Imported from Dxboy266/FIFA-WINNER-SKILL (skill/SKILL.md). Install upstream with npx skills add Dxboy266/FIFA-WINNER-SKILL --skill skill. Copyright stays with the author.

FIFA-WINNER-SKILL

FIFA-WINNER-SKILL is a reusable World Cup edition workflow. It keeps raw sources, compiled wiki notes, structured data, predictions, posters and post-match evaluations tied to the same edition and match ledger.

Agent-to-Agent Entry

Runtime agents should read AGENT_README.md before invoking commands. The machine-readable capability card is skill/AGENT_CARD.json; the tool/resource/prompt catalog is skill/TOOL_CATALOG.json; the quick operator guide is skill/RUNBOOK.md.

Use JSON reports as canonical audit artifacts. Use SQLite only as a query/index layer. If the two disagree, prefer the locked JSON report and report the mismatch.

Safety First

Every prediction is entertainment only.

Required disclaimer:

娱乐预测，非投注建议；不得作为投注、购彩或资金决策依据。

Never output stake sizing, odds advice, guaranteed wins, lottery advice, 稳赢, 稳胆, or similar gambling-oriented language.

Command Roots

If running as a standalone GitHub repository, use commands from the repo root with --root ..

If running inside the dxboy knowledge base, use the project path _meta/projects/世界杯预测/ and keep edition data isolated under:

• raw/体育/世界杯/<edition>/
• wiki/体育/世界杯/<edition>/
• _meta/projects/世界杯预测/data/editions/<edition>/

Quick Routes (Tool Layer CLI Command Reference)

• Initialize Edition & Structure:
  • python3 skill/scripts/worldcup_edition_init.py init --edition <edition> --root .
• Source Config & Plan Audits:
  • python3 skill/scripts/worldcup_source_readiness_auditor.py write --edition <edition> --root .
  • python3 skill/scripts/worldcup_prediction_evidence_planner.py write --edition <edition> --root .
• T0/T1 Web Source Snapshotting:
  • python3 skill/scripts/worldcup_source_snapshot_tool.py plan --edition <edition> --source-id <source-id> --root .
  • python3 skill/scripts/worldcup_source_snapshot_tool.py apply --edition <edition> --source-id <source-id> --root .
• Official Squad PDF Parser:
  • python3 skill/scripts/fifa_squad_pdf_parser.py parse --edition <edition> --pdf <path/to/pdf> --update-edition-teams --root .
• Initialize Team Profiles & Player Dossiers:
  • python3 skill/scripts/worldcup_profile_init.py init --edition <edition> --scope [teams|players|all] --root .
• FIFA Fixtures & Schedule Parsers:
  • python3 skill/scripts/worldcup_fixture_parser.py parse --edition <edition> --schedule-json <path/to/json> --root .
• FIFA Official Men's Ranking Parser:
  • python3 skill/scripts/worldcup_ranking_parser.py parse --edition <edition> --ranking-json <ranking.json> --snapshot-manifest <manifest.json> --root .
• Squad Depth & Features Aggregator:
  • python3 skill/scripts/worldcup_squad_depth_compiler.py build --edition <edition> --root .
• Roster Alignment compiler:
  • python3 skill/scripts/worldcup_roster_compiler.py compile --edition <edition> --root .
• Compile Tournament Evidence (recent form / H2H / injury check / rest / history paths):
  • python3 skill/scripts/compile_prediction_evidence.py
  • Rebuilds wiki/public/<edition>/evidence/*, copies history/rankings to expected paths, scaffolds matchday daily-evidence when needed.
• Adjust Daily Context Evidences (Weather, Injuries, Referee):
  • python3 skill/scripts/daily_evidence_input.py init --edition <edition> --date YYYY-MM-DD --root .
  • python3 skill/scripts/daily_evidence_input.py status --edition <edition> --date YYYY-MM-DD --root .
• Live Odds & News Sentiment Web Fetchers:
  • python3 skill/scripts/worldcup_live_fetcher.py fetch-odds --edition <edition> --date YYYY-MM-DD --root .
  • python3 skill/scripts/worldcup_live_fetcher.py fetch-news --edition <edition> --date YYYY-MM-DD --root .
• Historical Results Fetcher:
  • python3 skill/scripts/worldcup_history_fetcher.py fetch --edition <edition> --root .
• Physics Prediction Model (rankings, squad, rest, evidence completeness → outcome → confidence):
  • python3 skill/scripts/prediction_scoring_model.py predict --edition <edition> --date YYYY-MM-DD --root .
  • By teams: python3 skill/scripts/prediction_scoring_model.py predict --edition <edition> --teams "TeamA,TeamB" --root .
  • By match ID: python3 skill/scripts/prediction_scoring_model.py predict --edition <edition> --match-id <match_id> --root .
  • Optional: --now ISO-time --force to re-run pre-kickoff; never rewrite after kickoff lock.
• Daily Prediction Runner (E2E daily runner):
  • python3 skill/scripts/daily_prediction_runner.py run --edition <edition> --date YYYY-MM-DD [--now ISO-time] [--poster] --root .
• Unified Agent Entrypoint (Octopus Paul Agent):
  • python3 skill/scripts/octopus_paul_agent.py fetch-schedule --edition <edition> --root .
  • python3 skill/scripts/octopus_paul_agent.py predict --edition <edition> [--phase <phase> | --group <group> | --teams <teams> | --all] [--now ISO-time] --root .
• Prediction Report Prompt Builder:
  • python3 skill/scripts/prediction_report_prompt_builder.py build --edition <edition> --date YYYY-MM-DD --report-path <report.json> --match-id <match_id> --root .
• Poster Prompts Builder (Chinese Showdown Template):
  • python3 skill/scripts/poster_prompt_builder.py build --edition <edition> --date YYYY-MM-DD --style [prediction|showdown] [--match-id <match_id>] --root .
• Poster Generator & Rendering:
  • python3 skill/scripts/poster_generator.py generate --manifest <manifest.json> --backend image2 --root .
• Post-Match Predictions Evaluator:
  • python3 skill/scripts/prediction_evaluator.py write --edition <edition> --date YYYY-MM-DD --root .
• Prediction Accuracy Dashboard Compiler:
  • python3 skill/scripts/prediction_evaluation_dashboard.py write --edition <edition> --root .
• README & Calendar History Compiler:
  • python3 skill/scripts/update_readme_and_history.py --edition <edition> --date YYYY-MM-DD --now <now> --root .
• GitHub Public Readiness Auditor:
  • python3 skill/scripts/worldcup_github_readiness_auditor.py write --edition <edition> --root .
• Standalone Portable Export Tool:
  • python3 skill/scripts/worldcup_export_standalone.py --edition <edition> --output <target_dir> --root .

