Taobao UI Automation Skill

Core Behavior

This skill provides a reusable Taobao UI automation workflow that:

• validates normalized input payloads
• performs keyword search
• extracts and filters product candidates with deterministic rules
• optionally executes add-to-cart
• returns a stable structured result schema

Expected Input

Input is a JSON object with fields such as:

• platform
• keyword
• constraints
• action
• notify_channel
• max_candidates
• need_login

Example (sample only, not capability boundary):

{
  "platform": "taobao",
  "keyword": "蓝牙耳机",
  "constraints": {
    "positive_rate_gte": 99,
    "price_lte": 300
  },
  "action": "search_only",
  "notify_channel": "feishu",
  "max_candidates": 5,
  "need_login": false
}

Output Format

Standard response fields:

• success
• task_status
• selected_product
• message
• error_code
• error_detail

Internal task_id is used for tracing and logging, and is not part of the standard output schema.

Standard Workflow

1. Validate payload and normalize constraints.
2. Start runtime and open target page.
3. Check login/captcha/risk-control status.
4. Execute search and wait for parse-ready state.
5. Extract product candidates.
6. Filter and rank with constraints.
7. Return selected product for search_only, or continue with add-to-cart for add_to_cart.

Login / Captcha / Risk-Control Policy

• The skill does not implement captcha bypass.
• Login, captcha, risk-control, and complex SKU cases are handled with human-in-the-loop semantics.
• When login is required and unavailable, return LOGIN_REQUIRED.
• When captcha signals are detected, return CAPTCHA_DETECTED.
• When risk-control signals are detected, return RISK_CONTROL_PAGE.
• These human-required states should map to task_status=need_human_intervention.

Candidate Extraction Policy

• Candidate extraction is modular and replaceable.
• The current implementation favors deterministic DOM/locator parsing with safe fallbacks.
• Missing optional fields are allowed and represented as None.
• Title is required for a valid candidate.

Filtering Policy

• Filtering is driven by keyword and constraints.
• Keyword match is case-insensitive substring matching on candidate title.
• Constraint checks include positive rate, price range, and shop-type matching.
• No matched candidate should return NO_MATCHED_PRODUCT.

Add-to-Cart Policy

• Add-to-cart runs only for action=add_to_cart.
• The executor uses centralized selectors and success signals.
• For complex SKU pages requiring manual decision, return SKU_SELECTION_REQUIRED.
• If success cannot be confirmed after click + short polling, return ADD_CART_FAILED.

Error Codes

Typical structured codes include:

• INVALID_INPUT
• LOGIN_REQUIRED
• CAPTCHA_DETECTED
• RISK_CONTROL_PAGE
• SEARCH_TIMEOUT
• NO_MATCHED_PRODUCT
• ADD_CART_FAILED
• SKU_SELECTION_REQUIRED
• PAGE_STRUCTURE_CHANGED
• NETWORK_ERROR
• UNKNOWN_ERROR

Boundaries

• Platform scope is currently taobao only.
• No stealth or anti-bot plugin is required or enabled by default.
• No attempt is made to bypass captcha or risk-control restrictions.
• Complex SKU automation is intentionally limited and may require manual intervention.

Local Mock Smoke Usage

Use local mock pages for stable end-to-end demonstration without relying on the real Taobao website:

• From workspace parent:
• python taobao_ui_automation_skill/tools/run_mock_smoke.py
• From project root:
• python tools/run_mock_smoke.py

The mock smoke path validates:

• candidate extraction from local search page
• deterministic filtering and best-candidate selection
• add-to-cart execution and success confirmation on local detail page