machina-sports/sports-skills

• **Archetype:** domain-skill-pack
• **Stars:** 66
• **Last pushed:** 2026-04-12
• **License:** MIT
• **Discovered:** 2026-04-12
• **Source**: gh-search

Summary

Open-source sports data and prediction market skills. 19 SKILL.md files covering 14+ sports (NFL, NBA, MLB, NHL, WNBA, tennis, golf, F1, college sports, volleyball, football/soccer across 13 leagues) plus prediction markets (Kalshi, Polymarket) and betting analytics. Published as both npm (npx skills add) and Python (pip install sports-skills) packages.

Key differentiators: zero API keys required (wraps public ESPN, FPL, Understat, Transfermarkt, Kalshi, Polymarket APIs), includes a markets orchestration skill that bridges live ESPN schedules with prediction market odds, a betting analysis module (pure computation: odds conversion, de-vigging, Kelly criterion, arbitrage detection), and a sports reporter skill for generating journalism from live data.

Includes automated nightly health checks that test all upstream endpoints and generate structured reports with GitHub issue auto-filing on failures.

Assessment

Moderate extractable value. The betting analysis workflow (fetch odds -> de-vig -> compute edge -> Kelly sizing -> arbitrage scan) is a clear multi-phase analytical pipeline. The markets orchestration (ESPN schedule -> match to prediction markets -> normalize prices -> evaluate) is another. The nightly health check for upstream API monitoring is a reusable process pattern.

The skills themselves are primarily data-access wrappers (API reference + CLI commands) with limited multi-phase logic, but the orchestration and analysis layers on top are genuinely novel.

Extraction Priority

**MEDIUM** - The betting/markets analysis pipeline and API health monitoring process are extractable. The data-access skills themselves are reference material rather than processes.

---

Processes

1. Sports Betting Edge Analysis Pipeline

**Placement:** specializations/sports-analytics/betting-edge-analysis.js

Multi-phase analytical pipeline for evaluating betting opportunities across sportsbooks and prediction markets.

**Phases:** 1. **Data Collection** - Fetch odds from multiple sources: ESPN (American odds), Kalshi (integer probabilities), Polymarket (decimal probabilities) 2. **Normalization** - Convert all odds to implied probabilities. De-vig sportsbook lines to extract fair probabilities (remove the house edge) 3. **Edge Detection** - Compare fair probabilities against prediction market prices. Calculate edge percentage and expected value per dollar 4. **Sizing & Arbitrage** - Apply Kelly criterion for optimal bet sizing. Scan cross-platform price combinations for arbitrage (total implied probability < 1.0) 5. **Report** - Present findings with fair probability, edge, EV, Kelly fraction, arbitrage ROI if found, and line movement classification

**Tasks:**

• fetch-odds (node) - collect odds from configured sources for a given event
• normalize-devig (node) - convert formats, remove vig, compute fair probabilities
• compute-edge (node) - compare fair vs market, calculate EV and Kelly
• detect-arbitrage (node) - cross-platform arbitrage scan
• generate-report (node) - synthesize findings into actionable output

2. Upstream API Health Monitor

**Placement:** specializations/shared/api-health-monitor.js

Automated health monitoring process for multi-source data pipelines. Extracted from sports-skills' nightly health check but generalizable to any system that depends on multiple external APIs.

**Phases:** 1. **Endpoint Discovery** - Enumerate all configured upstream endpoints with expected response characteristics 2. **Parallel Probe** - Hit each endpoint with timeout, measure latency, validate response structure 3. **Classification** - Categorize each source as OK (responsive + valid), Degraded (responsive but slow > threshold), or Down (unreachable / invalid response) 4. **Report Generation** - Produce structured JSON results + human-readable markdown summary 5. **Alert** - If failures detected, generate GitHub issue body for auto-filing

**Tasks:**

• enumerate-endpoints (node) - load endpoint config, return probe list
• probe-endpoint (node) - HTTP request with timeout, return latency + status + body sample
• classify-results (node) - apply thresholds, categorize health status
• generate-health-report (node) - structured JSON + markdown output
• file-alert (node, conditional) - create GitHub issue if degraded/down sources found

Plugin Ideas

1. Sports Data Integration Plugin

**Format:** Babysitter marketplace plugin with install.md

**install.md description:** Installs sports data access skills for AI agents. Provides structured CLI interfaces to public ESPN, FPL, Understat, Transfermarkt, Kalshi, and Polymarket APIs. Includes 14+ sport modules (NFL, NBA, MLB, NHL, football/soccer, F1, tennis, golf, college sports), prediction market integration, betting analytics (odds conversion, de-vigging, Kelly criterion, arbitrage), and a sports reporter for generating journalism from live data. Zero API keys required.

**Key components:**

• Skills: per-sport data access (19 skills), markets orchestration, betting analysis, sports reporter
• Commands: sport-specific data queries, market comparison, edge analysis
• Scripts: nightly health check for upstream API monitoring
• References: per-skill API reference docs, team/player ID lookups, conference mappings

**Why this is a plugin:** Persistent data access tooling that agents use on-demand, not a one-shot workflow. The skills activate contextually when users ask about sports data or betting analysis.

Implicit Procedural Knowledge

• **De-vigging as a prerequisite:** Never compare raw sportsbook odds directly against prediction market prices. Always de-vig first to extract fair probability, then compare. This is the single most common analytical error in sports betting.
• **Source-aware price normalization:** ESPN uses American odds (-150/+130), Polymarket uses 0-1 decimals, Kalshi uses 0-100 integers. Normalization to implied probability must happen before any cross-source comparison.
• **Nightly health checks with threshold-based classification:** The OK/Degraded/Down trichotomy (using a 3-second latency threshold) is more useful than binary up/down for services that gracefully degrade.
• **Sport-aware entity search:** When bridging ESPN schedules to prediction markets, map through sport codes (e.g., NBA -> KXNBA for Kalshi, NBA -> series_id for Polymarket) rather than free-text search. Structured mapping eliminates false matches.

Library Mapping

Plugin Marketplace Mapping