Product and review research workflow
Product and review research workflow. Follow concrete contracts, examples and reviewable steps using TikTok, Xiaohongshu (RedNote).
Copy the agent prompt manually
Inputs, goal and deliverable
Connect product identifiers to review samples and distinguish opinions from sales facts.
Inputs consist of explicit platforms, keywords or object identifiers. Preserve the samples, request IDs, classifications and unresolved questions in the deliverable. The task does not schedule itself or paginate indefinitely.
Step-by-step plan
| Step | Capability | Input | Output |
|---|---|---|---|
| 1 | tiktok.shop.products.search | query | product / items |
| 2 | tiktok.shop.products.get | id | product / item |
| 3 | tiktok.shop.products.reviews.list | id | review / items |
Reusable task prompt
Ask the client to present a plan before you confirm each execution. This prompt bounds the sample; it does not call an external service for you.
Goal: Product and review research workflow Use only DataNavoAPI local demo/simulated mode. Planned capabilities: tiktok.shop.products.search → tiktok.shop.products.get → tiktok.shop.products.reviews.list Describe inputs and output fields and propose a bounded request count first. Execute explicitly, retaining request IDs. Stop on failures or unknown outcomes. Return a table, missing fields and sampling limits; never present fictional samples as market facts.
Plan, run and export
This bounded example has 3 steps. Successful customer simulation costs 6 TEST; anonymous demo costs zero. Provide each required identifier before generating a fixed plan; search results do not expand the request list automatically.
Use data.items[].id for the matching platform and object detail, account IDs for account content and comment IDs for replies. Check object types before reusing an ID. Fixture IDs do not establish a real relationship. The workbench exports per-step JSON and source-labelled CSV.
| Workbench field | Meaning | Example |
|---|---|---|
| query | 研究关键词 | 户外咖啡 |
| step2_id | 步骤 2 · id | 1732488007561286228 |
| step3_id | 步骤 3 · id | 1732488007561286228 |
{
"schemaVersion": 1,
"templateId": "products",
"runId": "00000000-0000-4000-8000-000000000001",
"inputs": {
"query": "户外咖啡",
"step2_id": "1732488007561286228",
"step3_id": "1732488007561286228"
},
"maxTotalCredits": 6,
"estimatedCredits": 6,
"priceVersion": "test-credit-v1",
"steps": [
{
"id": "step-1",
"title": "TikTok Shop 商品搜索",
"capability": "tiktok.shop.products.search",
"params": {
"query": "户外咖啡"
},
"maxCredits": 2,
"idempotencyKey": "wf:00000000-0000-4000-8000-000000000001:step-1"
},
{
"id": "step-2",
"title": "TikTok Shop 商品详情",
"capability": "tiktok.shop.products.get",
"params": {
"id": "1732488007561286228"
},
"maxCredits": 2,
"idempotencyKey": "wf:00000000-0000-4000-8000-000000000001:step-2"
},
{
"id": "step-3",
"title": "TikTok Shop 商品评价",
"capability": "tiktok.shop.products.reviews.list",
"params": {
"id": "1732488007561286228"
},
"maxCredits": 2,
"idempotencyKey": "wf:00000000-0000-4000-8000-000000000001:step-3"
}
]
}Review and stop conditions
- Each conclusion links back to a sample and request ID.
- Pagination stops at the agreed bound.
- Missing fields remain missing rather than being invented by the model.
- External writes, scheduled monitoring and automatic top-ups are not enabled by this workflow.
Clear inputs. Inspectable output.
Each click runs a local fixture. Advanced JSON inputs are checked against the selected method. Editing does not execute a request.
Copying does not execute. CLI quote is read-only and needs a local customer TEST key. Configure MCP in demo mode first. GET and POST batches both return only fictional local fixtures.
Download contract example (no execution) ↗For TEST budgets, Key scopes or failure scenarios, open the account console. Customer sandbox ↗