Video Account Oral Seeding Script Generator

Problem Addressed

In WeChat Video Account operations, oral seeding scripts are common for product or content promotion. Manually writing these scripts is time-consuming and lacks structural support, especially when adapting popular videos for secondary creation. This skill automates generation via a remote paid API, addressing the need for both original creation and iterative rewriting.

How It Works

The core capabilities include creation and recreation modes:

• Creation Mode: Generates scripts from scratch based on user-specified parameters such as direction, industry, and campaign purpose. It supports Chinese label display and English value mapping for enum fields to ensure compliance.
• Recreation Mode: Rewrites scripts using a WeChat Video Account share link as a skeleton. It only accepts links via --share-url and does not process local files.

Key steps involve:

1. Authentication: Reads the LY_API_KEY from config.json in the skill directory or falls back to environment variables. The request header uses the bare key (without Bearer). If missing, it guides the user to obtain and write the key.
2. Information Collection: Uses a two-step process: first fetches configuration options (e.g., enums for mode, direction), displays them to the user in Chinese labels; after user selection, maps back to English value for backend requests. The script type is fixed as oral_seeding and not user-configurable.
3. Task Creation and Polling: Assembles the request body (auto-injecting tracking parameters), calls the remote API to create a task. Polls status at --poll-interval intervals (e.g., running, completed), with intermediate results saved to response files.
4. Result Delivery: After completion, parses the data.result.markdown field from the response and outputs it verbatim as a Markdown report. It reports actual points used (via ===POINTS_USED=== marker block) and provides the file path.

Boundaries and Considerations

• Recreation Limitation: Only supports WeChat Video Account share links, not local video uploads. User-provided links must be passed via --share-url as-is; do not request or guess video_id.
• Billing Model: Each task incurs points, with an estimate around 98 points (indicative only; actual usage based on account records). Failed tasks have points refunded, and --task-id can resume the same task to avoid double billing.
• Parameter Validation: Enum fields must use English value for backend requests; Chinese labels are for user display only—sending Chinese causes 400 errors. The script type is fixed as oral_seeding and cannot be overridden.
• Error Handling: Specific exit codes (e.g., 8 for auth failure, 5 for polling timeout). Unified messaging attributes failures to network issues, guiding users to retry or resume without exposing backend error details.
• Configuration Dependency: Relies on remote API availability. If config fetch fails, --skip-config can bypass enum validation, but at the user's own risk.