Workflow

1. Initialize the edition if directories or match ledger are missing.
2. Run source readiness before claiming sources are usable.
3. Snapshot T0/T1 sources before parsing them; every snapshot needs URL, tier, hash and allowed-use metadata.
4. Parse official fixtures, rosters and rankings before stronger prediction claims.
5. Every remaining match uses the same pipeline (group → R32 → R16 → QF → SF → F):
   1. Update match-ledger (teams bound, kickoffs, final scores for finished games).
   2. compile_prediction_evidence.py then worldcup_prediction_evidence_planner.py write.
   3. Init/refresh daily-evidence/<date>.json; fetch odds/news when available.
   4. Predict outcome/score first via prediction_scoring_model.py / daily_prediction_runner.py / octopus_paul_agent.py.
   5. Then score confidence with _score_prediction_confidence (earned only — never force high / never patch frontend labels).
   6. Publish locked report + optional dashboard rebuild.
6. Mark missing evidence as partial or blocked; never pretend it is complete.
7. Only predict matches that have not kicked off. Prefer prediction_scoring_model.py for official reports.
8. Lock pre-match reports. Do not overwrite them after kickoff.
9. Build report prompts from structured prediction reports, not memory. Build poster prompts only when requested by the user.
10. After matches, append evaluation and update the aggregate dashboard.

Source Tiers

• T0: FIFA official schedule, FIFA official squad PDF, FIFA rankings, national FA official sites.
• T1: Wikidata, Wikipedia, OpenFootball and similar structured open sources.
• T2: football-data.org, API-Football, TheSportsDB; only after key, rate limit and license boundaries are recorded.
• T3: FBref, StatBunker, Transfermarkt, ESPN and similar references; cross-check only, no unauthorized bulk scraping.

