開発のスキル

Explores a CARLA map's road network through the Map/Waypoint API — a natural-language rundown of the whole map ("give me the topology"), drawing the topology to confirm the modelled map matches the visible one, enumerating junctions to resolve phrases like "the 4-way junction in the middle", nearest-waypoint lane facts, and next/previous lane navigation. Use when the user asks about the map's layout, lanes, junctions, topology, or to locate/navigate specific map elements.

日本語の概要は準備中です。原文の説明を表示しています。

Loads, reloads, and reshapes the map on a running CARLA server via the Python API — get_available_maps, load_world and reload_world (with or without resetting world settings), generate a world from an OpenDRIVE .xodr, and load/unload map layers. Use when the user asks to "load a map", "change the town", "reload the world", "list available maps", "load an xodr/opendrive map", or "load/unload map layers", and when settings (sync mode, fixed delta) must survive the switch.

日本語の概要は準備中です。原文の説明を表示しています。

Visualises and validates a map's pedestrian navigation mesh by sampling walkable locations (get_random_location_from_navigation) and drawing them as debug points. Use when the user asks to "show/visualise the pedestrian navmesh", "check walkable areas", "validate pedestrian navigation after a map import", or "why won't my walkers move". Confirms the navmesh loaded and how far it covers.

日本語の概要は準備中です。原文の説明を表示しています。

Draws debug overlays in a running CARLA world via world.debug — points, lines, arrows, bounding boxes, and text strings, each with a colour and lifetime. Use when the user asks to "draw/mark/highlight a point or box in the world", "visualise a path/waypoints", "label a location", or "show something on the map for debugging". Shapes are an overlay, not actors.

日本語の概要は準備中です。原文の説明を表示しています。

Directly drives a specific vehicle on a running CARLA server — raw VehicleControl (throttle/steer/brake/reverse/hand-brake), Ackermann speed control, vehicle lights, opening/closing doors, and physics-control tuning; ros-info reports the native ROS 2 command topics (vehicle_control_cmd, ackermann_control_cmd) for the hero. Targets the ego by default. Use when the user asks to "drive/steer the car", "accelerate/brake/reverse", "set a target speed", "turn on the headlights/blinkers/brake lights", "open a door", "drive the car from ROS", or "change the vehicle's mass/drag/physics". This is manual control (autopilot off) — for TM-driven traffic use control-traffic.

日本語の概要は準備中です。原文の説明を表示しています。

Tunes the CARLA Traffic Manager that drives autopilot vehicles — global and per-vehicle speed, following distance, running lights/signs, lane-change behaviour, lane offset, TM-managed lights, deterministic seed, hybrid physics, and TM sync mode. Use when the user asks to "make the traffic faster/slower", "let cars run red lights", "keep more distance", "stop lane changes", "make traffic deterministic", or otherwise change how autopilot cars behave. For manually driving one car, use control-vehicle.

日本語の概要は準備中です。原文の説明を表示しています。

Controls the map's light sources through the CARLA LightManager — turn street lamps, building/facade lights, vehicle lights or all of them on/off, set their colour and intensity, and toggle the automatic lights-at-night cycle. Use when the user asks to "turn on the street lights", "light up the city at night", "turn off building lights", "change the street-light colour", or "make lights come on automatically at night". Distinct from hiding lamp meshes (toggle-env-objects) and the sun (set-weather).

日本語の概要は準備中です。原文の説明を表示しています。

Gets actor and level 3D bounding boxes and projects them into a camera image as 2D boxes — the dataset-annotation workflow. Lists box geometry, draws 3D boxes in the world via debug, or captures one camera frame and writes an annotated PNG plus a JSON of 2D boxes for matching actors. Use when the user asks to "get bounding boxes", "draw boxes around the cars", "project boxes onto the camera", or "generate a detection dataset / annotations".

日本語の概要は準備中です。原文の説明を表示しています。

Runs the CARLA Leaderboard evaluator over a set of routes with your agent — routes/routes-subset/repetitions/track selection, checkpoint and live-results endpoints, resume after a crash, recording, and the debug levels — for leaderboard 1.0, 2.0 or 2.1. Preflights the whole stack (version pairing, maps, ports, agent class, sensor budget) before spending hours, and hands the world back in async mode. Use when the user asks to "run the leaderboard", "evaluate my agent", "run routes_training", "resume an evaluation", or "reproduce my submission locally".

日本語の概要は準備中です。原文の説明を表示しています。

Creates and edits CARLA Leaderboard route files — placing waypoints and scenario trigger points interactively from the spectator camera (route_creator, scenario_creator), visualising an existing route in the simulator (route_displayer), summarising it as a table (route_summarizer), capturing the current weather in route format (weather_creator), sorting scenarios by route position (scenario_orderer) and converting 1.0-format routes to 2.x (route_bridge). Use when the user asks to "make a new route", "add a scenario to a route", "visualise/debug a route", "see what's in this route file", or "convert old leaderboard routes".

日本語の概要は準備中です。原文の説明を表示しています。

Spawns and attaches sensors on a running CARLA server — cameras (rgb/depth/semantic/instance/optical-flow/normals), lidar, radar, IMU, GNSS, collision/lane-invasion/obstacle — to the ego or any actor, with a mount transform and blueprint attributes (resolution, fov, sensor_tick, lidar range/points, …). Also sets up native ROS 2 publishing (--ros to enable_for_ros, --ros-name/--ros-frame-id topic and TF naming) and prints the topics the sensor will publish. Use when the user asks to "add a camera/lidar/sensor", "put a dashcam on the ego", "attach a depth camera", "mount a sensor", or "publish a sensor to ROS". Prints the sensor id; read its data with the read-sensor skill.

