Doubao Listing Localization

Cross-Border Listing And Product Optimization

Scope

This skill covers two tightly connected work modes:

1. Listing creation and localization. Turn product facts, supplier titles, screenshots, competitor pages, or Chinese copy into platform-ready titles, bullets, descriptions, A+ content, tags, attributes, backend search terms, feed fields, and image briefs.
2. Product optimization diagnosis. Diagnose and improve the sellability of a product page or item setup across PDP content, main image, attributes, variants, price display, coupons/promotions, Buy Box/Featured Offer readiness, inventory-visible issues, listing quality, and product suppression symptoms.

Treat Listing SEO as one part of Product Optimization. Do not split these tasks into separate skills. When a diagnosis shows that content fields need rewriting, complete the rewrite inside this skill.

Core Workflow

1. Classify the product task. Identify whether the user needs listing copy, keyword research, localization, PDP audit, main image/image brief, attribute/category setup, variant/size-chart fix, price/offer diagnosis, conversion/click diagnosis, suppression/unpublished triage, or a combined product optimization plan.
2. Extract platform and product context. Pull out platform, target country/site, language, product/category, brand, specs, material, size, color, pack quantity, variants, fulfillment mode, price/promo state, issue symptom, objective, links, screenshots, exports, competitor examples, and requested output fields.
3. Choose evidence paths before writing. Read references/source-map.md for listing evidence and references/product-optimization-source-map.md for official product/page/offer evidence. Decide whether to use Seller Center, Seller University, Marketplace Learn, policy center, official help, API docs, public PDP/search pages, user files, competitor pages, ads reports, or keyword tools.
4. Check official product requirements. For product-page diagnosis or platform-specific claims, read references/product-policy-checklists.md. Official platform requirements and Seller Center notices beat competitor behavior and generic ecommerce experience.
5. Extract product facts and current state. Separate verified facts from assumptions. Build a table for product identity, title, images, bullets, description, A+ content, attributes, category, variants, size chart, price, coupons, rating/reviews, stock/fulfillment-visible state, offer status, traffic/conversion metrics, and unresolved backend-only fields. Record evidence scope as product family, parent, child SKU, or tested sample; do not silently generalize narrower evidence.
6. Build the keyword and positioning map. Read references/keyword-taxonomy.md. Create core keywords, attribute keywords, use-case terms, audience terms, occasion/gift terms, long-tail terms, backend/tags, and blocked or risky words. Avoid keyword stuffing.
7. Benchmark competitors when available. Extract 3-5 direct competitors or category leaders across title, image, price, offer, rating/reviews, bullets, A+ structure, attributes, variant setup, and review pain points. Use competitors for gap discovery, not as policy authority or copy source.
8. Diagnose root causes when optimizing an existing product. Read references/product-diagnosis-playbooks.md. Separate likely causes into traffic, click, trust, price, content, image, policy, inventory/fulfillment, category/attributes, offer, and data/reporting issues.
9. Generate platform-ready improvements. Read references/platform-listing-rules.md, references/output-templates.md, and references/product-optimization-output-templates.md. Produce the requested copy, field table, diagnosis table, image brief, and prioritized optimization plan. Complete every requested field or mark it blocked: missing evidence; do not omit weakly supported sections without saying so.
10. Localize beyond translation. Adapt buyer language, local category names, units, sizing, occasions, gift framing, benefit order, marketplace style, and claim boundaries for the target country/site.
11. Run risk and handoff checks. Read references/risk-and-handoff.md. Flag brand/IP terms, certification/medical claims, prohibited claims, price/promo/contact information, unsupported live data, policy uncertainty, and backend-only assumptions.
12. Give validation steps. Recommend the smallest next verification: check Seller Center notice, pull listing quality report, review Brand Analytics/Search Query Performance, compare PDP/search screenshots, test title/image, monitor CTR/CVR/sessions/orders, or gather missing backend exports.

Mandatory Quality Gates

Run all applicable gates before presenting final copy or a completed diagnosis.

1. Evidence And Claim Gate

• Create a claim ledger for material, dimensions, weight, packaging, compatibility, certification/test results, safety, comfort, durability, performance, audience, and use-case claims.
• For each claim, record its source and scope: product family, parent, child SKU, model/year, or tested sample.
• Use a claim in shared parent copy only when the evidence covers every included child. Keep sample-specific test results and child-specific facts out of shared copy.
• Delete or qualify claims such as comfortable all day, secure, safe, compatible, or unsupported use scenes unless direct evidence supports the exact wording.
• Treat review pain points as research signals, not product facts. Do not turn them into affirmative claims without product evidence.
• Maintain a verified, conditional, and blocked claim whitelist. Final buyer-facing copy may use only verified claims; place conditional claims in validation notes.

2. Parent, Child, And Multi-Model Gate

• Distinguish the current selected child, text visible on the current page, actual default landing child, and seller-configured default. Do not infer one from another.
• Determine for each field whether the platform stores it at parent, child, or shared family level. Check title, bullets, description, images, dimensions, weight, compatibility, and backend terms separately.
• When child-level maintenance is available, write or map child-specific copy and assets. When it is unavailable or unverified, use neutral family-level wording or explicitly name all supported variants.
• Never anchor shared copy or images to one child's dimensions, weight, package contents, test result, or compatibility.
• Do not merge nearby models, model years, item codes, standards, or compatibility rules from a product name alone. Build a SKU/model matrix and mark unresolved differences.

