Tencent Marketing Delivery Smart Create

The Problem This Solves

Manually creating a Smart Delivery ad on the Tencent Ads platform requires navigating a web of interdependent fields and enum values, including scenario identification, quadruple matching, asset association, conversion goal selection, and bidding strategy. The nested structures from rules_json, account-specific access restrictions, and platform-specific argument passing quirks make manual automation error-prone. A single wrong field choice triggers API errors, while inefficient polling or retry strategies can exhaust execution rounds.

Core Capabilities & Workflow

This skill structures the Smart Delivery ad creation process into a strict seven-step sequence, with dedicated scripts handling API complexity.

1. Script-Mediated API Calls: All API interactions (e.g., get-rules.mjs, get-assets.mjs) are executed via purpose-built scripts. These scripts handle API routing, data trimming, and format conversion (e.g., parsing rules_json, encoding geo lookups). The Agent focuses on business decisions derived from the script outputs, such as selecting the correct quadruple from multiple combinations.
2. The Non-Negotiable First Three Steps:
   • Step 1 (Scenario Identification): Map user intent to a smart_delivery_platform enum, e.g., Mini Game Promotion corresponds to SMART_DELIVERY_PLATFORM_EDITION_MINI_GAME_PROMOTION.
   • Step 2 (Fetch Quadruple): Call get-rules.mjs to retrieve the account-eligible combinations of marketing_goal, marketing_sub_goal, marketing_target_type, and marketing_carrier_type. Guessing is forbidden; the unique quadruple must be selected from the returned combinations list using defined priority rules (e.g., reverse inference via asset ID, user intent keywords).
   • Step 3 (Fetch Promotion Product & Carrier): Based on the selected marketing_target_type, call get-assets.mjs to get the list of available assets. The final output is a flat_params object whose fields (e.g., asset_id, carrier_id) are passed directly to the final creation script.
3. Key Decision Points: In subsequent steps, the Agent must make choices based on user input (e.g., ROI coefficients, bid amounts, conversion source) and system responses (e.g., site_set lists, conversion_id lists). For instance, conversion_id must come from a query result or be explicitly provided by the user.
4. Strict Execution Protocol:
   • Step Dependency: Before advancing, required data must be confirmed to originate from the previous step's output or explicit user input. Natural language descriptions cannot replace structured fields.
   • Error Handling & Retry Limits: Each script call allows a maximum of 1 retry. If both attempts fail, the step is skipped, and execution continues with available data to avoid infinite loops.
   • Explicit Evidence Allowlist: Only for query sub-actions in Steps 4-6 is skipping permitted if the user provides complete IDs or the data was fetched in the current session. Steps 1-3 have no such exemption.

Boundaries & Caveats

• Prerequisites: Requires tencentads-cli (Node.js ≥ 20) installed globally.
• Cross-Platform Compatibility: The script call format node scripts/.mjs '' varies by terminal. Adhere to the documented escaping rules, particularly for Windows PowerShell 5.x (using --% stop-parsing token).
• Scenario & Enum Specificity: Enum names may diverge from Chinese business names (e.g., SINGLE_PRODUCT = "Short Video & Live Dual-Link"). Always refer to the Step 1 table for precise matching.
• Decision Transparency: Before entering Step 7, the origin of every critical field (e.g., conversion_id, asset_id)—either user-provided or returned by a preceding script—must be traceable. Fabrication is prohibited.