This is a public, open-source repository. Every commit, file, and message is visible to anyone. - 28 skills, all NHL/hockey focused - MCP server at mcp.puckapi.com/mcp - Data starts from the 2010-11 season. 16 completed seasons plus the one in progress. Do not claim 2008-09; it is not there - 32 active NHL teams (includes Utah Hockey Club, not Arizona Coyotes) - Odds from the books PuckAPI tracks (DraftKings, FanDuel, BetMGM, ESPN BET and others). The count moves as books…
Teaches how to use Claude and MCP tools effectively for hockey analytics -- exploratory analysis, hypothesis testing, model iteration, and report generation. Use when user asks how to analyze hockey data with AI, how to structure an analysis session, how to test a hypothesis about team or player performance, how to improve a model systematically, or how to generate a performance report. Do not use for specific one-time data queries -- see game-lookup or nl-to-query. Do not use for the mechanics of model building -- see feature-engineering or model-building directly.
Walk-forward historical testing of betting strategies to determine whether a model's edge is real. Use when user asks about backtesting, historical simulation, ROI over a season, whether a strategy works, bootstrap confidence intervals, edge compression, or model audit over time. Refuses in-sample testing -- walk-forward only. Do not use for live bet tracking -- see bet-tracker. Do not use for finding edges on today's slate -- see edge-detection. Do not use for model training -- see model-building.
Track predictions vs actuals in production: log bets, compute CLV, monitor ROI, detect edge decay, and run significance tests. Use when user asks about tracking bets, bet logging, closing line value, CLV, ROI tracking, win rate, drawdown monitoring, or whether their model has an edge right now. Do not use for finding today's bets -- see daily-card or edge-detection. Do not use for historical backtesting before live betting -- see backtesting. Do not use for model training or retraining -- see model-building.
Full slate analysis for tonight's games with edge rankings, odds across books, and recommended stakes. Use when user asks about tonight's slate, best bets tonight, daily card, full slate, tonight's games, which games are worth betting, or wants all games ranked by edge. Do not use for a single game breakdown -- see game-preview. Do not use for historical backtesting -- see backtesting. Do not use for tracking bets placed -- see bet-tracker.
Builds automated sports analytics pipelines -- daily data pulls, scheduled prediction generation, model versioning, prediction tracking, and drift alerting. Use when user asks how to automate their model, how to run predictions every morning, how to set up a cron job for sports data, how to track model performance over time, or how to version their model. Do not use for one-time data pulls -- use game-lookup or puckapi-tool directly. Do not use for building the model itself -- see model-building. Do not use for backtesting historical performance -- see backtesting.
Routes hockey analytics requests to the right 2-3 skills. Load this FIRST when a hockey data or analytics request comes in and you are unsure which skills to activate. Prevents loading all 28 skills when only 2-3 are needed. Also handles first-use persona detection.
Compares model probabilities against market odds to find positive expected value bets. Use when user asks about EV calculation, edge magnitude, Kelly criterion, bankroll sizing, CLV tracking, closing line value, fractional Kelly, simultaneous bets, or which bets are worth placing. Do not use for odds exploration without a model -- see odds-explorer. Do not use for validating model accuracy -- see backtesting. Do not use for converting odds formats or computing implied probability -- see odds-analysis.
Builds multi-variant Elo rating systems for sports prediction. Use when user asks about Elo ratings, rating systems, team strength, K-factor, home field advantage, season carryover, margin-of-victory adjustments, or mentions 'Fading Elo', 'Form Elo', 'Component Elo'. Do not use for general feature construction -- see feature-engineering instead. Do not use for team stats without ratings -- see team-analysis instead.
Transforms raw game, player, and goalie data into model-ready features with built-in leakage detection. Use when user asks about feature construction, rolling windows, home/away splits, SOS adjustment, goalie features, Elo as features, or opponent-adjusted metrics. Strictly enforces .shift(1) before any rolling calculation. Do not use for raw data exploration -- see game-lookup or team-analysis. Do not use for understanding hockey metrics -- see hockey-analytics. Do not use for xG feature construction -- see xg-model-building.
Finds games by date, team, matchup, or season -- past results, tonight's slate, upcoming schedule, head-to-head history. Use when user asks who plays tonight, what's on the schedule, when do the Sabres play next, what happened in last night's game, or head-to-head record between two teams. Do not use for team performance evaluation -- see team-analysis. Do not use for odds or betting lines -- see odds-explorer. Do not use for player stats -- see player-scouting.
