Zhitai Temu Capture

Zhitai Temu Capture

Use this skill to补抓当天蜘泰新品榜的合格候选，并把真实 Temu 商品详情回写到 AI 商品流水线中台。Do not build an external scraper or save a separate export unless the user explicitly asks.

Defaults

• Project: ~/Documents/Playground/flow-task-center-pro
• Console: http://127.0.0.1:4173/pipeline-console.html?workspaceId=default-workspace
• Store: data/workspaces/default-workspace/pipeline-store.json
• Zhitai page: https://www.data.izhitai.net/#/usregional/newproduct/index
• Date scope: local today only. Prefer captureBatchDate; fall back to sourceSnapshotDate.
• Item scope: only candidates that pass the project’s existing demand/risk eligibility checks and can be resolved to a real Temu product detail URL.
• Console retention: do not leave Zhitai-only candidates, pending Temu links, search-result URLs, or failed candidates in the project console as usable demand rows. Keep raw scratch data out of the main store unless it is attached to a verified Temu detail capture.

Workflow

1. Read ~/Documents/Playground/AGENTS.md and CURRENT_TASK.md if this is a new thread.

2. Work in ~/Documents/Playground/flow-task-center-pro.

3. Run the read-only status check:

   node ~/.codex/skills/zhitai-temu-capture/scripts/check_today_zhitai_targets.mjs \
     --project ~/Documents/Playground/flow-task-center-pro
   

4. Start or reuse the console server and open the console. If the Chrome extension code changed recently, reload the unpacked extension first because this flow needs the extension content/background scripts and tabs permission.

5. Do not import Zhitai rows directly into pipeline-store.json. First screen rows into a temporary candidate file:

   node scripts/import-zhitai-demand-painpoints.mjs \
     --input data/research/zhitai/YYYY-MM-DD-live-newproduct/raw-visible-rows.jsonl \
     --qualified-output tmp/zhitai-qualified-YYYY-MM-DD.json \
     --screen-only \
     --limit 30
   

6. Only for top screened candidates, use the Zhitai page/extension to open Temu. If Temu lands on search_result.html?search_key=PRODUCT_ID, extract the real product detail link from the results page and continue to the -g-PRODUCT_ID.html or goods_id=PRODUCT_ID detail page.

7. Capture true Temu detail fields: product URL, product ID, title, image URLs, attributes/spec text, size text, and material text.

8. Submit only verified captures back through /api/zhitai/product-link-captures; failed, skipped, product-id-only, constructed, or search-result captures must not be submitted.

9. To materialize a fresh batch from temp files, commit only verified captures:

   node scripts/import-zhitai-demand-painpoints.mjs \
     --input data/research/zhitai/YYYY-MM-DD-live-newproduct/raw-visible-rows.jsonl \
     --qualified-output tmp/zhitai-qualified-YYYY-MM-DD.json \
     --commit-verified data/research/zhitai/YYYY-MM-DD-live-newproduct/product-link-captures.jsonl \
     --limit 30
   

10. Remove or avoid writing unverified Zhitai-only rows from pipeline-store.json; a row without a verified Temu detail URL, title, image, and spec/attribute text must not remain in the main console as pending inventory.

11. Re-run the read-only status check and verify the console shows the updated status.

Writeback Rules

• Successful capture: store the real Temu detail URL and sourceDetails.temuProductPage; console should show Temu详情已抓.
• Success requires captureStatus: captured, a real Temu detail URL, a Temu title, at least one image URL, and spec/attribute text.
• Not eligible: do not retain it as a main console row. If an audit trail is required, mark temuDetailCaptureStatus: skipped with a concrete skip reason in a non-user-facing run/audit record.
• Jump or detail extraction failure: do not retain it as a main console row and do not fabricate a Temu detail URL.
• Never treat search_result.html, merchant/store pages, copied search pages, or constructed placeholders as the final product URL.
• Before finishing, audit the current store and remove same-day Zhitai records that do not have a verified Temu product URL.
• Do not process historical batches by default. Skip 06-11, 06-08, and any non-today batch unless the user explicitly changes scope.
• Do not create ProductTruth, supplier, listing, or publish-ready green states from Zhitai/Temu evidence alone.

Useful Project Checks

Use these after code changes or after validating a meaningful capture milestone:

cd ~/Documents/Playground/flow-task-center-pro
npm test -- zhitaiProductLinkCapture
npm run check
npm run build

After a verified milestone worth recording:

cd ~/Documents/Playground/flow-task-center-pro
npm run memory:sync

Troubleshooting

• If no targets appear, confirm today’s date and compare captureBatchDate versus sourceSnapshotDate.
• If Zhitai rows are not found, switch the table to 100 rows/page and rely on title, image key, rank, and page metadata.
• If Temu opens a search page, use the product ID in search_key to find the first matching real detail URL.
• If Chrome automation becomes unresponsive, stop live browser work and rely on store/API checks until the user refreshes the page.
• Keep secrets, cookies, credentials, and runtime config out of logs and final answers.