What problem does it solve? When working with a running CARLA simulation, requests like "the red prius" or "the nearest walker" are ambiguous — this Skill turns such descriptions into a concrete, stable actor id that other action skills (telemetry, control-vehicle, bounding-boxes) can act on, and also reads world-level data like snapshots, level bounding boxes, raycasts, and ROS 2 topic mappings. ## Core Features & Use Cases - Actor resolution: Find actors by type/blueprint filter, role, or color, or pick the single closest actor to a point or another actor with --nearest, yielding one stable id. - World queries: Read the world snapshot (frame/time/actor count), static level bounding boxes by CityObjectLabel, semantic raycast hits, and ground projection of a point. - ROS 2 topic diagnosis: List the native ROS 2 topics each actor should publish and explain why any expected topic is silent, without needing ROS 2 installed. - Use Case: A user asks "get the prius closest to my ego vehicle" — run actors --filter '*prius*' --near-id <ego> --nearest to get one id, then hand it to the telemetry skill. ## Quick Start Ask the agent to find the red prius in the running CARLA simulation and report its actor id.
