╭─╮╭──╮
 ╭╯ ╰╯   ╮
 ╰═══════╯
SKILL.md

Football Match Report Skill

Turns any WhoScored match URL into a two-page post-match team report by merging two data sources:

  • WhoScored (headless Chromium) → full event stream (passes, tackles, carries, recoveries). Powers passing networks, defensive blocks, progressive passes/carries, xT momentum and all zone charts.
  • FotMob (plain requests, no token) → shots with xG/xGOT, native match momentum, official stats, player of the match, and real team colors.

The FotMob match id is auto-resolved from the WhoScored match date + team names — a single WhoScored URL is all you need. Both sources are required for the complete report; a FotMob-only scrape (--fotmob-id) renders a reduced report (shots/momentum/stats) without the event panels.

Charts follow the Post-Match-Report-2.0 blueprint (Adnan Ahmed): UEFA pitch, black background, shirt numbers inside player nodes (circle = starter, box = sub), line-height markers, xT momentum.

Two commands, end to end:

python scripts/scrape_match.py "<whoscored-url>" --out ./data
python scripts/render_report.py --data ./data --out ./report

A third command adds the player-level reports (Top Players dashboard, per-player dashboards, per-player stats CSV):

uv run football-match-report players --data ./data --out ./report

When to Use

  • A user asks for a post-match report, match dashboard, or football analytics breakdown and provides a WhoScored match URL (or match id).
  • A recurring match-report job (e.g. after each round of fixtures).
  • Rebuilding/updating an old notebook-based report into the modern pipeline.

Don't use for: season-long datasets (use soccerdata/StatsBomb open data), live in-play streams, or non-WhoScored competitions (Understat etc.).

Prerequisites

  • Python 3.10+ with uv (or pip).
  • Install deps: uv sync --extra scrape (or pip install -e ".[scrape]").
  • One-time browser install: uv run playwright install chromium (WhoScored is bot-walled; the script launches headless Chromium to read the embedded matchCentreData JSON).
  • No API keys. FotMob's public endpoint (/api/data/matchDetails) is used for shots/xG/momentum/colors and does not require a token.

How to Run

1. Scrape (network required, ~10-30s)

python scripts/scrape_match.py \
  "https://www.whoscored.com/matches/1873310/live/international-fifa-club-world-cup-2025-salzburg-real-madrid" \
  --out ./data

Produces data/{matchdict.json, events.csv, shots.csv, meta.json, xt_grid.csv}. The FotMob match id is auto-resolved from date + team names. Pass --fotmob-id <id> to skip resolution, or --no-browser to reuse a cached matchdict.json. Any match works — teams, colors, scores and stats all come from the scraped data; nothing is hardcoded per match.

2. Render (offline, a few seconds)

python scripts/render_report.py --data ./data --out ./report

Produces two 4x3 report figures:

  • report/match_report_1.png — Team report: passing networks, shot map with xG stats bar, defensive blocks, goalkeeper saves, progressive passes, xT momentum, progressive carries, match stats.
  • report/match_report_2.png — Zones report: final-third entries, box entries, Zone 14 & half-spaces, crosses, pass-end-zone heatmaps, high turnovers, chance-creating zones, congestion map.

3. Player-level reports (offline, optional)

uv run football-match-report players --data ./data --out ./report
# or, only specific players:
uv run football-match-report players --data ./data --out ./report \
  --players "Lamine Yamal,Mikel Oyarzabal"

Produces:

  • report/match_report_3.png — Top Players dashboard (4x3): top ball progressor pass maps, passes received by the center-forward, top defender actions, goalkeeper pass maps + Top10 stacked bar charts (ball progressors, shot-sequence involvement, defenders, threat creators via xT).
  • report/players/<name>.png — one 2x3 individual dashboard per starter (or the names in --players): pass map, carries & take-ons, shot map with xG/xGOT, passes received, defensive actions, touches heatmap with distance covered.
  • report/player_stats.csv — merged wide per-player stat table.

4. As a CLI (same thing)

uv run football-match-report scrape "<whoscored-url>" --out ./data
uv run football-match-report render --data ./data --out ./report

Quick Reference

Task Command
Scrape any match (WhoScored + FotMob auto-merged) python scripts/scrape_match.py "<whoscored-url>" --out ./data
Render both report pages python scripts/render_report.py --data ./data --out ./report
Render player-level reports (top players + per player + CSV) uv run football-match-report players --data ./data --out ./report
Limit per-player dashboards to specific names ... players --players "Lamine Yamal,Mikel Oyarzabal"
Explicit FotMob id ... --fotmob-id 4685754
FotMob-only reduced bundle (no browser) python scripts/scrape_match.py --fotmob-id 4685754 --out ./data
Skip browser (reuse cache) ... --no-browser
Run tests uv run pytest
Lint uv run ruff check src tests

