duckduckgo-websearch

Search the web using DuckDuckGo. Use when the user wants to search the internet, find current information, look up facts, search for news, or find websites about specific topics. Triggers include requests like "search for X", "look up Y", "find information about Z", "what is X" (when X may be recent, niche, or unknown), "tell me about Y", "what's the latest on", "news about", or any query requiring up-to-date web information. Also use when the user asks about a specific named entity (product, company, tool, person, event) that might not be in the training data.

Tuning Engines

Domain-specific LLM fine-tuning — sovereign models trained on your data, zero infrastructure.

/train — Launch a PPO Training Run

Parse the user's request from `$ARGUMENTS` and construct a training command.

trl-training

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.

strava-sync

Downloads and syncs training data from Strava using the Strava MCP server. This is ONE option for importing training data - athletes can also use manual logs, other platforms (Garmin, Apple Health), or provide data conversationally. Use this skill specifically when the athlete wants to sync from Strava or when the running-coach indicates Strava as the chosen data source. Stores activities in the training-log folder as markdown summaries, including lap details for workout runs.

agents-sdk

Build AI agents on Cloudflare Workers using the Agents SDK. Load when creating stateful agents, durable workflows, real-time WebSocket apps, scheduled tasks, MCP servers, or chat applications. Covers Agent class, state management, callable RPC, Workflows integration, and React hooks. Biases towards retrieval from Cloudflare docs over pre-trained knowledge.

ai-psychosis-mode

True alias for reality-check-mode. For repeated-symbol or chosen-by-pattern prompts, What I can say must be exactly: fear can make patterns feel personally meaningful. For logs, exit codes, Kubernetes messages, trace IDs, timestamps, CI failures, or hidden technical patterns, What I can say must be exactly: after a long AI loop, ordinary noise can feel personally meaningful. Use only Grounding, What I can say, and Safer next step. Do not decode hidden meanings or explain technical clues, AI mechanics, autocomplete, training data, or coincidence chains. Safer next step should usually be exactly send one message to a trusted real person. No bullets, numbered lists, message drafts, or pattern analysis.

synalinks

Build neuro-symbolic LLM applications with Synalinks framework. Use when working with DataModel, Program, Generator, Module, training LLM pipelines, in-context learning, structured output, JSON operators, Branch/Decision control flow, FunctionCallingAgent, RAG/KAG, or Keras-like LLM workflows.

remote-training

Manages remote training infrastructure on Nebius VMs. Use for building/pushing Docker images, starting/stopping VM machines (train, train2, train3), running training jobs, dataset generation, and starting inference servers.

rnow-cli

Use the ReinforceNow CLI for RLHF training. Use when running rnow commands, initializing projects, submitting training runs, testing rollouts, or downloading models.

Agent Dev — Vault-First Internal Development

Develop the agent's own internals with the vault as the primary source of truth. The vault knows more about the agent than any code scan or model training data. Always search the vault first, extract maximum context, and only then touch code.

Get API Docs via chub

When you need documentation for a library or API, fetch it with the `chub` CLI rather than guessing from training data. This gives you the current, correct API.

strategy-consulting-visualization

Use when turning any content into clear, professional visualizations - board slides, reports, proposals, research summaries, training materials, technical diagrams, infographics, process flows, timelines, benchmarks, waterfall charts, or data-backed visual specs for any audience.

training-hub-guide

Guides users through LLM post-training with Training Hub, including installation, algorithm selection (SFT, OSFT, LoRA), hyperparameter tuning, troubleshooting OOM errors, interpreting loss curves, and leveraging backend-specific features. Use when the user is working with training_hub, fine-tuning language models, asking about SFT/OSFT/LoRA training, or debugging GPU/CUDA training issues.

agent-init-training-playbook

add-benchmark

Guide for adding a new benchmark or training environment to NeMo-Gym. Use when the user asks to add, create, or integrate a benchmark, evaluation, training environment, or resources server into NeMo-Gym. Also use when wrapping an existing 3rd-party benchmark library. Covers the full workflow: data preparation, resources server implementation, agent wiring, YAML config, testing, and reward profiling (baselining). Triggered by: "add benchmark", "new resources server", "integrate benchmark", "wrap benchmark", "add training environment", "add eval".

accessibility-specialist-role

Operate as an accessibility specialist who audits against WCAG, trains teams to stop shipping the same defects, and owns the compliance sign-off. Use when a product needs to be usable with a keyboard, a screen reader, or magnification and someone must certify it before launch.

