diff --git a/tools-sync/scripts/sync_repos.ps1 b/tools-sync/scripts/sync_repos.ps1
index eca6f60..ed064a0 100644
--- a/tools-sync/scripts/sync_repos.ps1
+++ b/tools-sync/scripts/sync_repos.ps1
@@ -51,7 +51,7 @@ function Invoke-Doctor([string]$Why) {
     $doctor = Join-Path $PSScriptRoot "gitea_doctor.ps1"
     if (-not (Test-Path $doctor)) { Write-Host "gitea_doctor.ps1 not found; cannot repair the server."; return $false }
     Write-Host ""
-    Write-Host "=== transport fault ($Why) - running gitea_doctor -Fix ==="
+    Write-Host "=== gitea transport check/repair ($Why) ==="
     $dargs = @("-RepoUrl", $RepoUrl, "-Fix")
     if ($AllowDestructive) { $dargs += "-AllowDestructive" }
     & $doctor @dargs
@@ -143,6 +143,16 @@ if ($DryRun) {
     }
     if ($r.Code -ne 0) { Write-Host $r.Out; Fail "push failed (auth? see skill Phase 1)." }
     Assert-RemoteInSync $ClaudeSkills $Branch
+    # ls-remote only proves the ref plane works. A sync run can push happily while the
+    # server hands out empty packs, and then no other machine can fetch what we just
+    # pushed - which is precisely how the 2026-08 outage hid. The doctor's transport
+    # probe is a few KB and returns without touching ssh when everything is fine, so
+    # "in sync" is only reported once fetching actually works.
+    if (-not $NoDoctor) {
+        if (-not (Invoke-Doctor "post-sync fetchability probe")) {
+            Fail "this machine's work is on the server, but the server cannot serve packs: no other machine can fetch it. Not a successful sync - see the skill's Phase 1.5."
+        }
+    }
 }
 
 # ---------- Phase 2: junction reconciliation (~\.codex\skills) ----------