Which layer the error belongs to
The message dns.fake-ip-filter[n]: invalid domain is emitted while the core reads the configuration. It is not a connection error, not a DNS resolution error at runtime, and not evidence about the remote server. The bracketed number is the index of the offending entry inside the array, and the quoted string is the value the parser refused.
| What you see | What it establishes | Next step |
|---|---|---|
dns.fake-ip-filter[2]: invalid domain "stun*.*.*" | The Fake IP filter entry failed domain syntax validation; the core may not have loaded the configuration at all | Locate the generated config and the entry in the source subscription, back both up, then fix the line |
the remote profile data is invalid yaml | The YAML text failed to parse, so semantic validation never ran | Use the invalid YAML path instead — different stage, different fix |
| Core started, but enabling system proxy kills connectivity | Configuration loading succeeded; the fault is in ports, rules, DNS or the node | Use the system-proxy-no-internet path |
| Same node fails only when TUN is on | Traffic capture is involved, which is past configuration loading | Use the TUN troubleshooting path |
What the index number does not tell you
[2] is a position in an array, not a label. It cannot tell you whether the bad entry came from the airport subscription, from a client-side override, or from a local script that rewrites DNS settings. Only a line-by-line comparison of the source configuration against the generated runtime file answers that.
There is a second trap in the same log: the file named clash-verge-check.yaml is the temporary copy used for validation. Editing that copy does not change what the client generates on the next start. Fix the source — the subscription, the override, or the merge script — or the error returns.
Why a wildcard inside a label is rejected
Mihomo's domain syntax treats * as a wildcard that must occupy an entire label. stun*.*.* places the asterisk inside the label stun*, which is not a legal domain label, so the whole entry is refused. This is a syntax rule, not a policy about which domains you may filter.
Two documented forms behave differently and are not interchangeable:
| Pattern | What it covers | When it is the right tool |
|---|---|---|
*.example.com | One label of subdomain under example.com | You want a specific depth only |
+.example.com | example.com and every depth below it | You want the apex and all descendants |
stun*.*.* | Nothing — invalid label | Never; this is the failing form |
Quoting matters as well. A pattern containing * or + should be quoted in YAML so the parser does not read it as an alias or a flow indicator.
Recover the configuration in order
- Preserve a working connection first. Turn off TUN and the system proxy, keep a full copy of the error text, and mask the subscription URL, token, username and local paths before sharing it anywhere. Confirm the log actually contains
dns.fake-ip-filterandinvalid domain, not only the generic “core failed to start” summary. - Back up both sides. In Clash Verge Rev, export the remote profile in use and any local override. There are two files in play: what the provider sends and what the client generates from it.
- Compare the source first. If the source subscription already contains the bad entry, the fix belongs to the provider or to a local corrected copy. If only the generated file contains it, look at overrides, merge rules and scripts.
- Confirm the failure by removal, not by rewriting. Temporarily delete the exact entry named in the log, save, and re-run the client's own config check. Success here proves the entry was the blocker; it does not yet prove the node works.
- Then write the pattern you actually need. Decide which domains must match before choosing
*.or+., and add quotes. Do not substitute one form for the other mechanically. - Enable one capture method only. With the config passing, turn on system proxy or TUN — one, not both — and make a single HTTPS request to a known site. Configuration passing only clears the syntax gate.
- Handle the subscription cycle. If the error appeared after a subscription update, keep the old subscription and your corrected local copy side by side and ask the provider to fix the source. Otherwise the next update reintroduces it.
A real report, and its limits
Clash Verge Rev issue #8462 documents one Windows 11 case on client 2.5.5 and later, where stun*.*.* appeared in dns.fake-ip-filter and runtime validation failed. The same log shows the client attempting to start with a default minimal configuration, which is why “the core never ran” is the wrong conclusion from this line alone. A repository collaborator stated the configuration could not pass core validation; the issue was later closed as not planned, which is not the same as an automatic fix for every similar configuration. The thread does not establish a corresponding Mihomo core version.
Mihomo issue #3116 is a separate record: Linux Docker, Mihomo 1.19.30, a rrn-sw-* pattern producing the same class of error. It is not a version mapping for the client case above. A maintainer note in that thread adds a detail worth keeping: older versions did not reject this syntax, which does not mean it ever matched as intended.
Both records are public issue reports, not reproductions on your device. The steps here are a reversible locating procedure, not a claim that a particular release repairs old subscriptions.
When this page is the wrong page
- No exact
invalid domainstring. Do not delete DNS rules hoping to get lucky.dns resolve failedis a runtime resolution failure;invalid domainis a load-time syntax failure. - Several configuration errors at once. Handle the first clearly identified failure, then re-validate. Fixing them in parallel hides which change mattered.
- The error appears after an update. Distinguish a validation failure from a failed restart of the app after the installer ran. Those are different stages with different logs.
- You only need one browser. System proxy mode can serve as a temporary path, but its capture scope differs from TUN and the two are not equivalent.
What to record before asking for help
- Client version, core version and operating system build.
- The exact log line including the array index and the quoted string.
- Whether the entry exists in the source subscription or only in the generated file.
- The override or script in use, described without its contents if it contains credentials.
- What happened after removing the entry: did validation pass, did the core start, did one HTTPS request succeed.
- Whether the error returned after the next subscription update.
Mask the subscription URL to its domain, and never publish a token, UUID, password or full query string. A config error is cheap to share safely; a leaked subscription is not.