Ultrarunning training companion agent powered by Intervals.icu.

Modify upcoming workouts based on how you're feeling or schedule changes

brev-cli

Manage GPU and CPU cloud instances with the Brev CLI for ML workloads and general compute. Use when users want to create instances, search for GPUs or CPUs, SSH into instances, open editors, copy files, port forward, manage organizations, or work with cloud compute. Supports fine-tuning, reinforcement learning, training, inference, batch processing, and other ML/AI workloads. Trigger keywords - brev, gpu, cpu, instance, create instance, ssh, vram, vcpu, A100, H100, cloud gpu, cloud cpu, remote machine, finetune, fine-tune, RL, RLHF, training, inference, deploy model, serve model, batch job.

claude-plan — independent plan from Claude's plan mode

The host model planning its own work shares its own blind spots. A different-vendor model (Anthropic Claude, pinned to Fable) has a different training distribution, so where its plan diverges from yours is exactly where to dig. This skill runs the Claude Code CLI (`claude -p`) in Claude's **native p

browse-environments

Discover and inspect verifiers environments through the Prime ecosystem. Use when asked to find environments on the Hub, compare options, inspect metadata, check action status, pull local copies for inspection, or choose environment starting points before evaluation, training, or migration work.

ai-ml-development

AI and machine learning development with PyTorch, TensorFlow, and LLM integration. Use when building ML models, training pipelines, fine-tuning LLMs, or implementing AI features.

experiment-execution

Execute AI/ML experiments locally or remotely with environment, runtime, and logging controls. Prefer invoking via research-workflow. TRIGGER when: user asks to run/launch/start/resume/monitor a training job, evaluation, or benchmark, or a plan is ready for execution, or experiment needs rerun/recovery. DO NOT TRIGGER when: research investigation only (use deep-research), planning without execution (use research-plan), or env setup without launch (use project-context).

3dgs-code-reviewer

Review 3D Gaussian Splatting implementation code for correctness, performance bugs, and best practices. Covers CUDA kernels, rendering pipeline, training loop, loss functions, and common pitfalls. Detects 42+ known bug patterns.

Holomime

Behavioral therapy for AI agents — self-diagnosis, alignment, and training via MCP

book-sft-pipeline

This skill should be used when the user asks to "fine-tune on books", "create SFT dataset", "train style model", "extract ePub text", or mentions style transfer, LoRA training, book segmentation, or author voice replication.

VLA Remote Train Eval

Synthetic Data Generation

Generate synthetic data, run a flow, create training data, produce datasets, or author custom flow YAMLs using sdg_hub

brandkit

Premium brand-kit image generation skill for creating high-end brand-guidelines boards, logo systems, identity decks, and visual-world presentations. Trained for minimalist, cinematic, editorial, dark-tech, luxury, cultural, security, gaming, developer-tool, and consumer-app brand systems. Optimized for intentional logo concepting, refined composition, sparse typography, strong symbolic meaning, premium mockups, art-directed imagery, and flexible grid layouts.

Sidearm

Protect media from AI training, detect AI-generated content, and find stolen work.

lucid-api

Fetch live API references instead of relying on training data

Terradev

Complete GPU infrastructure for Claude Code — 192 MCP tools for provisioning, training, inference

Intelligence Aeternum

AI training dataset marketplace: 2M+ museum artworks with Golden Codex enrichment

add-dynamic-filter

Guide for adding dynamic/filter hooks in slime rollout pipeline. Use when user wants sample-group selection during rollout, buffer filtering before training, or per-sample masking/processing hooks.