「rl」の検索結果

Método da Anthropic para treinar IA inofensiva através de auto-aprimoramento. Abordagem em duas fases - aprendizado supervisionado com auto-crítica/revisão, depois RLAIF (RL from AI Feedback). Use para alinhamento de segurança, redução de outputs prejudiciais sem rótulos humanos. Alimenta o sistema de segurança do Claude.

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。

調査、裏取り、検証、コードリーディング、外部URL確認、GitHub PRやIssue確認、BigQueryなどのコマンドやクエリ実行を行う際に自動的に使用。清書用ドキュメントとは分離して、.agent_work/investigations/ にエージェント向けの内部調査ログを根拠URL、実行コマンド、クエリ、結果、未確認事項つきで残します。

Discover and list all URLs on a website without extracting content, via the Tavily CLI. Use this skill when the user wants to find a specific page on a large site, list all URLs, see the site structure, find where something is on a domain, or says "map the site", "find the URL for", "what pages are on", "list all pages", or "site structure". Faster than crawling — returns URLs only. Essential when you know the site but not the exact page. Combine with extract for targeted content retrieval.

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。

Route ordinary public-web information tasks to the correct Tavily CLI operation. Use when the task concerns publicly accessible webpages and needs web discovery, clean extraction, URL mapping, same-site crawling, or public-web-only research, but the exact Tavily operation is not yet clear or several operations must be sequenced. Prefer the specific leaf skill when intent is clear: tavily-search for discovery without a URL, tavily-extract for known URLs, tavily-map for URL/site-structure discovery, tavily-crawl for content from many pages on one site, tavily-research for cited multi-source synthesis, and tavily-dynamic-search for programmatic filtering with context isolation. Do not use for named social/community/media/career/code platforms, login or cookie-dependent content, platform-native discussions, or cross-platform collection; use Agent Reach there. Do not use for local files, Git operations, deployments, or ordinary code editing.

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。

Extract clean markdown from any URL, including JavaScript-rendered SPAs. Use this skill whenever the user provides a URL and wants its content, says "scrape", "grab", "fetch", "pull", "get the page", "extract from this URL", or "read this webpage". Handles JS-rendered pages, multiple concurrent URLs, and returns LLM-optimized markdown. Use this instead of WebFetch for any webpage content extraction.

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。

GitHub Actions の CI 失敗を調査するスキル。失敗した run / PR / job の URL（例: https://github.com/OWNER/REPO/actions/runs/RUNID、…/pull/PR、…/job/JOBID）や run_id・PR 番号を貼られて「このCIが失敗している、原因を調査して」「ビルド/テストがコケた、なぜ落ちたか調べて」「workflow が failed、修正案を出して」「job のログを見て」と頼まれたら必ず使う。WebFetch では github.com のログは取れない（PreToolUse hook が差し戻し、HTML しか返らない）ため、gh（`gh run view --log-failed` / `gh run view --job` / `gh pr checks` / `gh api`）で失敗ジョブ・ステップと失敗ステップのログだけを決定論スクリプトで収集し、失敗原因を特定して修正案を提示する。GitHub Actions 以外（GitLab CI 等）や、URL がログではなくソース閲覧目的のときは対象外。GitHub(gh) 前提。`/gh-ci-investigate` で明示起動も可。

Calculate and schedule quarterly estimated tax payments (Form 1040-ES) for self-employed individuals. Covers self-employment tax calculation (15.3%), income tax estimation, safe harbor rules (100%/110% of prior year), payment deadlines, and underpayment penalties. Trigger on "estimated taxes", "quarterly taxes", "1040-ES", "how much should I pay quarterly", "self-employment tax", "avoid tax penalty".

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。

Create or refresh FriendlyBet Story of the World Cup image assets and app overlays. Use when generating new 9:16 World Cup story artwork, replacing story-assets PNGs, writing image-generation prompts, checking real player shirt numbers, placing Hebrew/English story captions, or validating that story text does not cover player faces or get cropped in WhatsApp previews.

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。

Train and fine-tune transformer language models using TRL (Transformers Reinforcement Learning). Supports SFT, DPO, GRPO, KTO, RLOO and Reward Model training via CLI commands.

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。

ClaudeDesign（claude.ai/design）とClaude Codeの連携。DesignSyncツールで デザインプロジェクトからプロトタイプHTML等を取得（インポート）、または ローカルのコンポーネントをデザインシステムプロジェクトへ同期（プッシュ）する。 「ClaudeDesignからデザインを取得して」「claude.ai/design のURLを取り込んで」 「デザインプロジェクトに同期して」「デザインシステムをプッシュして」 などのリクエスト、または claude.ai/design のURLが渡されたときに使用する。 取得したデザインのコード実装への適用は apply-design を使う。

Turn any URL — landing page, blog post, product page, documentation page, changelog, public PDF, or web article — into a finished AI-generated video using the Ozor MCP. Use this skill whenever the user gives a link and wants a video from its content: 'make a video from this URL', 'video from my landing page', 'turn this blog post into a video', 'video from https://...', 'video from our homepage', 'generate a video from this product page', 'create a video from this article'. This skill runs analyze_document with the url parameter (no scraping needed on our side), presents the resulting plan to the user, then generates → exports → embeds. Do NOT use this skill for a short text brief (use ozor-generate) or for an attached local file (use ozor-document-video).

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。

Performing security reviews of serverless functions across AWS Lambda, Azure Functions, and GCP Cloud Functions to identify overly permissive execution roles, insecure environment variables, injection vulnerabilities, and missing runtime protections.

