Z-Sight APP Skills (zmg-zsj-skill)

Core capabilities for the Z-Sight APP, including live streaming search, short video and variety show browsing, trending topics, app page navigation, content publishing, and more.

Capability List

Usage Scenarios

Search

• User wants trending topics -> use search_hot
  • No parameters required, returns trending search keywords
  • Example: "What's trending now"
• User wants to find a live stream -> use search_live with a keyword
  • Example: "Search for Jay Chou's live stream"
• User wants to find a variety show -> use search_variety with a keyword
  • Example: "Search for Keep Running"

Content Browsing

• User wants hot live streams -> use get_hot_live_list
  • Returns current hot live streams with ID, title, subtitle, and watch URL
  • Example: "What live streams are on now"
• User wants to enter a specific live room -> use get_live_detail with liveId
  • Returns live detail H5 page URL and app schema
• User wants to watch TV live -> use watch_live_channel with channel name
  • Supported channels: Zhejiang TV, Qianjiang, Economy & Life, Education, Minsheng, News, Kids, Zhejiang International, Haoyigou, Zhijiang Documentary
  • Returns channel live H5 page URL
  • Defaults to Zhejiang TV if no match
• User wants trending variety shows -> use get_variety_list
  • Returns variety show ranking with name and play page URL
• User wants trending short videos -> use get_short_video_list
  • Returns short video ranking with title and share page URL

App Navigation

• User wants to open the app to a specific page -> use open_app_page with schema
  • Returns a deep link (H5 page), displayed as hyperlink for users to click

Content Creation

• User wants to publish a text & image post -> use publish_post
  • Required: title (post text content)
  • Optional: img_array (image list with pic/height/wide), tags, description
  • Returns article_id on success
• User wants to publish a short video -> use publish_short_video
  • Required: title, video_url
  • Optional: cover_img, width/height, play_time (["HH","MM","SS"]), tags, description
  • Returns article_id on success

Messages

• User wants to see mentions -> use get_message_list with msg_type: "mention"
  • Returns mention list with username, avatar, and post content
• User wants to see comments/replies -> use get_message_list with msg_type: "comment"
  • Returns comment list with content and post info
• User wants to see likes -> use get_message_list with msg_type: "like"
  • Returns like list

User Content

• User wants to see their short videos -> use get_user_content_list
  • Returns short video list with title, content, cover, likes, comments, views, and content type
  • User identity is auto-detected from the auth token. Do NOT pass uid, accessUuid, or other user identifiers.
• User wants to see their posts -> use get_user_post_list
  • Returns post (text & image) list
  • User identity is auto-detected. Do NOT pass uid or accessUuid.
• User wants to see their audio -> use get_user_audio_list
  • Returns audio list (e.g. KTV recordings)
  • User identity is auto-detected. Do NOT pass uid or accessUuid.

Content Management

• User wants to delete an article -> use delete_article with article_id
  • Must first call get_user_content_list, get_user_post_list, or get_user_audio_list to get the article ID (id field)
  • Deletion is irreversible. AI should confirm with the user before deleting.

MCP Configuration

MCP Server address (built into skill.json):

{
  "mcp": {
    "transport": "sse",
    "url": "https://zmg-mcp.cztv.com/sse",
    "auth": {
      "type": "bearer",
      "header": "Authorization",
      "token_prefix": "zmg_",
      "acquire_url": "https://zmg-mcp.cztv.com"
    }
  }
}

Client configuration example (replace \x3Cai_token> with actual token):

{
  "mcpServers": {
    "zmg-zsj-skill": {
      "url": "https://zmg-mcp.cztv.com/sse",
      "headers": {
        "Authorization": "Bearer \x3Cai_token>"
      }
    }
  }
}

Authentication Model

The MCP Server is stateless and does not manage user sessions. Authentication is handled via ai_token sent with each request:

• Source: ai_token is issued after the user scans a QR code in the app (format: zmg_xxxxx), stored in Redis
• How to get: Visit https://zmg-mcp.cztv.com, scan the QR code, then copy the ai_token into client config
• How to pass: Set Authorization: Bearer \x3Cai_token> in MCP request headers
• Validation: NG Gateway validates the ai_token via Redis, retrieves sessionId, and injects it into business requests

Notes

• channel parameter accepts channel names (e.g. "Zhejiang TV"), supports fuzzy matching, defaults to Zhejiang TV
• get_live_detail requires a live ID to get the detail page URL
• open_app_page returns a link that can be used as a hyperlink to launch the app
• Publishing APIs (publish_post / publish_short_video) require user authorization; return 401 if unauthorized or expired
• No need to pass uid, sessionId, or other sensitive info -- the system handles this automatically

401 Authorization Error Protocol (Highest Priority)

When ANY API returns code: 401, immediately follow this procedure:

1. Stop the current task immediately -- do not continue calling any APIs
2. Tell the user to scan the QR code and display:
   • Prompt: "Authorization required. Please scan the QR code with the Z-Sight APP"
   • Display the authorization URL as a hyperlink: Click to authorize
   • Authorization page URL: https://zmg-mcp.cztv.com
3. After the user scans and authorizes, update the new ai_token in the MCP config Authorization: Bearer \x3Cai_token>
4. Retry the original request after updating the token

401 Response Format:

{
  "code": 401,
  "message": "Authorization expired, please re-scan to authorize",
  "action": "REAUTHORIZE",
  "authorize_url": "https://zmg-mcp.cztv.com"
}

Strictly prohibited:

• Returning "I don't have permission" or "I can't do this" after receiving a 401 -- must guide the user to scan the QR code
• Ignoring 401 and retrying the request
• Returning only the error message without the authorization link