wc26-predict-text-only

name

wc26-predict-text-only

description

Use this skill whenever the user needs a pure md/txt distributable version of the WC26 Predict project that can reconstruct a runnable local prediction app without downloading any upstream repository. This skill is specifically for rebuilding the text-only ELO-capable WC26 project from bundled markdown references. Also use it when the user says "只支持 md/txt", "给朋友发一个纯文本 skill", "安装后重建 WC26 预测项目", "不要依赖上游仓库", or wants the earlier modified project packaged in a text-only skill format.

WC26 Predict Text-Only Skill

This skill rebuilds a runnable WC26 Predict project from bundled markdown references only.

It exists for environments where the distributable package may contain only:

• .md
• .txt
• or folders containing only those file types

The generated project is based on the already modified local WC26 Predict build, but expressed as text templates.

What this skill should do

When triggered, reconstruct a local project by creating these files from the bundled references:

• elo/app.R
• elo/simulation.R
• data/teams.csv
• data/matches.csv
• data/venues.csv
• data/score_dist.csv
• data/cache/results.tsv

Then help the user install dependencies, run the app, and verify the output.

Important scope

This is the text-only version of the project.

• It does not download the upstream GitHub repository.
• It does not require any binary model files to be bundled.
• It is designed to run the ELO-based workflow and the Chinese/weather/stadium enhancements.
• The Dixon-Coles UI may still appear in the generated app, but without the DC model file it is expected to be unavailable.

That trade-off is intentional so the project can be distributed in a pure markdown/text format.

Reconstruction workflow

Follow this sequence exactly:

1. Create the target folder structure:
   • elo/
   • data/
   • data/cache/
2. Read the bundled reference files in references/.
3. Write each referenced file to disk with the exact contents from the code fences.
4. Install required R packages if the machine does not already have them.
5. Verify with a source check.
6. Launch the Shiny app if requested.

Reference files

Use these reference files:

• references/app-r.md
• references/simulation-r.md
• references/data-core.md
• references/data-schedule.md

Always copy the file contents exactly from those references. Do not paraphrase or regenerate them from memory if the reference exists.

Required R packages

Install these if missing:

• shiny
• dplyr
• DT
• future
• promises
• jsonlite

Recommended install command:

Rscript -e "options(repos=c(CRAN='https://mirrors.tuna.tsinghua.edu.cn/CRAN/')); install.packages(c('shiny','dplyr','DT','future','promises','jsonlite'))"

Verification

Always verify in two passes:

1. Source check

Rscript -e "setwd('/path/to/project/elo'); source('app.R', local = new.env()); cat('source_ok\\n')"

2. Runtime check

Rscript -e "shiny::runApp('/path/to/project/elo', host='127.0.0.1', port=3840, launch.browser=FALSE)"

Then confirm that the rendered app includes:

• ⚙️ 参数设置
• ▶ 开始模拟
• 🏅 小组积分榜
• 🌦️ 天气与球场

Output expectations

When you finish, report:

• where the project was written
• which reference files were used
• whether R dependencies were installed
• whether the app was started
• the localhost URL if running
• that weather and stadium data are auxiliary analysis by default
• that Dixon-Coles is not bundled in this text-only version

Example prompts this skill should handle

Example 1 User: 给我一个只含 md/txt 的 WC26 预测 skill。

Example 2 User: 朋友那边只能装纯文本 skill，你帮我重建一个可运行的 WC26 预测项目。

Example 3 User: 不要下载上游仓库，直接用这个 text-only skill 生成预测项目。