diff --git a/.github/workflows/ci.yml b/.github/workflows/ci.yml
index cea29f3..92a99e8 100644
--- a/.github/workflows/ci.yml
+++ b/.github/workflows/ci.yml
@@ -4,15 +4,11 @@ on:
   pull_request:
 
 permissions:
-  pull-requests: write
-  issues: write
   contents: read
 
+# Label check lives in label-check.yml on pull_request_target so it can
+# label fork-submitted PRs; plain pull_request tokens are read-only there.
 jobs:
-  label-check:
-    name: Label Check
-    uses: ApolloAutomation/Workflows/.github/workflows/label-check.yml@main
-
   ci:
     name: CI
     uses: ApolloAutomation/Workflows/.github/workflows/esphome-ci.yml@main
diff --git a/.github/workflows/label-check.yml b/.github/workflows/label-check.yml
new file mode 100644
index 0000000..96a9751
--- /dev/null
+++ b/.github/workflows/label-check.yml
@@ -0,0 +1,20 @@
+name: Label Check
+
+# pull_request_target (not pull_request) so the job gets a write token on
+# fork-submitted PRs too; plain pull_request runs from forks are read-only
+# and cannot add labels. Safe because the called workflow only reads the PR
+# body and never checks out or executes PR code. The "edited" type re-runs
+# the check when the template checkboxes are changed.
+on:
+  pull_request_target:
+    types: [opened, edited, reopened, synchronize]
+
+permissions:
+  pull-requests: write
+  issues: write
+  contents: read
+
+jobs:
+  label-check:
+    name: Label Check
+    uses: ApolloAutomation/Workflows/.github/workflows/label-check.yml@main