Search Log: job-search dashboard with weekly Gmail refresh

• index.html: the dashboard (static, saves to your browser's localStorage, has Import/Export CSV, dedupes on import)
• scripts/sync_gmail.py: reads recent Gmail, detects applications/interviews/offers/rejections, updates data/applications.json, emails you a weekly summary
• .github/workflows/weekly-refresh.yml: runs the script every Monday (and on demand), commits the updated JSON
• data/applications.json: pipeline output; the dashboard merges it in on load

Privacy: read this first

data/applications.json contains your job-search history. Keep the repo private. GitHub Pages on private repos needs a paid plan, so on the free plan either: run the dashboard locally (python3 -m http.server, then open http://localhost:8000; git pull after each weekly run), or skip Pages and just use the weekly summary email plus Export CSV.

One-time setup (about 15 minutes)

1. Google Cloud: create a project at console.cloud.google.com, enable the Gmail API.
2. OAuth consent screen: user type External, add yourself as a test user, add scopes gmail.readonly and gmail.send. Important: while the app is in "Testing", refresh tokens expire after 7 days. Click Publish app (it stays "unverified", which is fine for personal use) so the token lasts.
3. Credentials: create an OAuth client ID of type Desktop app, download client_secret.json.
4. Get a refresh token (on your computer):

   pip install -r scripts/requirements.txt
   python scripts/get_refresh_token.py client_secret.json
   

   Sign in, approve the scopes, and copy the three printed values.
5. GitHub secrets: repo, Settings, Secrets and variables, Actions, add GOOGLE_CLIENT_ID, GOOGLE_CLIENT_SECRET, GOOGLE_REFRESH_TOKEN.
6. First run: Actions tab, "Weekly job-search refresh", Run workflow. Enter a higher "lookback days" value (e.g. 90) in the Run workflow form to backfill. Scheduled runs scan the last 14 days.

After that it runs every Monday at 13:00 UTC (edit the cron line to change it).

Your own list and LinkedIn's Archived tab

LinkedIn's Job Tracker has no API and isn't in LinkedIn's data export, so the refresh can't read it, including the Archived tab. Keep your real list in data/manual.csv:

1. In the dashboard, set each archived job's Outcome: Ghosted (never heard back), Not moving forward, or No longer accepting. If you're still interviewing, leave Outcome as "Still active" and put it in Notes. A note like "still interviewing" also keeps the row active.
2. Click Export CSV, then upload the file to the data folder on GitHub as manual.csv, replacing the old one.

Rows in manual.csv are locked. The refresh never deletes them or changes their stage, outcome or notes. It only adds email rows for applications newer than the newest one in your list.

How it works and its limits

• Classification is keyword-based (confirmation, screen, round 1/2/3, final, offer, rejection). It is a good first pass, not perfect. Spot-check new rows.
• Source: LinkedIn emails are tagged linkedin; a human interview email with no prior application is tagged recruiter; everything else is ats. Networking can't be detected from email, so change those manually.
• A rejection never overwrites a reached interview stage, so interview rounds stay counted.
• Company and role are parsed from subjects and sender names; odd formats can come out wrong or as "Unknown role".
• LinkedIn is only seen through its application-update emails ([email protected]). InMail, newsletters and job alerts are ignored, so a recruiter who reaches you only by InMail needs adding by hand.
• Your own sent mail and Google Calendar invites are skipped so they don't create duplicate rows.
• Stage rules have tests: python -m unittest discover scripts.
• Rows you first imported from CSV may duplicate pipeline rows if the applied dates differ slightly.
• Your Gmail credentials live only in GitHub Actions secrets.