Weights & Biases Skills
Official W&B agent skills for model training and agentic app observability.
About
Weights & Biases Skills is the official agent-skills repository from Weights & Biases. It guides Claude Code, Codex, and other coding agents on using the W&B AI developer platform to train models and build agents: logging metrics and rich media during training, tracking and analyzing experiments, tuning hyperparameters, tracing agentic applications, classifying trace failure modes, and running evaluations.
The repo is Apache-2.0 licensed and publishes benchmark badges showing how well major coding agents perform with the skills installed. Install with npx skills add wandb/skills, then set a WANDB_API_KEY.
Honest limits: the skills assume a W&B account and API key, so they are only useful inside the W&B ecosystem; the repo is smaller and newer than the biggest vendor skill collections.
What it can do
- Training metric and media logging
- Experiment tracking and analysis
- Hyperparameter tuning guidance
- Agentic app tracing with Weave
- Model evaluation with labeled datasets
Setup
npx skills add wandb/skills, then export WANDB_API_KEY from wandb.ai/authorize.
Prerequisites
A Weights & Biases account and API key.
Use cases
Works with
Trust score — 6/10
Scored 6/10 (metrics tier, metrics-v1). Evidence: Verified via vendor site/docs/GitHub and reputable coverage, wave 87; Trending score 64/100 from researcher assessment.
★ 69 GitHub stars
Sources
Every fact above was checked against these sources. Where a claim comes from the provider and couldn't be independently verified, the text says so.