Each entry pairs a category slug with the number of retrievable
    records it holds. Call this before retrieve_articraft to discover
    valid category names.

    Args:
        contains: Case-insensitive substring filter on the category
            slug; omit to list every category.

    Returns the matching categories, each with its record count.
    

    Each record is a URDF asset with meshes and physical joints that
    generate_scene can place. Use list_categories first to find valid
    category slugs.

    Args:
        category: Exact category slug from list_categories.
        query: Free-text description used to rank records within the
            category; omit to keep catalog order.
        top_k: Maximum number of records to return.

    Returns the ranked records with their ids and metadata.
    

    Runs the full Shell->SceneReview pipeline; spends real LLM budget
    (bounded by the resolved budget cap) and takes minutes.
    

    Re-runs the pipeline from start_stage using checkpoints already on
    disk, writing a new revision and leaving prior revisions untouched.
    Cheaper than generate_scene when only later stages need to change;
    spends real LLM budget and takes minutes.

    Args:
        scene_hash: Session id from list_sessions or generate_scene.
        start_stage: First stage to re-run, one of "shell", "furniture",
            "physics", "scene_author", "scene_review".
        stop_stage: Optional last stage to run; omit to run to the end.

    Returns the new revision id, status, quality status, and cost.