Skills for your AI
Builds production-ready Model Context Protocol servers in C# using the ModelContextProtocol SDK, dependency injection, and async patterns. Use when creating, implementing, debugging, optimizing, integrating, or testing a C# MCP server with tools, prompts, or r
Handles data entry tasks end to end — extraction, cleaning, formatting, validation, migration, integration, categorization, enrichment, analysis, and automated entry from emails, forms, documents, images, audio, spreadsheets, web pages, and business cards — wi
Generates Ray Data code for loading, transforming, batch inference, writing, optimization, and ML framework integration. Use when the user needs to read data into a Ray Dataset, preprocess it, run a trained model over it, save results, tune a slow or large pip
Audit datasets for completeness, consistency, accuracy, and validity. Profile data distributions, detect anomalies and outliers, surface structural issues, and produce an actionable remediation plan. Use when the user asks to check data quality, profile a data
Loads, featurizes, splits, trains, and evaluates molecular datasets with DeepChem for drug discovery and materials science, and predicts properties for new SMILES. Use when the user provides molecular data files, asks to featurize molecules, split data, train
Turns raw defect data from logs, chats, and tracking systems into categorized, prioritized analyses and reports for QA managers. Use when the user needs defect intake and severity assessment, root cause analysis, trend analysis, prioritization, resolution stat
Converts plain-English diagram requests into blueprint-style SVG files by building a JSON spec and rendering it with the diagrammer CLI. Use when the user asks for an SVG architecture sketch, request flow, state machine, neural net, or pipeline diagram for doc
Chain document operations into reusable pipelines
Produces mergekit YAML configurations, run commands, method recommendations, and validation for combining fine-tuned language models. Use when the user wants to merge two or more HuggingFace models, pick a merge method, validate a merge config, or review past
Applies one-shot LLM pruning techniques (Wanda, SparseGPT, N:M, magnitude, structured) to compress models and speed up inference without retraining. Use when the user wants to reduce model size, hit a target sparsity, or compare sparsity patterns for specific
Implements and compares speculative decoding methods (draft model, Medusa, lookahead decoding) to accelerate LLM inference, measuring real speedups. Use when the user wants to speed up LLM generation, run or benchmark draft model speculative decoding, Medusa,
Index of 37 advanced engineering agent skills for Claude Code, Codex, Gemini CLI, Cursor, OpenClaw. Use when browsing or choosing among the POWERFUL-tier engineering skills: agent design, RAG, MCP servers, CI/CD, database design, observability, security auditi
Validates episode payloads, routes them through a configured sequence of specialized agents, and returns a consolidated JSON draft. Use when an episode payload needs validation, multi-agent dispatch, consolidated JSON output, or duplicate-run checking.
Generates and analyzes proteins with ESM3 and ESM C models for sequence completion, structure prediction, inverse folding, embedding extraction, function-conditioned design, chain-of-thought refinement, and batch processing. Use when the user provides protein
Evaluate and rank agent results by metric or LLM judge for an AgentHub session. Use when the user runs /hub:eval or asks to score, compare, or pick a winner among completed AgentHub agents.
Builds the Phase 0 evaluation harness for fine-tuning projects—traces, goldens, per-bucket graders, judge calibration, drift suite, and base-model baseline. Use when starting a fine-tuning effort, defining failure buckets, writing graders, calibrating LLM judg
Extracts, creates, modifies, and compares Excalidraw diagrams by delegating all file reads and writes to subagents so verbose JSON never enters the main context. Use when the user asks to explain, summarize, add to, change, create, or compare .excalidraw or .e
This skill enables advanced Excel automation using xlwings - a library that can interact with live Excel instances. Unlike openpyxl (file-only), xlwings can control Excel in real-time, execute VBA, update dashboards, and automate complex workflows.
Builds, configures, and deploys production-ready MCP servers in Python with FastMCP 3.0, covering scaffolding, tools, resources, auth, middleware, providers, deployment, and 2.x upgrades. Use when the user asks to create a new MCP server, add a tool, resource,
Guides Axolotl fine-tuning with YAML configs, training method selection, debugging, and advanced features. Use when configuring Axolotl training, choosing LoRA/QLoRA/DPO/KTO/ORPO/GRPO, debugging NCCL or FSDP errors, or handling long sequences and custom integr
Guides fine-tuning of LLMs with LLaMA-Factory WebUI, covering model selection, QLoRA bit levels, dataset formatting, debugging, and export. Use when the user wants to configure a fine-tuning job, prepare text or multimodal data, fix training errors, export or
Routes fine-tuning decisions to the right method (SFT, DPO/ORPO/KTO, GRPO/RLVR, continued pretraining) and base-model size class, after checking off-ramps and eval readiness. Use when someone asks whether to fine-tune, which fine-tuning method fits their data,
Generates PEFT fine-tuning code and configuration (LoRA, QLoRA, adapter loading/merging, method and parameter selection) for LLMs from 7B to 70B on limited GPU memory. Use when the user wants to fine-tune a model with minimal memory, asks which PEFT method or
Provides step-by-step guidance for fine-tuning large language models with Unsloth, covering setup, training scripts, speed and memory optimization, and debugging. Use when a user wants to fine-tune an LLM with Unsloth, asks for a training script, hits slow tra
