football-data

Football Data

Before writing queries, consult references/api-reference.md for endpoints, ID conventions, and data shapes.

Setup

Before first use, check if the CLI is available:

which sports-skills || pip install sports-skills

If pip install fails (package not found or Python version error), install from GitHub:

pip install git+https://github.com/machina-sports/sports-skills.git

The package requires Python 3.10+. If your default Python is older, use a specific version:

python3 --version

No API keys required.

Quick Start

Prefer the CLI — it avoids Python import path issues:

Python SDK (alternative):

CRITICAL: Before Any Query

CRITICAL: Before calling any data endpoint, verify:

• Season ID is derived from get_current_season(competition_id="...") — never hardcoded.
• Team ID is resolved via search_team(query="...") and passed as the numeric team_id. For get_head_to_head, get_team_strength, and get_match_forecast, always pass IDs — ambiguous names (e.g. two "Paris" clubs) can resolve to the wrong team.
• The endpoint actually covers the league in question — see the Coverage & Source Map below. Coverage is uneven across sources; an uncovered call returns an empty payload with a message, not data.
• get_event_xg and get_event_players_statistics (with xG) are only called for top-5 leagues (EPL, La Liga, Bundesliga, Serie A, Ligue 1).
• get_season_leaders and get_missing_players are only called for Premier League seasons (season_id must start with premier-league-).

Choosing the Season

Derive the current year from the system prompt's date (e.g., currentDate: 2026-02-16 → current year is 2026).

• If the user specifies a season, use it as-is.
• If the user says "current", "latest", or doesn't specify: Call get_current_season(competition_id="...") to get the active season_id. Do NOT guess or hardcode the year.
• Season format: Always {league-slug}-{year} (e.g., "premier-league-2025" for the 2025-26 season). The year is the start year of the season, not the end year.
• MLS exception: MLS runs spring-fall within a single calendar year. Use get_current_season(competition_id="mls").

Coverage & Source Map

This skill stitches several free sources together. Coverage is not uniform — each endpoint works only where its underlying source has data. Check this before promising an answer; when an endpoint isn't covered, it returns an empty payload with an explanatory message (never an error) — read that message and fall back.

Rule of thumb: ESPN answers "what happened" everywhere; the enrichment sources ( Understat/FPL/ClubElo/football-data.co.uk ) add depth only in their coverage zone. ESPN is always the fixture/score authority — never let an enrichment source override an ESPN score.

Gotchas (from live testing)

• get_team_profile returns the squad. data.players[] carries the current roster with ESPN athlete ids, shirt numbers and ages — use it instead of collecting names match by match.
• get_player_season_stats takes the same league slug as everything else (serie-a-brazil, not only ESPN's bra.1), and its gamelog is the last ~5 matches across competitions, not a season total.
• Scored penalties are penalty_goal in the timeline. Count goal + penalty_goal + own_goal when reconciling with the score.
• Pass IDs, not ambiguous names. For H2H/strength/forecast, resolve teams with search_team first and pass the numeric team_id. Names like "Paris Saint-Germain" can collapse onto the wrong club (Paris FC) during name resolution.
• ClubElo off-season gaps: current-date get_team_strength can miss clubs in the summer break (a club's weekly Elo period may not span today). If a well-known club returns unresolved, pass an in-season date (e.g. date="2026-03-01").
• ClubElo outages: get_team_strength falls back to locally computed Elo and sets source: "local-elo". Check that field before comparing numbers across calls — the local scale is division-local, so a rating means nothing outside its own division and cross-division comparisons are refused. The fallback honours date (it rates the division as of that date, and each entry's as_of is the last match counted). get_match_forecast has no fallback and stays empty.
• get_match_forecast is short-horizon: ClubElo only forecasts ~a week ahead — empty between matchdays / off-season. That's expected, not a failure.
• H2H is same-division only: two clubs that met in a cup or across tiers won't show; it counts league meetings in the resolved division.
• H2H tells "unresolved" apart from "never met": football-data.co.uk uses short exonyms/abbreviations ("FC Koln", "M'gladbach", "Sp Lisbon"). Each club in teams[] reports resolved + matched_as; if a club is resolved: false, zero meetings means the lookup failed, not that the clubs never played.

Combining Endpoints (mix-and-match)

Compose sources for richer answers. Run independent calls in parallel.

• Match preview (X vs Y): search_team ×2 → get_head_to_head (recent record) + get_team_strength(team_id, team_id_2) (Elo gap / favorite) + get_match_forecast (if within ~a week: W/D/L + scoreline). For a top-5 fixture add historical get_event_xg context from recent meetings.
• Match report (post-game): get_event_summary + get_event_statistics + get_event_timeline, and for top-5 leagues get_event_xg + get_event_players_statistics.
• Team form + context: get_team_schedule (recent results) + get_team_strength (current Elo & rank) + get_missing_players (PL only) + per-match get_event_xg (top-5).
• Rivalry / derby deep dive: get_head_to_head (all-time-ish record + goals) + get_team_strength comparison for the current power balance.
• Odds sanity-check: get_match_forecast gives a free model baseline (W/D/L) to compare against the kalshi / polymarket betting skills.

When a piece of the composition isn't covered (e.g. xG outside the top 5, H2H for MLS), skip it silently and deliver the parts that are covered — don't block the whole answer on one missing source.

Commands

See references/api-reference.md for full parameter lists, return shapes, and data coverage table.

