[Hiddify-Official]

Builds on #86 (the notifications bell) and includes its commit until that one is merged. Only the last commit is this fix.

Problem

Config health on the dashboard could fail with a raw error that overflowed the card: Could not fetch the subscription: HTTPSConnectionPool(...) SSLError(CertificateError("hostname '<ip>' doesn't match ...")). It didn't say what was wrong or where to look.
Root cause: the check fetched the subscription from the first domain, which can be the server's bare IP. The server's certificate is for its name, not that IP, so the fetch failed although the configs were fine.

Change

• Address (proxy_v3/config_health.py, config_tester.py):
  • a domain name is preferred over a bare IP;
  • with only an IP, the server fetches its own subscription without the certificate check (fetch(verify=False)), since no certificate names a bare IP.
• Errors (config_health.py):
  • each error gets a kind: certificate, timeout, dns, unreachable, http, empty, core, other;
  • the stored message no longer holds the subscription path or the UUID;
  • the result keeps the host that was tested.
• Card (ConfigHealthCard.vue, new shared/composables/useConfigHealth.ts):
  • an explanation per kind, saying where to look (e.g. "the TLS certificate of is not valid … check this domain in Domains");
  • the technical message on demand, wrapped inside the card;
  • a reason in words for each failed config ("The server closed the connection", "The TLS handshake failed", …).
• Bell (usePanelNotifications.ts):
  • Config health problems also show as notifications, errors first: which app, what's wrong, and where the details are;
  • only for super admins, refreshed every 10 minutes.
• Strings in en and fa (the only languages that have the Config health strings).

Screenshots

Before	After

With the technical details and a failing app opened:

Tested

• On main (panel restart):
  • the exact error reproduces when fetching with a name the certificate doesn't cover;
  • verify=False fetches it;
  • _host() picks the domain name;
  • a full real check completed with no fetch error.
• Error kinds and cleanup: checked on sample messages (certificate, timeout, HTTP 404, refused, no config).
• UI: checked with a sample certificate error put in redis, then the real result was restored: desktop, mobile, Persian RTL, and the bell (the error first, the failed-config count from total - ok, not from the 60-item list).
• Deploy: the panel is deployed to main, node1 and node2.

Review

• Confidence: High (>90%) — root cause reproduced and fixed, and the result format only gains fields (error_kind, host). Older results still render.
• Complexity: Simple — the Config health module, its card and the bell.

🤖 Generated with Claude Code

[github-actions]

🥷 Ninja i18n – 🛎️ Translations need to be updated

Project /project.inlang

lint rule	new reports	level	link
Missing translation	125	warning	contribute (via Fink 🐦)
Identical pattern	1	warning	contribute (via Fink 🐦)