What problem does it solve? Replaying an entire CARLA simulation recording just to answer a simple question — what map was used, which vehicles collided, which actors got stuck — wastes time. This Skill reads facts directly out of a recorded .log file on the server and returns text reports, so you can diagnose a run without touching the simulation world. ## Core Features & Use Cases - Recording overview: Run the info query to get the map, duration, frame count, and full actor list from a .log header, with an optional --all flag for per-frame detail. - Collision queries: Find recorded collisions between actor categories (hero, vehicle, walker, traffic light, other, any) using show_recorder_collisions, with each row giving frame, time, and actor ids. - Blocked-actor detection: Identify vehicles that moved less than a threshold distance (in centimetres) over a minimum time (in seconds) to spot gridlock or wedged actors. - Use Case: After recording a traffic scenario, ask "did any car hit a pedestrian?" — the collisions query with --type1 v --type2 w returns the exact frames and actor ids, which you then feed to the replay-recording skill to watch that moment. ## Quick Start Ask the agent to show what is inside your recording, for example: "what's in /tmp/run.log and did any vehicles collide with walkers?"