Prediction Evidence

Check these before daily predictions: official fixtures, official rosters, FIFA rankings, historical World Cup results, recent form, squad depth, injury availability, venue/rest/travel, head-to-head and player identity enrichment.

Statuses must be complete, partial or blocked.

Compile from disk when families are stale:

python3 skill/scripts/compile_prediction_evidence.py
python3 skill/scripts/worldcup_prediction_evidence_planner.py write --edition 2026 --root .

Artifact roots (public edition data):

• wiki/public/<edition>/evidence/recent-form.json
• wiki/public/<edition>/evidence/injury-availability.json
• wiki/public/<edition>/evidence/head-to-head.json
• wiki/public/<edition>/evidence/rest-travel-features.json
• wiki/public/<edition>/history/team-wc-history.json
• wiki/public/<edition>/daily-evidence/<YYYY-MM-DD>.json
• wiki/public/<edition>/prediction-evidence-plan.json

Standard Match Prediction (all remaining fixtures)

Use this for any unstarted match (including SF/F). Same code path — no special-case high labels.

# 1) evidence
python3 skill/scripts/compile_prediction_evidence.py
python3 skill/scripts/worldcup_prediction_evidence_planner.py write --edition 2026 --root .
python3 skill/scripts/daily_evidence_input.py init --edition 2026 --date YYYY-MM-DD --root .
python3 skill/scripts/worldcup_live_fetcher.py fetch-odds --edition 2026 --date YYYY-MM-DD --root .   # when feed available
python3 skill/scripts/worldcup_live_fetcher.py fetch-news --edition 2026 --date YYYY-MM-DD --root .

# 2) predict (single match OR whole day OR phase)
python3 skill/scripts/prediction_scoring_model.py predict --edition 2026 --match-id <match_id> --root .
# or whole day:
python3 skill/scripts/daily_prediction_runner.py run --edition 2026 --date YYYY-MM-DD --root .
# or phase / all open:
python3 skill/scripts/octopus_paul_agent.py predict --edition 2026 --phase semi_final --root .

# 3) dashboard (optional)
python3 skill/scripts/prediction_visual_dashboard.py write --edition 2026 --root .
# dashboard-v2 rebuild if using wiki/public/2026/dashboard-v2/build_data.py

玩法卡片

Every daily prediction should include play_card with share title, match hook, watch points, risk flags, poster angle, confidence meter and gameplay tags. Keep it fun and shareable, but never gambling-oriented.

Prediction Rules

• 数据模型权重 (基本面 + 市场)：60%（阶段权重可在模型内微调；以报告 JSON 为准）。

• 天纪气运娱乐层权重：上限 40%。

• Order is fixed: compute outcome + scoreline first, then multi-factor confidence. Never force high via display policy or frontend badges.

• Confidence comes only from prediction_scoring_model._score_prediction_confidence:

  • Inputs: data readiness, edge tier, evidence quality, evidence gaps, scoreline mode, market status, track votes, KO adjustments.
  • edge_tier == coinflip hard-blocks high (even with complete evidence).
  • High needs score ≥ ~72, multi-support factors, no critical gaps / unusable market.

• Missing roster, injury, lineup or recent-form evidence must downgrade confidence.

• odds_unavailable is no market (status none), not mock-invalid — do not mark evidence unusable solely for missing odds.

• Knockout: no primary draw unless true coinflip lean; prefer directional + both-score scoreline pool (knockout_policy / scoreline_tuning in model-hyperparameters.json).

• Reports must keep the entertainment disclaimer.

• Exact score must come from a scoreline distribution, not from a single default template.

• Encode scoreline failure modes explicitly: repeated 2-1, 1-2, and 1-1 overuse is a model issue, not a harmless style choice.

• Treat result correct + score wrong as a separate learning signal; do not only tune winner/loser direction.

• If evaluated matches show more clean sheets than predictions, increase shutout branches and suppress automatic loser-goal assumptions.

• Keep scoreline heuristics only when post-match evidence supports them; validation beats intuition.

Poster Rules

• image2 is a configurable backend alias.
• User-facing image2 prompts must be plain .txt, not JSON.
• Do not build poster prompts unless the user explicitly asks for poster material.
• Missing backend must return blocked_missing_backend.
• Poster manifests must keep prompt, source report, backend, output path and provenance.