Complete pre-game analysis for a single NHL (or other sport) matchup: team records, goalie matchup, key stats, odds snapshot, model edge, and head-to-head history in one shareable report. Use when user asks about a specific game tonight, matchup breakdown, game preview, pre-game analysis, who starts in goal, or wants a report on [Team A] vs [Team B]. Do not use for full slate analysis -- see daily-card. Do not use for odds-only lookup -- see odds-explorer. Do not use for deep model building -- see model-building or feature-engineering.
Goalie-specific analysis for NHL: leaderboard rankings, workload tracking, starter identification, tandem splits, and matchup history. Returns xG-adjusted metrics -- GSAX, high-danger save%, rolling form -- straight from the API, which predict future performance better than raw save percentage. Use when user asks about goalie stats, save percentage, GAA, quality starts, goalie fatigue, back-to-back starts, or which goalie is starting tonight. Do not use for skater stats -- see player-scouting. Do not use for team-level shot metrics -- see team-analysis. Do not use for building goalie model features -- see feature-engineering.
Teaches hockey analytics metrics -- Corsi, Fenwick, PDO, xG, zone entries, high-danger chances, RAPM, WAR -- with real data context. Use when user asks what Corsi means, how to interpret PDO, what xG tells you, why Fenwick matters, or asks to explain advanced stats for a team or player. Do not use for betting concepts or odds interpretation -- see odds-analysis. Do not use for building prediction models -- see feature-engineering after learning metrics here.
Trains and validates prediction models for sports game outcomes. Use when user asks about building a prediction model, training a classifier, choosing between logistic regression or XGBoost, model selection, hyperparameter tuning, feature importance, or ensemble methods. Always uses walk-forward methodology -- refuses k-fold. Do not use for feature construction -- see feature-engineering. Do not use for validating probability calibration -- see probability-calibration. Do not use for xG models specifically -- see xg-model-building.
Translates natural language hockey queries into structured data filters and executes them via puckapi-tool. Use when user asks a research question in plain English: 'show me games where the home team was outshot but won', 'find all back-to-back losses', 'how often do goalies with 2 days rest outperform their season SV%'. Includes a self-correction loop -- retries with adjusted parameters if the first query returns unexpected results. Do not use for specific known queries where the right tool is obvious -- just call game-lookup, team-analysis, or goalie-analysis directly. Do not use for model building -- see feature-engineering.
Converts betting odds between formats and removes bookmaker margin to find true implied probabilities. Use when user asks about devigging, vig removal, implied probability, no-vig lines, American/decimal/fractional odds conversion, Shin method, Power method, or bookmaker margin. Prerequisite for edge detection. Do not use for viewing current odds -- see odds-explorer instead. Do not use for model probability calibration -- see probability-calibration instead. Do not use for finding edges -- see edge-detection instead.
Multi-book odds comparison, best price identification, vig calculation, line shopping, and line movement analysis for NHL. Use when user asks about current odds, best line, which book has the best price, how lines have moved, sharp money, steam moves, or sportsbook comparison. Do not use for devigging or implied probability math -- see odds-analysis. Do not use for comparing model output vs market -- see edge-detection. Do not use for game results -- see game-lookup.
Searches players and surfaces season stats, per-game rates, multi-season comparisons, roster snapshots, and prospect NHLe translations. Use when user asks how a player is performing, wants to compare two skaters, needs a team's top scorers, is evaluating a prospect from another league, or needs stats for fantasy or DFS. Do not use for goalie stats -- see goalie-analysis. Do not use for team-level analysis -- see team-analysis. Do not use for building WAR or GAR -- see war-gar-decomposition.
Monte Carlo playoff and season simulator for NHL. Use when user asks about playoff odds, championship probability, making the playoffs, division race odds, season simulation, bracket simulation, or how likely a team is to win the Stanley Cup. Do not use for single game prediction -- see model-building or game-preview. Do not use for team stats without simulation -- see team-analysis. Do not use for player-level analysis -- see player-scouting.
Plain text files in a repository that tell a coding agent how the project works: commands to run, conventions to follow and things to avoid. CLAUDE.md, AGENTS.md, cursor rules and skills are the common kinds.
CLAUDE.md or AGENTS.md?
CLAUDE.md is read by Claude Code. AGENTS.md is an open format that Codex, Cursor and other agents read. Many projects keep one and point the other at it.
What is a skill?
A folder with a SKILL.md that describes one capability, such as filling PDFs or reviewing code. The agent loads it only when the task calls for it.
Can I search my own team's files too?
Your agents already can, over MCP, limited to the files you're allowed to read. Searching them from this page is coming.
PuckAPICLAUDE.md