3. Listing Completeness And Search-Term Gate

• Check the user's requested deliverables one by one. For a full Amazon listing, include title, 3-5 bullets or the category-allowed count, description, backend search terms, keyword layering, image sequence/brief, and risk notes unless the user narrows scope.
• Keep backend terms relevant and evidence-supported. Exclude brand/trademark terms without authorization, repeated visible-title terms, subjective claims, punctuation where disallowed, and speculative audiences, body locations, or use cases.
• Validate exact platform limits using the current official rule when limits matter. For byte-limited fields, calculate encoded bytes from the final text, state the encoding/method, and do not substitute a character count.

4. Final Consistency Gate

• Re-read the final copy, tables, attachments, and summary as one deliverable.
• Confirm variant facts, claim scope, field limits, SKU count, recommendations, and status labels agree everywhere.
• Remove contradictory actions such as "leave unchanged" in one section and "replace now" in another.
• Never claim that a file, formula check, platform validation, or upload-ready output is complete unless it was actually created and checked.

5. Mixed-Task Boundary Gate

• Identify when the request also requires PPC/bidding, inventory/replenishment, procurement allocation, margin/forecast modeling, logistics, tax, or legal eligibility.
• Complete the listing-side work that is supported, then explicitly route each remaining workstream to the appropriate skill or agent. State that the overall task is not complete until those workstreams return.
• Do not improvise spreadsheet formulas, replenishment decisions, ad budgets, or financial conclusions inside this skill and present them as validated business outputs.

Input Rules

• Platform and target country/site are the highest-value inputs. If missing and the answer depends on platform rules or local language, ask a concise question. If proceeding asynchronously, default to Amazon US only when the query implies Amazon; otherwise use global marketplace and mark assumptions.
• Product/category is required for final listing copy. If missing, give an input checklist and template rather than inventing product details.
• For product diagnosis, ask for the product link, screenshot, item ID/ASIN/SKU, visible symptom, Seller Center notice, or export only when that evidence materially changes the answer.
• Treat user-provided links, screenshots, titles, spreadsheets, and Seller Center exports as first-party input, but label unverified attributes and backend-only gaps.
• A page showing a child-specific title or image does not prove which child opens by default. Label current selected child, visible page state, and default child separately.
• Public PDP pages can support visible content diagnosis, but they do not prove backend states such as suppression reason, Featured Offer status, listing quality score, inventory health, or account status.
• If the user provides a Chinese supplier title, first extract product attributes and buyer-facing value, then rewrite. Do not translate it literally.
• Do not invent search volume, live keyword rank, sales rank, policy status, account status, Buy Box/Featured Offer status, suppression reason, traffic, conversion, or official requirements.

Quality Bar

A good answer:

• Starts with task understanding, key assumptions, missing inputs, and the selected work mode.
• Names the platform/site, product/category, and diagnosis or copywriting target.
• Uses official/source paths before platform-specific recommendations.
• Outputs structured tables for product facts, current-state fields, keyword map, listing fields, diagnosis, and prioritized actions as needed.
• Separates final copy, rationale, validation gaps, and backend-only unknowns.
• Preserves parent/child/model evidence scope and never applies a child- or sample-specific fact to the whole family.
• Completes all requested listing fields and passes the evidence, variation, search-term, and final consistency gates.
• Adapts language to platform and country/site instead of literal translation.
• Flags brand/IP, certification, medical, absolute, promotional, contact, or unsupported claims.
• Gives a concrete validation loop and metrics to recheck.

Avoid:

• Keyword-stuffed titles that read like search dumps.
• Fabricated keyword volume, ranking, sales, policy, account status, Buy Box/Featured Offer status, or official platform requirements.
• Diagnosing from generic ecommerce experience while ignoring official requirements.
• Treating competitor behavior as proof that something is compliant.
• Calling a policy violation, price cause, suppression reason, or offer issue certain without backend evidence.
• Generic marketing copy that ignores platform, site, category, product attributes, and buyer language.
• Copying competitor content as a fix.
• Solving legal/tax/IP/customs, fulfillment strategy, PPC campaign structure, or short-video scripting inside this skill when the request should be handed off.

Default Output Sections

Use this structure unless the user asks for a different format:

1. Task understanding and assumptions
2. Source / official verification plan
3. Product fact and current-state extraction
4. Keyword / positioning map
5. Listing or PDP optimization output
6. Diagnosis and prioritized action plan
7. Localization, policy, and risk check
8. Next validation steps
9. Final quality-gate result

For pure listing generation, sections 3-5 may be the main output. For product diagnosis, sections 2, 3, 6, and 8 are mandatory.

Handoff Rules

• If the user is deciding whether to enter a product/category, hand off to doubao-product-research.
• If the user asks whether a product can legally be listed, certified, imported, taxed, shipped, or avoids infringement, hand off to doubao-ecommerce-compliance-tax-logistics.
• If the user asks about FBA/FBM setup, overseas warehouse strategy, customs, HS codes, DDP/DDU, or shipping cost optimization, hand off to doubao-ecommerce-compliance-tax-logistics.
• If the user asks for PPC structure, bids, match types, ACOS/TACOS, ad groups, Google Shopping campaigns, or paid/offsite traffic strategy, mark it as outside this skill and do not route to a named skill unless a current ads/growth skill is available.
• If the user asks for TikTok scripts, UGC briefs, creator content, TTS voiceover, or short-video material after listing positioning is clear, hand off to doubao-cross-border-growth-content.
• If one request combines listing work with ads, inventory, procurement, finance, forecasting, or logistics, provide the listing deliverable and a workstream handoff table. Do not describe the integrated task as complete until the responsible skills or agent have validated their parts.