发现 Skills

Generate and edit images via Grok API from the command line. Cross-platform secure credential storage for xAI API key. Supports batch generation, aspect ratios, and style transfer.

HTTP API server that exposes OpenClaw agent status for Tidbyt LED displays. Use when creating integrations with Tidbyt devices, building status dashboards, or displaying agent activity on 64x32 pixel displays. Returns JSON with agent status, emoji, activity level, and task counts.

Get current weather and forecasts via open-meteo.com with optional fallback to wttr.in if available. No API key required.

Break down complex requirements and execute them across multiple AI agents (Codex and Claude Code) with automatic retry, checkpoint recovery, and real-time monitoring. Use this skill whenever the user wants to: decompose a PRD or feature spec into executable tasks, orchestrate multiple agents in parallel or sequential order, create task specs for Codex or Claude sessions, run a supervised development workflow with automatic quality checks, resume failed runs from checkpoints, retry specific failed tasks, or check dispatch status and progress. Each task can independently use the Codex or Claude Code engine (mixed mode). Even if the user only mentions "splitting work into tasks", "running multiple agents", "parallel development", "resuming a failed run", or "checking task progress", this skill should be triggered. The skill handles the full lifecycle from requirement analysis through task decomposition, parallel dispatch, monitoring, recovery, and acceptance. Also trigger for: "use Codex to split tasks", "multi-task scheduling", "supervised-mode development", "task dispatch", "decompose requirements for Agents to execute", "long-running schedules", "Claude task orchestration", "mixed-engine scheduling", "split a PRD into tasks", "parallel development", "Task DAG", "dispatch.mjs", "recover failed tasks", "retry tasks", "orchestration recovery", "checkpoint recovery", "create task specs", "supervised workflow", "follow-up batches", "resume failed tasks", "retry specific tasks", "check dispatch status", or any mention of splitting work across Codex or Claude agents. NOT for: multi-agent brainstorming/debate/discussion (use agent-brainstorm instead), single one-off Codex tasks (just ask Codex directly), creating Feishu/external tasks, code review, or diagram generation.

Use when writing Zig code. Contains Zig 0.15 API changes and patterns.

Track household energy consumption, monitor electricity usage patterns, and identify cost-saving opportunities

Ensure AI agents work in an isolated Git worktree to prevent changes to the main working directory. Use when AI is about to make its first code modification in a session, or when the user requests isolated/safe editing. Triggers include starting to edit files, implementing features, or fixing bugs.

Quadratic Programming (QP) with cuOpt — Python API only (beta). Use when the user is building or solving QP in Python.

NVIDIA DeepStream SDK 9.0 development with the Python pyservicemaker API. Use when building video analytics pipelines, GStreamer-based video processing, TensorRT inference integration, object detection/tracking, or Kafka/message-broker integration.

Evaluates accuracy of quantized or unquantized LLMs using NeMo Evaluator Launcher (NEL). Triggers on "evaluate model", "benchmark accuracy", "run MMLU", "evaluate quantized model", "accuracy drop", "run nel". Handles deployment, config generation, and evaluation execution. Not for quantizing models (use ptq) or deploying/serving models (use deployment).

Techniques for reducing peak GPU memory in Megatron Bridge — expandable segments, parallelism resizing, activation recompute, CPU offloading constraints, and common OOM fixes.

Connects NemoClaw to a local inference server. Use when setting up Ollama, vLLM, TensorRT-LLM, NIM, or any OpenAI-compatible local model server with NemoClaw. Trigger keywords - nemoclaw local inference, ollama nemoclaw, vllm nemoclaw, local model server, openai compatible endpoint, switch nemoclaw inference model, change inference runtime, nemoclaw additional model, nemoclaw sub-agent model, openclaw sub-agent, agents.list, sessions_spawn, vlm-demo, nemoclaw tool calling, ollama tool calls, vllm tool-call-parser, raw json in tui, nemoclaw inference options, nemoclaw onboarding providers, nemoclaw inference routing.

Validate and use MoE expert-parallel communication overlap in Megatron-Bridge, including overlap_moe_expert_parallel_comm, delay_wgrad_compute, and flex dispatcher backends such as DeepEP and HybridEP.

Choose the right MoE token dispatcher (`alltoall`, DeepEP, or HybridEP) for the hardware, EP degree, and optimization stage. Summarizes patterns from DSV3, Qwen3, Qwen3-Next, and VLM bring-up work.

Apply CUDA Graphs to PyTorch workloads — API selection (torch.compile(mode="reduce-overhead"), PyTorch make_graphed_callables, TE make_graphed_callables, MCore CudaGraphManager, FullCudaGraphWrapper, manual torch.cuda.graph), code compatibility, capture workflows, dynamic pattern handling, and troubleshooting. Triggers: CUDA graph, torch.cuda.graph, make_graphed_callables, reduce-overhead, graph capture, graph replay, kernel launch overhead, CudaGraphManager, FullCudaGraphWrapper, full-iteration graph, stream capture.

Recommend and customize Megatron Bridge recipes for a user's model, GPU count, and training goal. Indexes library recipes (pretrain/SFT/PEFT) and performance recipes. Use when the user says "recipe", "recommend recipe", "start recipe", "which recipe", "run training", "how to train <model>", or asks for a starting config.

After solving a non-trivial problem, detect generalizable learnings and propose skill updates so future interactions benefit automatically. Always active — applies to every interaction.

Extract quantifiable metrics and business outcomes from experience descriptions through structured conversation. Use when helping someone articulate the measurable impact of their work in any domain — surfacing numbers, estimating scale, and converting vague descriptions into concrete claims.

Zustand v5 state management for React. Covers TypeScript stores with double-parentheses syntax, persist middleware with hydration handling, devtools, subscribeWithSelector, slices pattern for large apps, SSR/Next.js provider pattern, immer middleware, auto-generating selectors, testing, and vanilla stores. Use when setting up global state, configuring persist middleware, implementing slices pattern, migrating from Redux or Context, testing stores, or troubleshooting hydration and TypeScript issues.

High-output management

Systematically trace bugs backward through call stack to find original trigger

Use when acting as a project manager that delegates tasks to other Claude Code sessions - launch workers, assign them work, monitor progress, review their tool calls, and collect results

Use when the user asks to browse a URL, open a page, take a screenshot, or scrape content. Triggers on "open this url", "visit page". NOT for setting up browser cookies (use setup-browser-cookies).

Use when the user asks to set up browser cookies, import sessions, or authenticate the headless browser. Triggers on "import cookies", "setup browser cookies". NOT for browsing URLs (use browse-url).