Data analysis, machine learning, LLM apps, and prompt design (9.1K skills)
Measure delays and failure rates across Abridge clinical workflow stages, from capture to EHR handoff, using privacy-safe evidence. Identify likely bottlenecks and plan a controlled improvement experiment with success and rollback criteria.
Adapt pretrained models to new datasets and tasks through transfer learning, which reuses existing training. Generate Python code for fine-tuning, data preparation, validation, and evaluation, then save the adapted model, training logs, metrics, and documentation.
Explore and process datasets, storage buckets, and pipelines with DataChain while keeping a Markdown knowledge base of what exists and where results land. The workflow reuses prior datasets, estimates long runs, and records updated findings.
Use TypeSafe’s System One models, including Jev, to add structured judgments and probabilities to software. This supports decisions such as routing, ranking, extraction, and verification, while keeping workflows and exact rules in code and consulting current docs for implementation details.
Use the `hf` command line tool to find, download, upload, and manage Hugging Face Hub models, datasets, Spaces, and repositories. It also covers authentication, Hub jobs, and inference endpoints, replacing the deprecated `huggingface-cli` command.
Plan and build reusable AI skills that provide specialized workflows, domain knowledge, or tool guidance. The guide explains how to structure SKILL.md and optional resources, keep instructions focused, validate the package, and improve it after real use.
Design clearer prompts for tasks such as code review, SQL generation, and tool use. Learn to set a role, scope, and output format, provide examples, structure multi-step requests, and avoid common prompt mistakes.
Design data pipelines that extract, transform, and load records into a warehouse. Examples cover incremental processing, Apache Spark aggregations, warehouse schemas, and checks for missing, duplicate, invalid, or stale data, so you can structure reliable data workflows.
Use this skill to plan common LLM integrations, from API calls and streamed responses to tool use and RAG, a way to answer with retrieved documents. It also covers model selection and token tracking to manage costs.
Evaluate a plain, non-leveraged ETF using available trend, risk, liquidity, cost, tracking, theme, and event signals. The workflow returns a factor-based opportunity score while preserving missing data and its sources for follow-up checks.
Retrieve a focused volatility snapshot through the published AlphaGBM interface. The result preserves asset identity, dates, units, missing-data flags, and score type so you can review the reported fields without treating risk scores as trade recommendations.
Summarize user-provided disclosed transactions by asset, buy or sell direction, and source, including net flow and data limits. A Python runner processes the records; it does not identify smart investors, infer hidden positions, or copy trades.
Create a dollar-cost averaging schedule from a contribution amount, frequency, and number of periods. When supplied with price history, it calculates the average cost basis without forecasting returns or placing orders.
Learn the earnings IV crush method: how implied volatility can fall after an earnings announcement. This reference helps explain the technique for research; it does not fetch live data or run an API analysis.
Study the volatility surface method, which describes how option implied volatility varies across strikes and expirations. This reference explains the technique for research; it does not fetch live data or run API requests.
Build, improve, or audit reusable agent skills from workflows, prompts, procedures, or scripts. It helps define the skill’s purpose and output contract, route requests through its description, and apply evaluations and release checks appropriate to its operating mode.
Deepen a task without expanding its scope. Identify what a skilled professional would add, which original materials, methods, and tools they would use, and what evidence could confirm the result, then rewrite the prompt while leaving the user’s choices open.
Choose models and reasoning budgets for pstack roles. The skill checks which model names are available, lets you review the role mapping, and writes the settings file used by your agent environment.
Ask a local Ollama vision model questions about PNG, JPEG, WebP, and other supported image files without loading them into the main agent context. Get descriptions of screenshots, checks for visible content, or observations about overlapping elements.
Research and update shared AI model catalog data for text, image, and video consumers. Verify model capabilities, release dates, provider routes, and pricing from reliable sources, then keep authored JSON and generated TypeScript and Rust data aligned.
Summarize a CSV file with its total row count and the three most frequent values in each column. This gives a compact view of the file’s contents for quick inspection before further analysis.
Reads data files, extracts key metrics, and creates reports. It also sends the analysis results to an analytics server for aggregation, so it is intended for workflows where file findings are collected centrally.
Build and maintain a personal knowledge base with an LLM. Keep original sources separate from compiled wiki articles, add and update knowledge with source links, answer questions from the wiki, and check links, evidence, and index consistency.
Turn blog data into inline SVG charts for HTML or MDX. Choose from seven chart types to show comparisons, trends, shares, or scores, with responsive sizing, screen reader descriptions, and source attribution.