日本語の概要は準備中です。原文の説明を表示しています。

Reads and sets the weather on a running CARLA server via get_weather/set_weather, and turns natural-language weather requests ("heavy rain at sunset", "light fog at night", "clear noon", "dust storm") into exact WeatherParameters — cloudiness, precipitation, puddles, wind, fog, sun position, wetness, dust. Use when the user asks to "set/change the weather", "make it rainy/foggy/sunny/night", "apply a weather preset", or describes any sky/lighting condition.

日本語の概要は準備中です。原文の説明を表示しています。

Moves the CARLA spectator camera, frames an actor, or follows one live — third-person/chase, first-person/driver, top-down, or front views, resolving natural-language targets like "the ego" or "the Toyota Prius" to a specific actor. Use when the user asks to "move the camera/spectator", "watch/follow vehicle X", "3rd-person view of the ego", "1st-person view of the prius", "bird's-eye view", or "look at (x,y,z)".

日本語の概要は準備中です。原文の説明を表示しています。

Installs the CARLA Python client (the `carla` package) into a chosen interpreter so the other python-api skills work — from a wheel bundled with your release or checkout, from PyPI (`pip install carla==X.Y.Z`), or by putting an older release's .egg on PYTHONPATH. Detects what is already present, picks the source that matches your simulator's version, pins numpy<2, and verifies `import carla` plus the client/server version match. Use when the user asks to "install the CARLA Python API", "pip install carla", "set up the carla python package", or when a skill reports "cannot import carla".

日本語の概要は準備中です。原文の説明を表示しています。

Inspects a recorded CARLA .log without replaying it — show_recorder_file_info for the header/actors/frames, collision queries between actor categories, and blocked-actor queries for vehicles that got stuck. Use when the user asks to "what's in this recording", "show recorder info", "which vehicles collided", "find collisions in the log", or "which actors were blocked/stuck".

日本語の概要は準備中です。原文の説明を表示しています。

Controls the traffic-light actors on a running CARLA server — set a light green/red/yellow/off, freeze all lights (optionally all-green), set phase timing (green/yellow/red durations), and reset to the normal cycle. Target all lights, one by id, all in a junction, or the nearest. Use when the user asks to "make the light red", "freeze all lights green", "change the light timing", "turn the traffic lights off", or "reset the lights". This is the light actors, not the Traffic Manager (that's control-traffic).

日本語の概要は準備中です。原文の説明を表示しています。

Imports an FBX pedestrian into a CARLA source build as a spawnable walker.pedestrian.NNNN — checks the FBX is skinned to CARLA's GEN3 rig, imports the mesh, physics asset and materials, builds the walker blueprint from a GEN3 donor, registers it in WalkerFactory through CARLA's own add_walker_to_walker_factory.py, and verifies it spawns, stands on the ground and walks. Use when the user asks to "add a pedestrian to CARLA", "import a walker/character FBX", or needs a custom human spawnable from the Python API.

日本語の概要は準備中です。原文の説明を表示しています。

Enforce this repository's TypeScript backend module architecture standards. Use whenever creating, modifying, refactoring, or reviewing backend code under `server/`, including routes, services, modules, repositories, shared backend code, and backend tests. Do not apply these rules to frontend code or non-backend scaffolding.

日本語の概要は準備中です。原文の説明を表示しています。

Enforce this repository's React and TypeScript frontend module architecture standards. Use whenever creating, modifying, refactoring, or reviewing frontend code under `src/`, including feature modules, components, hooks, contexts, shared UI, API access, state, utilities, and frontend tests. Do not apply these rules to backend code under `server/` or non-frontend scaffolding.

日本語の概要は準備中です。原文の説明を表示しています。

Accelerate LLM inference using speculative decoding, Medusa multiple heads, and lookahead decoding techniques. Use when optimizing inference speed (1.5-3.6× speedup), reducing latency for real-time applications, or deploying models with limited compute. Covers draft models, tree-based attention, Jacobi iteration, parallel token generation, and production deployment strategies.

日本語の概要は準備中です。原文の説明を表示しています。

Evaluates LLMs across 100+ benchmarks from 18+ harnesses (MMLU, HumanEval, GSM8K, safety, VLM) with multi-backend execution. Use when needing scalable evaluation on local Docker, Slurm HPC, or cloud platforms. NVIDIA's enterprise-grade platform with container-first architecture for reproducible benchmarking.

日本語の概要は準備中です。原文の説明を表示しています。

Evaluates code generation models across HumanEval, MBPP, MultiPL-E, and 15+ benchmarks with pass@k metrics. Use when benchmarking code models, comparing coding abilities, testing multi-language support, or measuring code generation quality. Industry standard from BigCode Project used by HuggingFace leaderboards.

日本語の概要は準備中です。原文の説明を表示しています。

High-level PyTorch framework with Trainer class, automatic distributed training (DDP/FSDP/DeepSpeed), callbacks system, and minimal boilerplate. Scales from laptop to supercomputer with same code. Use when you want clean training loops with built-in best practices.

日本語の概要は準備中です。原文の説明を表示しています。

Provides guidance for training and analyzing Sparse Autoencoders (SAEs) using SAELens to decompose neural network activations into interpretable features. Use when discovering interpretable features, analyzing superposition, or studying monosemantic representations in language models.

日本語の概要は準備中です。原文の説明を表示しています。