SAS Sports Analytic Skills
FREE · OPEN SOURCE · AGENT SKILLS

Sports analytics skills your agent can actually run.

Install the pack, point an agent at your data, and get focused guidance for EDA, time-safe features, baselines, walk-forward validation, leakage checks, calibration, and honest reporting.

Install the pack Read the docs →
24standalone skills
NFL · NBA · MLBpublic data paths
$0price forever
MITopen source
WHAT YOU GET

A jumpstart for serious sports modeling.

Not a black-box picks engine. Operator manuals your agent can follow — on data you own.

Install into any agent

Works with Claude Code, Cursor, and agents that support the Agent Skills standard. One command. No clone required for skill-only use.

Point it at your data

Skills consume CSV, Parquet, or JSON you already have. Each skill documents the grain and fields it needs.

Time-safe by default

Decision-time legality, walk-forward validation, leakage audits, and calibration checks are first-class.

Baselines before complexity

Constant rates, home-field, and simple statistical baselines come before fancy models.

Public data paths

Skills for nflverse, sportsdataverse, and pybaseball. Optional sports_ds toolkit if you want prebuilt loaders.

Honest reporting

Model cards, experiment logs, miss analysis, sample sizes, uncertainty, and limits. Built to kill sloppy analytics.

COMPOSABLE PATH

Use only what the task needs.

Skills are independently invocable. Chain them when the question demands it.

READ THIS FIRST

What this is / is not.

Scaffolding so you can learn the workflow and build your own work. No claimed betting edge. No guaranteed profit. No finished market models.

This is

  • A free jumpstart for sports analytics with agents
  • Operator manuals for EDA, features, models, validation
  • A way to learn leakage, calibration, and honest reporting
  • Optional public-data helpers for NFL, NBA, MLB
  • Yours to fork, extend, and break on your own data

This is not

  • A betting tip service or guaranteed edge engine
  • A black-box model that prints money
  • Custom sport/market build-on-request support
  • A hosted dashboard or pick seller
  • Permission to skip learning how validation works
24 SKILLS

The full pack.

Browse here. Full operator manuals live in each skill’s SKILL.md. Docs skill index →

Foundation & data
sports-modeling-doctrineLock question, target, decision time, baseline
environment-setupPortable analysis environment
data-sourcesChoose public source and grain
nflreadpyLoad NFL data from nflverse
sportsdataverse-pyMulti-sport public data
pybaseballStatcast and MLB tables
sports-ds-bridgeOptional sports_ds toolkit bridge
Exploration & features
eda-sportsCoverage, grain, missingness, red flags
sports-visualizationHonest charts with uncertainty
anti-slop-analyticsKill chartjunk and bad claims
feature-rulesDecision-time-legal features
time-series-sportsShifted rolling and EWMA form
ratings-strength-modelsAs-of Elo and strength ratings
Modeling & validation
baseline-modelsConstant, home, simple baselines
statistical-modelingGLMs, diagnostics, uncertainty
predictive-modelingModels under honest time splits
validation-designWalk-forward folds and metrics
leakage-auditLook-ahead and join leakage
calibration-checkProbability reliability
simulation-sportsSeason and matchup simulation
Interpretation & reporting
model-interpretationDrivers, slices, largest misses
results-reportingReproducible results and limits
model-cardDurable model contract
experiment-logReproducible experiment history
GET STARTED

One command.

No repo clone required for skill-only use.

# Interactive
npx skills add WalrusQuant/sports-analytic-skills

# One skill
npx skills add WalrusQuant/sports-analytic-skills --skill eda-sports -y

# Entire pack
npx skills add WalrusQuant/sports-analytic-skills --all

Full getting-started guide →

OPEN SOURCE RESEARCH SOFTWARE

Learn the craft. Build your own.

View the repository

This is a free jumpstart for sports analytics workflows — not a betting product and not a claim of edge.