Case Comp Tracker

A static website that lists case competitions as open, upcoming or closed, with filters for status, UG/PG, domain and host type (college/company). A nightly agent researches new competitions and updates the data.

How it fits together

nightly agent (10 PM IST)  ->  commits data/competitions.json to GitHub
                                       |
                          Vercel / Netlify / GitHub Pages auto-redeploys
                                       |
                              index.html + app.js read the JSON

• No build step and no dependencies. It is plain HTML, CSS and JavaScript.
• Status is computed in the browser from dates, not stored. Open = registration started and deadline not passed. Upcoming = registration has not started yet. Closed = deadline passed.
• Entries with "needs_review": true are hidden from the public site until you flip them to false.

Files

File	Purpose
index.html, style.css, app.js	The website
programmes.js	Code for the Programmes tab
data/competitions.json	All case competitions (the agent edits this and programmes.json, plus logs)
data/programmes.json	All programmes for MBA students and aspirants (Programmes tab)
schema.json	Field definitions and allowed values for competitions
programmes.schema.json	Field definitions and allowed values for programmes
agent/nightly-prompt.md	The prompt the nightly agent runs (competitions, every night 10 PM IST)
agent/weekly-programmes-prompt.md	The prompt the weekly agent runs (programmes, every Monday 10:15 PM IST)
logs/	One short log per nightly run (created by the agent)

Run locally

cd case-comp-tracker
python3 -m http.server 8000
# open http://localhost:8000

(Opening index.html directly will not work, because browsers block fetch on file:// pages.)

Deploy

1. Create a GitHub repo and push this folder.
2. Connect the repo to Vercel, Netlify or GitHub Pages. Publish directory is the repo root; no build command.
3. Every commit to the default branch redeploys the site.

Set up the nightly agent

1. The repo name is already set in agent/nightly-prompt.md.
2. Run the prompt manually a few times first and check the diff it produces.
3. Create a scheduled task in Claude that runs the prompt daily at 10:00 PM IST (CRON_TZ=Asia/Kolkata 0 22 * * *). Each run is a fresh session, so the prompt must stay self-contained.
4. Review logs/ and any entries with needs_review: true once a day at first.

Before launch

• Delete the five SAMPLE: entries in data/competitions.json.
• Add 15-20 real, hand-verified entries, or let the first agent runs fill them in and review them.

Adding or approving an entry by hand

Add an object to competitions in data/competitions.json following schema.json, or change needs_review from true to false to publish a flagged entry.