, texture, mechanism, accessory, specification, certification, performance …

    line 62
  10. Distinctive:

    Do not require competitor collection, reviews, keywords, or a product link for the quick workflow.

    line 67
  11. Very specific:

    Use scripts/init_job.py when a local anchor path is available.

    line 77
  12. Very specific:

    Run scripts/validate_pack.py <job-directory> before handoff.

    line 94

≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Ecom One Photo Page

Create a useful first version from minimal input without pretending that one photo proves unseen structure or performance.

Route the Request

  • If the user asks for planning, copy, page structure, or prompts, use plan mode and do not generate images.
  • If the user explicitly asks to generate, produce, render, or make the images, use produce mode after the plan is internally coherent.
  • If the user provides an existing page for review, diagnose it only when the request is about the one-photo workflow; otherwise use the relevant ecommerce review skill.

Read the references needed for the selected work:

  • For minimum intake, evidence states, selling-point choice, and low-input fallbacks, read references/decision-engine.md.
  • For the 1+5 main-image sequence and detail-page decision flow, read references/page-architecture.md.
  • For reference roles, SKU/count locking, consistency routing, and identity QA, read references/product-consistency.md.
  • For observable category cues and safe screen substitutions, read references/category-routing.md when category-specific guidance would help.
  • For claim classes and short Chinese copy, read references/copy-and-claims.md when writing page copy.
  • For writing independent generation prompts, read references/prompt-contract.md.
  • For actual image production, folder delivery, and visual QA, read references/production-and-qa.md.
  • When the user asks where this method came from or the Skill is being revised, read references/method-provenance.md.

Minimum Input

One clear product photo is enough to start. A short product description is helpful but not mandatory when the category is obvious. Platform, audience, price band, style, and verified claims are optional.

Do not interrogate the user for optional fields. Ask only when:

  1. the image does not identify the product;
  2. multiple products or SKUs make the target ambiguous;
  3. the requested result depends on an unseen structure or exact claim; or
  4. the image is too poor to preserve product identity.

Otherwise create a version-one proposal immediately and label assumptions.

Required Workflow

  1. Inspect the anchor photo and build one Product Identity Card: target SKU, intended product count, visible shape, color blocking, proportions, parts, texture, labels, accessories, fragile details, and unseen boundaries.
  2. Assign every supplied image one source role. Keep product identity authority separate from style, fact evidence, seller pages, and mixed-SKU material.
  3. Separate visible_facts, user_provided_facts, proposals, and missing_or_unverified.
  4. Run the product-consistency preflight and select a route: close-angle generation, original-pixel hybrid composition, or downgrade/request-more-views.
  5. Choose one primary purchase reason by combining a visible difference, a buyer benefit, an evidence cue, and a plausible scene. Keep secondary points subordinate.
  6. Plan the 1+5 main images and detail-page screens by buyer decision role, not by copying source layouts.
  7. Compile one independent prompt per image. Every prompt must carry the same SKU/count lock and identity constraints but a different screen job and composition.
  8. In produce mode, generate and inspect a hero plus one evidence screen before continuing the batch. Reject identity drift before assembling previews.

Non-negotiable Invariants

  • The product photo constrains what the product is; it does not constrain page composition or visual style.
  • Product identity authority, creative authority, and factual authority are separate. A style reference never becomes a product reference.
  • Lock the target SKU and intended visible product count before prompting. Do not let a reference layout silently duplicate, remove, or substitute products.
  • One screen performs one buyer-decision job. Repeated hero poses with different text do not count as different screens.
  • Never invent a back view, internal layer, exploded structure, texture, mechanism, accessory, specification, certification, performance level, or use result that the input does not support.
  • Replace an unsupported structure or proof screen with a visible-feature, usage-logic, scale, care, or scene screen.
  • Claims from the user may be used as user_provided_facts; claims merely visible in third-party marketing images remain unverified until the user confirms them.
  • Existing seller or competitor pages may provide facts or category context, but their composition, slogans, brand assets, and visual system are not generation references.
  • A prompt can request consistency but cannot prove it. Fine packaging text, logos, product count, and fragile geometry remain output-level QA items; use original-pixel composition when those details are business-critical.
  • Do not require competitor collection, reviews, keywords, or a product link for the quick workflow.

Default Delivery

Unless the user chooses another location, create the job under:

./outputs/ecommerce\one-photo-page\<product>_<YYYY-MM-DD>\

Use scripts/init_job.py when a local anchor path is available. Deliver:

input/
strategy/product-card.md
strategy/page-map.md
prompts/main/
prompts/detail/
renders/main/          # produce mode only
renders/detail/        # produce mode only
preview/               # produce mode only
qa/qa-report.md
job.json

For a standard full pack, create six main-image prompts and six detail-page prompts. Honor a shorter user-requested pack.

Run scripts/validate_pack.py <job-directory> before handoff. A valid folder structure does not replace visual QA for generated images.

Completion Boundary

State whether the delivery is plan only or produced and visually checked. Do not describe prompts as completed images. Report any screen that was downgraded because one-photo evidence was insufficient.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.