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。

bu skill covers security hardening for serverless compute platforms including AWS Lambda, Azure Functions, and Google Cloud Functions. It addresses least privilege IAM roles, dependency vulnerability scanning, secrets management integration, input validation, function URL authentication, and runtime monitoring to protect against injection attacks, credential theft, and supply chain compromises.

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 URL paths and proposes, tests, orders, and applies project-level path cleaning rules so dynamic segments (numeric IDs, UUIDs, slugs, dates) collapse into readable aliases. Use when the user says "clean the paths", "normalize URLs", "group similar pages", "too many distinct paths", "/users/123 and /users/456 are the same page", "set up path cleaning", or asks why a Web analytics or Paths breakdown is fragmented across thousands of nearly-identical URLs. Covers regex syntax (re2), alias placeholder convention, rule ordering, the test workflow, and applying rules via the project-settings-update MCP tool.

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。

Tells you what CARLA on UE 5.5 (branch ue5-dev) cannot do that UE 5.8 (ue58-dev) can, so you do not attempt an unsupported workflow there — no Autoware integration, no ROS 2 middleware choice (FastDDS only), no DLSS or ray-traced-lens camera, no OFPA large-map mount so Town12/13/15 load black, and a few missing World/Actor methods. Also states what is identical, so the ue58 skills can be used directly for everything else. Use when working against a 5.5 checkout, when a ue58 skill's step has no counterpart, or when deciding whether a feature request is possible on 5.5 at all.

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。

Create, migrate, improve, or debug detailed vehicle assets for CARLA using Blender and Unreal, including reference-based geometry, materials, wheels, physics, lights, and runtime validation. Use for vehicle asset authoring rather than ordinary traffic spawning or map decoration.

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。

Answers "can CARLA do X here, and is there a vetted procedure for it" before you promise anything, and detects client/server/tree version skew — a support matrix over the whole collection (covered by a skill / works but undocumented / present but untested / broken or removed), an index of the features that are verified BROKEN on 0.10.0 (gbuffers crash the server, map layers do nothing, Landmark.waypoint is None, standalone asset packages are gone), and a probe that reports what this machine's build flags and running server actually offer. Use when the user asks whether CARLA supports something, when a feature has no skill, when a UE4 tutorial's steps do not exist, or before improvising a procedure.

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。

Adds a new topic to CARLA's native ROS 2 interface — a publisher (BasePublisher + PublisherImpl, topic suffix, QoS, registration in ROS2::GetOrCreateSensor, a ProcessDataFrom* entry point and its call site in the UE4 plugin) or a subscriber that lets ROS drive an actor. Also covers adding a middleware (RMW) behind IPublisherMiddleware. Use when the user asks to "publish X to ROS from CARLA", "add a ROS topic/publisher/subscriber", "make the obstacle/lane-invasion sensor publish", "add odometry to ROS", or "add another RMW".

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 pedestrians that wander autonomously via WalkerAIController, and destroys them in the correct order. Pairs each walker.pedestrian.* with a controller.ai.walker, places them on the navigation mesh, and sends them to random navmesh points at walking speed. Use when the user asks to "spawn pedestrians/walkers", "add people/crowds", "make pedestrians walk around", or "remove the walkers". Reads the world settings first: if the world is already synchronous it never ticks it, and if it is asynchronous it makes it synchronous and drives the clock itself.

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 Scenic, the probabilistic scenario description language, for the CARLA in use — pip installing it into the same interpreter that holds the CARLA client, optionally cloning the upstream repo for the example scenarios the wheel does not ship, and verifying that the `scenic` CLI, the client and Scenic's version-keyed blueprint table all agree. Use when the user asks to "install Scenic", "set up Scenic", "download Scenic", "pip install scenic", wants to run `.scenic` files, or has a Scenic whose model import or blueprint lookup fails.

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 a single ScenarioRunner Python scenario (or a whole group) against a running CARLA server — FollowLeadingVehicle, ControlLoss, DynamicObjectCrossing, SignalizedJunctionLeftTurn and the rest of srunner/examples — in sync or async mode, with result output, recording and manual/agent control of the ego. Lists what is runnable on the checked-out branch and on the loaded map. Use when the user asks to "run a scenario", "test FollowLeadingVehicle", "list scenarios", "run a CARLA traffic scenario", or reports a scenario that is "not supported".

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。

nuqs (Type-safe URL search params state manager for React / Next.js) リファレンス。 useQueryState, useQueryStates, parsers (parseAsString / parseAsInteger / parseAsBoolean / parseAsArrayOf 等)、 Next.js App Router / Pages Router、SSR、シャロールーティング、history 制御。 createSerializer, createStandardSchemaV1, inferParserType ユーティリティ、URL 長さ制限 (limits)。

visionOS / iOS 空間コンピューティングリファレンス。RealityKit、ARKit、visionOS SwiftUI API。 Entity, ModelEntity, AnchorEntity, Component, RealityView, MeshResource, Material, System, EntityAction, GaussianSplatComponent, ComputeGraphComponent, PhysicsJointsComponent, ForceEffectComponent, ClothBodyComponent, NavigationComponent, NavigationMeshComponent, NavigationController, ImmersiveSpace, ImmersionStyle, Model3D, WindowGroup, VolumetricWindowStyle, ObjectTrackingProvider, RoomTrackingProvider, ARSession, ARWorldTrackingConfiguration, ARKitSession, WorldTrackingProvider, HandTrackingProvider, SceneReconstructionProvider。