Status Page and Uptime Monitor (Python, nothing to install)

Read CLAUDE.md. Create sites.json with "interval": 60 and a "sites" list of two entries, each with a "name" and a "url" (use https://example.com and https://www.python.org). Create monitor.py with a function check(url) that sends a GET request with a 10 second timeout and returns (ok, status_code, ms, error). Any 2xx or 3xx is ok; timeouts, connection errors and 4xx/5xx are not. Add a command `python monitor.py --once` that checks every site once and prints one line per site: name, UP or DOWN, status code and response time.

Store each check in SQLite (status.db next to monitor.py). Table checks: id, site, ts (UTC ISO 8601), ok (0/1), status, ms, error. Create the table on start if it is missing, turn on WAL mode, and add an index on (site, ts). --once should save its results too.

Make `python monitor.py` run forever: every "interval" seconds check all sites in parallel (one thread each), save the results and print them. Add an --interval N flag that overrides sites.json (useful for testing). Ctrl+C must stop it cleanly with a short goodbye, no traceback.

Add an incidents table: id, site, started (UTC ISO), ended (NULL while open), reason. After a check, if a site has failed twice in a row and has no open incident, open one with the last error as the reason. On the first successful check, close the open incident. Print a clear line when an incident opens or closes. Never open two incidents for the same site at once.

Add `python monitor.py --serve` (with --port, default 8080) that runs the checks in a background thread and serves a status page at / with http.server. The page shows: a banner ("All systems operational" in green, or "Partial outage" in red when any site has an open incident); for each site its name, current status, uptime % over 90 days, and a strip of 90 small bars, one per UTC day (green 100%, amber 95 to 99.99%, red below 95%, grey for no data) with a tooltip giving the date and %; and the last 10 incidents for the sites in sites.json with start, end or "ongoing", duration and reason. Escape every value you put into HTML. Work in light and dark mode (prefers-color-scheme) and on a phone.

Add GET /api/status.json returning {"generated": ts, "operational": bool, "sites": [{"name", "url", "up", "uptime_90d", "last_check": {"ts", "status", "ms"}, "open_incident": {...} or null}]}. Send Content-Type application/json and Cache-Control: no-store. Any other path returns 404.

Read an optional "webhook" URL from sites.json, or the STATUS_WEBHOOK environment variable (the environment wins). When an incident opens, POST JSON {"text": "🔴 <name> is down: <reason>"}; when it closes, POST {"text": "🟢 <name> is back after <duration>"}. Use a 10 second timeout and never let a failed webhook stop the monitor: log it and carry on. Send exactly one message per open and per close.

Once a day (and on start) delete checks older than 90 days. Then add test_monitor.py using unittest only: test that one failure does not open an incident, two in a row do, a success closes it, a site name containing <script> is escaped on the status page, and the JSON API reports operational=false while an incident is open. Use a temporary database file in the tests.