Examples

Example 1: Premier League table User says: "Show me the Premier League table" Actions:

1. Call get_current_season(competition_id="premier-league") to get the current season_id
2. Call get_season_standings(season_id=<season_id from step 1>) Result: Standings table with position, team, played, won, drawn, lost, GD, points

Example 2: Match report User says: "How did Arsenal vs Liverpool go?" Actions:

1. Call get_daily_schedule() or get_team_schedule(team_id="359") to find the event_id
2. Call get_event_summary(event_id="...") for the score
3. Call get_event_statistics(event_id="...") for possession, shots, etc.
4. Call get_event_xg(event_id="...") for xG comparison (EPL — top 5 only) Result: Match report with scores, key stats, and xG

Example 3: Team deep dive User says: "Deep dive on Chelsea's recent form" Actions:

1. Call search_team(query="Chelsea") → team_id=363, competition=premier-league
2. Call get_team_schedule(team_id="363", competition_id="premier-league") → find recent closed events
3. For each recent match, call in parallel: get_event_xg, get_event_statistics, get_event_players_statistics
4. Call get_missing_players(season_id=<season_id>) → filter Chelsea's injured/doubtful players Result: xG trend across matches, key player stats, and injury report

Example 4: Player market value User says: "What's Saka's market value?" Actions:

1. Call get_player_profile(tm_player_id="433177") for Transfermarkt data
2. Optionally add fpl_id for FPL stats Result: Market value, value history, and transfer history

Example 5: Non-PL club User says: "Tell me about Corinthians" Actions:

1. Call search_team(query="Corinthians") → team_id=874, competition=serie-a-brazil
2. Call get_team_schedule(team_id="874", competition_id="serie-a-brazil") for fixtures
3. Pick a recent match and call get_event_timeline(event_id="...") for goals, cards, subs Result: Fixtures, timeline events (note: xG, FPL stats, and season leaders NOT available for Brazilian Serie A)

Example 6: Match preview (mix-and-match) User says: "Preview Arsenal vs Man City this weekend" Actions:

1. Call search_team(query="Arsenal") and search_team(query="Manchester City") → team_ids 359, 382
2. In parallel: get_head_to_head(team_id="359", team_id_2="382") (recent record + goals), get_team_strength(team_id="359", team_id_2="382") (Elo gap + favorite), get_match_forecast(team_id="359", team_id_2="382") (W/D/L + likely scoreline, if within ~a week)
3. Synthesize: form/record + power balance + model odds. Skip any piece that returns empty (e.g. forecast if the match is >1 week out). Result: A preview blending head-to-head history, current strength, and a free model forecast

Commands that DO NOT exist — never call these

• get_standings — the correct command is get_season_standings (requires season_id).
• get_live_scores — not available. Use get_daily_schedule() for today's matches.
• get_team_squad / get_team_roster — use get_team_profile: data.players[] is the current roster with ESPN athlete ids (see Gotchas). get_season_leaders + get_player_profile remain the path for career data.
• get_transfers — the correct command is get_season_transfers (requires season_id + tm_player_ids).
• get_match_results / get_match — use get_event_summary with an event_id.
• get_player_stats — use get_event_players_statistics for match-level stats, or get_player_profile for career data.
• get_scores / get_results — use get_event_summary with an event_id.
• get_fixtures — use get_daily_schedule for today's matches or get_season_schedule for a full season.
• get_league_table — use get_season_standings with a season_id.

If a command is not in the Commands table above, it does not exist. Do not try commands not listed.

Error Handling

When a command fails (wrong event_id, missing data, network error, etc.), do not surface the raw error to the user. Instead:

1. Catch it silently — treat the failure as an exploratory miss.
2. Try alternatives — if an event_id returns no data, call get_daily_schedule() or get_team_schedule() to discover the correct ID.
3. Only report failure after exhausting alternatives — use a clean message (e.g., "I couldn't find that match — can you confirm the teams or date?").

Troubleshooting

Error: sports-skills command not found Cause: Package not installed Solution: Run pip install sports-skills. If not on PyPI, install from GitHub: pip install git+https://github.com/machina-sports/sports-skills.git

Error: ModuleNotFoundError: No module named 'sports_skills' Cause: Package not installed or path issue Solution: Install the package. Prefer the CLI over Python imports to avoid path issues

Error: get_season_leaders or get_missing_players returns empty for a non-PL league Cause: These commands only work for Premier League; they silently return empty for other leagues Solution: Check the Data Coverage table in references/api-reference.md. For other leagues, use get_event_players_statistics for player data

Error: get_team_profile returns no players Cause: ESPN has no roster for that team id in the given league (wrong league_slug, or a national team / youth side) Solution: Pass the league_slug the club plays in (e.g. serie-a-brazil); for match-day squads use get_event_lineups

Error: Wrong season_id format Cause: Season ID must follow the {league-slug}-{year} format Solution: Use get_current_season(competition_id="...") to discover the correct format. Example: "premier-league-2025", not "2025-2026" or "EPL-2025"

Error: No xG data for a recent match Cause: Understat data may lag 24-48 hours after a match ends Solution: If get_event_xg returns empty for a recent top-5 match, retry later. Only available for EPL, La Liga, Bundesliga, Serie A, Ligue 1

Error: Team or event ID unknown Cause: ID was guessed instead of looked up Solution: Use search_team(query="team name") to find team IDs, or get_daily_schedule / get_season_schedule to find event IDs. Never guess IDs.