1.1K skills · sorted by relevance
Track machine learning experiments with SwanLab by recording training settings, metrics, and media such as images and audio. Learn to compare runs, inspect training charts, and connect existing training code to local, cloud, or self-hosted dashboards.
Learn how GPT language models work by preparing text, training a small model, and generating samples. Covers CPU experiments, custom datasets, further training of pretrained GPT-2 models, and reproducing GPT-2 with multiple GPUs.
Evaluate language models on knowledge, math, coding, safety, and image understanding benchmarks. The skill guides reproducible runs in containers using local Docker, Slurm clusters, or cloud execution, with shared configurations for model comparison and result export.
Set up reinforcement learning, training guided by rewards, for language models with slime. The skill covers Megatron-LM training, SGLang response generation, data preparation, asynchronous execution, and custom workflows for agents that use tools across multiple turns.
Evaluate Hugging Face Hub models using inspect-ai or lighteval, with local inference or a provider-backed option. Choose an inference engine, start with a small test, and switch engines when model compatibility requires it.
Learn how to organize an AI agent skill using a sample SKILL.md file. It covers basic metadata, procedural instructions, supporting folders, and relative paths, and serves as an example for testing the OpenSkills loader.
Set up Convex snapshot backups and test recovery by restoring into a temporary environment. Verify table counts and sample records, then define a backup schedule, retention period, and production recovery instructions based on acceptable data loss.
Tests a specific Convex feature with sample data and simulated users, without deployment. It checks expected results, rejects unauthorized callers, and verifies that queries return only data belonging to the current user.
Create a Notion-style team dashboard that brings together key metrics, seven-day trend lines, an activity feed, and a task table. When no data source is available, sample data fills out the dashboard.
Build a new agent skill from a structured template. Copy the directory and edit SKILL.md to document its purpose, inputs, outputs, processing steps, precautions, and examples, replacing placeholders with concrete content.
Generate character images, backgrounds, synthetic speech, and music for product videos featuring 小省导购员, a digital shopping presenter. Coordinate these assets with InfiniteTalk to produce a 25-second vertical video made from five scenes.
Create a temporary workbench for one existing component, using realistic mock data and a grouped state switcher. View loading, empty, error, access, and other implemented states individually, with a link to each state.
Audit an Excel workbook before editing, then verify the saved file with Python and openpyxl. Each requested requirement receives an explicit status, so completion depends on checks of the actual file.
Check which H.264 encoders, the components that compress video, are available in FFmpeg before building a conversion script. Choose an encoding approach and test a small sample before processing long videos or batches.
Extract edit times from documents, inspect audio for clicks and level issues, and guide production of stems, or separate audio parts. Each stage includes checks, ending with a mixed export verified for duration, format, and audio levels.
Recover from failed Excel creation or editing by switching from execute_code_sandbox to shell_agent. It guides clear task descriptions so the agent can install missing libraries, run code, retry errors, and confirm file creation.
Investigate Python failures in a sandbox, an isolated execution environment, by capturing partial logs, testing functions separately, and checking output files one at a time. Use the findings to identify failing steps and rebuild a working script.
Set up libFuzzer to test C/C++ code with varied inputs guided by which code paths run. Build test harnesses, manage sample inputs and dictionaries, enable error detectors, and reproduce crashes or investigate stalled coverage.
Start shotgun metagenomics workflows for mixed-community DNA reads in FASTQ files. Check read quality, remove host sequences, and profile organisms and biological functions using supported tools and databases, producing tables and visual summaries when results are available.
Validate FASTQ sequencing files, run local quality checks, and interpret reports to decide whether trimming is needed. Optionally remove adapters or low-quality read ends while preserving original reads and their quality reports.
Plan DNA variant detection workflows by identifying the assay and sample model, then routing to germline, cancer somatic, or molecular-barcode panel skills. It also supports focused local BAM/CRAM checks with variant files, depth summaries, and optional filters.
Prepare single-cell or single-nucleus RNA-seq analysis by checking inputs and selecting workflows that turn FASTQ sequencing files into count matrices. Route quality checks, cell annotation, clustering, and UMAP visualization to the embedded downstream skill.
Route data analysis requests to focused workflows for metric checks, business decisions, charts, dashboards, and reports. Identify usable data sources and guide the next step, with delivery choices based on the available environment.
Prepare images and videos for social platforms using Adobe tools. Create image crops that keep the subject in view, expand backgrounds when available, preview a full-set sample, and resize videos while preserving their aspect ratio.