Vs Recommend

Viking Recommend

When to Use

Use this skill for recommendation runtime checks, recommend scene management, and first-pass verification of the recommendation path.

Preconditions

• an application-id is available
• a recommendation request will usually also need scene-id and user-id
• if the scene does not exist yet, inspect the existing scene list first and only create a new one when reuse is not possible

Commands

• recommend run: send a production-style recommendation request
• recommend scene create / recommend scene list / recommend scene get: manage recommend scenes
• recommend scene update: update scene configuration

Workflow

1. Confirm application-id, scene-id, and user-id 2. Run recommend scene list first and prefer an existing/default scene before creating a new one 3. Before recommend scene create or recommend scene update, explicitly confirm the target page / module and the required BhvSceneTypes with the user 4. Use recommend run for the first verification request 5. Read recommendation items from the raw response structure, especially result.rec_results 6. If the result looks wrong, inspect the scene with recommend scene list/get 7. Update the scene configuration when needed, then rerun the request

Customer Environment Principle

• In customer environments, assume repository source code is unavailable.
• Execute tasks using only the installed skills, the packaged vs CLI surface (--help, command output, and observed runtime behavior), and explicit user-provided information.
• Do not rely on reading local repository source files, generated repo snapshots, or implementation details to decide runtime actions.
• If the installed CLI behavior conflicts with a skill, trust the installed CLI behavior first.
• If the skills and the packaged CLI still do not provide enough information to proceed safely, stop and ask the user instead of searching source code.

Constraints

• Start with the scene when debugging recommendation behavior; do not jump to raw API calls first
• If the user only needs a first-pass conclusion, prefer recommend run
• Do not create or update a recommend scene until the user has confirmed the target page / module and BhvSceneTypes
• When reporting results, summarize the scene, the user context, and the raw response before proposing tuning changes
• Do not invent item titles or explanations. Ground every recommendation summary in the actual response payload
• If you show only a subset such as Top 5, explicitly say that the full response contains more items