Sourcing Selection
Follow shared release-shell rules in:
Use this skill when the user is not just asking for platform data, but for a real sourcing or product-selection judgment.
Typical requests:
- 这个产品值不值得找货源
- 适合先上 Amazon 还是 TikTok Shop
- 1688 上有货，但有没有需求侧证据支撑
- 帮我把供给侧和需求侧拼起来做判断
- 给我一个更接近真实决策的找货源结论
This skill is an orchestration and synthesis layer.
It should not replace platform skills.
It should decide which evidence to collect, in what order, and how much judgment is justified.
Read first:
- product-selection preferences
Design Goal
Keep the current version simple, but make the interface extensible.
The skill should think in capability groups, not hard-coded platforms:
Today, these groups may map to:
- for supply-side
- for search-intent
- for search-led demand
- for marketplace demand
- for content and audience language
In the future, the same groups may map to:
- , , , , offline vendor lists
- , , ad-library, search-console-like, , , ,
Do not write the skill as if
1688 + Amazon + TikTok Shop
are permanent.
Core Rule
A sourcing judgment is only as strong as its weakest missing layer.
Always separate:
If the user asks for a yes/no decision and the evidence is incomplete, return:
- the provisional judgment
- what supports it
- what is still missing
Do not fake certainty.
Minimal Decision Model
Use this order unless the user explicitly asks otherwise:
- demand proof
- competition shape
- channel fit
- merchant-model fit
- supply-side feasibility
- unit-economics pressure
- compliance or operational risk
This is intentionally simple.
Do not turn it into a giant scorecard unless the user asks.
Input Shapes
Classify the request first:
1. Product Idea Validation
Use when the user asks:
Default route:
- collect demand-side proof
- collect supply-side feasibility
- synthesize
2. Supply-Led Opportunity Check
Use when the user already has a supply-side signal:
- 1688 上看到很多货
- 某工厂或类目看起来很便宜
- 已经有一批供应商候选
Default route:
- inspect supply-side evidence
- collect matching demand-side proof
- test channel fit
- synthesize
3. Demand-Led Sourcing Check
Use when the user already has a demand-side signal:
- Amazon 上卖得不错
- TikTok Shop 上很多人在卖
- 某类内容在 TikTok 很火
Default route:
- inspect demand-side proof
- collect supply-side feasibility
- synthesize
4. Shortlist Comparison
Use when the user has:
- several products
- several niches
- several supplier options
Default route:
- normalize each candidate into the same decision frame
- compare strongest evidence and biggest gaps
- rank cautiously
Capability Routing
Choose sources by role, not by habit.
Supply-Side Source
Use for:
- factory options
- supplier variety
- MOQ
- tiered pricing
- customization
- location
Current preferred route:
skills/20-research/1688-research
Demand-Side Source
Use for:
- listings
- pricing
- reviews
- order or ranking proof
- bestseller shape
- channel-native competition
Current preferred routes:
- Amazon search-led demand ->
skills/20-research/amazon-research
- TikTok Shop marketplace demand ->
skills/20-research/tiktok-shop-research
Search-Intent Source
Use for:
- early demand signals
- topic or keyword momentum
- geo search interest
- rising-query discovery
Current preferred route:
- Google search-intent ->
skills/20-research/google-trends-research
Treat this as an early signal layer.
Do not confuse it with transaction demand or channel-native competition proof.
Content-Language Source
Use when content-led selling matters:
- what hooks are working
- what user language repeats
- what visual demo style fits the product
Current preferred route:
skills/20-research/tiktok-research
If the request is broader than one named platform and the goal is to compare social proof or audience language across networks, route first through:
skills/10-routing/social-media-extractor
Use this layer only when it changes the decision.
Do not force it into every sourcing task.
When social proof is cross-platform, do not let one familiar network stand in for the whole market.
Use the extractor to decide which platform-specific research skill should collect first.
Extensibility Rule
When a new platform appears, do not rewrite the decision model.
Instead, map it into one of these roles:
Then state:
- what role the source covers
- what role is still missing
This keeps the skill stable while letting the source set expand.
Good Output
Return a compact decision memo with:
- product or niche
- target merchant model
- target channel
- observed evidence
- provisional judgment
- biggest risks
- missing layer
- recommended next step
Good recommendation shapes:
promising, but demand proof still thin
good Amazon search fit, weak TikTok demo fit
cheap supply exists, but competition is commodity-price-led
strong demand and workable sourcing, but returns or compliance may kill margin
If the result is going to move into execution, keep the handoff explicit:
- sourcing judgment -> merchant or channel decision
- merchant or channel decision -> research expansion, supplier outreach, or brief creation
Do not blur evidence collection, business judgment, and execution prep into one opaque step.
Failure Modes To Avoid
Do not:
- treat one platform's popularity as universal demand proof
- treat cheap 1688 supply as a recommendation by itself
- jump from TikTok content heat to Amazon launch logic without search proof
- jump from Amazon demand to TikTok Shop without content-demo fit
- hide missing finance or compliance layers
Current Workspace Default
At the moment, this skill should usually compose existing skills rather than create a brand-new collection workflow.
Current building blocks:
- supply-side:
skills/20-research/1688-research
- search-intent:
skills/20-research/google-trends-research
- search-led demand:
skills/20-research/amazon-research
- marketplace demand:
skills/20-research/tiktok-shop-research
- content-language fit:
skills/20-research/tiktok-research
Future sources should be slotted into the same roles.