Procedure

  1. Confirm the URL is a WhoScored match URL. The match id is the number after /matches/. If the user only has a FotMob link or id, --fotmob-id still works but event panels degrade.
  2. Run the scrape (command above). Completion criterion: events.csv exists with 500+ rows and meta.json has teams.home.name.
  3. Run the render (command above). Completion criterion: both match_report_1.png and match_report_2.png exist.
  4. Inspect the PNGs (vision_analyze in Hermes) for empty panels or overlapping text. The most common failure is a missing FotMob id, which empties the shot map, xT momentum and GK panels — re-scrape with --fotmob-id.
  5. Deliver both PNGs inline to the user.

Metrics included (per team)

  • Passing network — nodes at median positions with shirt numbers, circle = starter / square = sub, line thickness ∝ pass volume, verticality %, defensive/forward line heights + shaded zone.
  • Shot map — football markers for goals, hatched saves, orange posts, big-chance scaling, plus a Goals/xG/xGOT/Shots/On Target/BigChance/ BigC.Miss/xG-Shot/Avg.Dist comparison bar.
  • Defensive block — KDE heatmap, action-height line, compactness %.
  • Goalkeeper saves — goal-mouth view of shots faced per keeper.
  • Progressive passes & carries — comet lines / dashed arrows with left-center-right zone split.
  • xT momentum — average xT per minute (home above / away below zero) with goal and red-card markers.
  • Match stats — possession, field tilt, passes, long balls, corners, GK kick length, tackles, interceptions, clearances, aerials, PPDA.
  • Zone charts — final-third entries (by pass/carry), box entries, Zone 14 & half-space passes, crosses (acc./unacc.), pass-end-zone heatmaps, high turnovers (led to goal/shot), chance-creating zones (key passes = violet, assists = green), congestion map.
  • Top Players dashboard (figure 3) — top ball progressor pass maps (all/progressive/key/assist passes + progressive carries), passes received by the center-forward, top defender action maps, GK pass maps (open play vs goal kicks/free kicks), Top10 stacked bars: ball progressors, shot-sequence involvement, defenders, xT threat creators.
  • Individual player dashboard — pass map (accuracy, progressive, chances created, assists, final third, penalty box, crosses, longballs, xT), carries & take-ons (progressive, led to shot/goal, box entries, dispossessed, success rate), shot map with xG/xGOT (FotMob) and inside/outside box split, passes received (final third, box, progressive, cutbacks, ball retention, most passes from), defensive actions (tackles won, dribbles past, recoveries, blocks, aerials, possession wins per third), touches heatmap with distance covered and area covered.

See references/metrics.md for definitions and interpretation notes.

Pitfalls

  • Both sources are needed for the full report. WhoScored supplies the event stream; FotMob supplies shots/xG/momentum/colors. A FotMob-only scrape renders a reduced report.
  • WhoScored blocks plain requests (403). Use the Playwright path; if playwright install chromium was skipped, the script exits with install instructions. --no-browser only works when matchdict.json already exists in the data dir.
  • Coordinate spaces. WhoScored events are 0-100 and scaled to a UEFA 105x68 pitch (x*1.05, y*0.68) — the blueprint convention. Both teams attack toward x=105 in the raw data; the away team's axes are inverted for display and home shots are flipped in the shot map.
  • FotMob match id resolution needs the match date. If the date endpoint returns nothing, pass --fotmob-id explicitly (from the FotMob URL #<id>).
  • Empty shot map = missing FotMob id. Event panels still render; only shot/xG/momentum/GK panels degrade.
  • Player names in passing networks come from WhoScored's playerIdNameDictionary; without matchdict.json only counts are shown.
  • Player-level panels need real names to match. FotMob shot xG/xGOT match on playerName (fallback: shortName); if a player's name differs between sources, xG shows 0. Without shots.csv the individual shot map falls back to WhoScored events (positions but no xG).
  • Only players who touched the ball appear in player_stats.csv and the per-player dashboards — bench/roster players with zero events are skipped by design.
  • Players with <3 touches get a 0 Total_Area_Covered (convex hull guard) instead of crashing the heatmap.
  • Be polite: scrape one match at a time; don't loop over hundreds of URLs.

Verification

  • scrape_match.py prints per-step status lines and a final ✅ Done in Xs. Bundle ready in ....
  • render_report.py prints ✅ <path> for both PNGs.
  • Open match_report_1.png: every panel populated, black background, team colors consistent (home left, away right), shirt numbers legible inside nodes, xT momentum in the center row.
  • Open match_report_2.png: all 12 zone panels populated with data.
  • uv run pytest passes (offline, uses synthetic match data).