diff --git a/.github/workflows/publish-claude-plugin.yml b/.github/workflows/publish-claude-plugin.yml
index 4d54bb9b4..b76ef5479 100644
--- a/.github/workflows/publish-claude-plugin.yml
+++ b/.github/workflows/publish-claude-plugin.yml
@@ -1,6 +1,11 @@
 # publish-claude-plugin syncs the generated marketplace bundle from
 # this repo to gortexhq/claude-plugin on every gortex release.
 #
+# release.yml dispatches it after each release: releases are created
+# with GITHUB_TOKEN, which does not fire the `release: published`
+# trigger below. That trigger only covers releases published without
+# GITHUB_TOKEN, such as by hand.
+#
 # How it works:
 #   1. Checkout gortex at the released tag
 #   2. Regenerate the bundle via `make claude-plugin`
diff --git a/.github/workflows/release.yml b/.github/workflows/release.yml
index 9ccefd070..a2051de8d 100644
--- a/.github/workflows/release.yml
+++ b/.github/workflows/release.yml
@@ -786,3 +786,22 @@ jobs:
             ./dist/*.rpm
             ./dist/*.apk
           update_release_body: true
+
+  # Releases are created with GITHUB_TOKEN, and events from GITHUB_TOKEN
+  # do not start other workflows, so publish-claude-plugin's
+  # `release: published` trigger never fires for them. workflow_dispatch
+  # is the documented exception, so start it explicitly once the
+  # release exists.
+  publish-claude-plugin:
+    needs: release
+    runs-on: ubuntu-latest
+    permissions:
+      actions: write
+    steps:
+      - name: Dispatch publish-claude-plugin
+        env:
+          GH_TOKEN: ${{ secrets.GITHUB_TOKEN }}
+        run: |
+          gh workflow run publish-claude-plugin.yml \
+            --repo "$GITHUB_REPOSITORY" \
+            -f ref="$GITHUB_REF_NAME"
diff --git a/scripts/validate-cask.sh b/scripts/validate-cask.sh
index 25fafef92..da5d68527 100755
--- a/scripts/validate-cask.sh
+++ b/scripts/validate-cask.sh
@@ -41,13 +41,32 @@ rendered="$("$repo_root/scripts/render-cask.sh" 0.0.0 \
 # untouched, and the trap removes it even when brew fails.
 tap_user="gortexcaskcheck"
 tap_dir="$(brew --repository)/Library/Taps/$tap_user/homebrew-tap"
-cleanup() { rm -rf "$(brew --repository)/Library/Taps/$tap_user"; }
+trust_home=""
+cleanup() {
+	rm -rf "$(brew --repository)/Library/Taps/$tap_user"
+	if [ -n "$trust_home" ]; then rm -rf "$trust_home"; fi
+}
 trap cleanup EXIT
 
 cleanup
 mkdir -p "$tap_dir/Casks"
 printf '%s\n' "$rendered" >"$tap_dir/Casks/gortex.rb"
 
+# Current Homebrew refuses to load casks from a tap nobody has trusted, and
+# `brew info` does not trust it on the way like `brew install user/tap/x`
+# does. Trust the scratch tap in a throwaway trust store (brew keeps it under
+# $XDG_CONFIG_HOME/homebrew when that is set), so the developer's own trust
+# store and brew.env are neither read nor changed. HOMEBREW_NO_REQUIRE_TAP_TRUST
+# is deprecated, and its warning would trip the check below, so it is unset
+# too in case a developer exported it to get past this same error. Older brew
+# has no `trust` command and needs none.
+trust_home="$(mktemp -d)"
+export XDG_CONFIG_HOME="$trust_home"
+unset HOMEBREW_NO_REQUIRE_TAP_TRUST HOMEBREW_REQUIRE_TAP_TRUST
+if brew trust --help >/dev/null 2>&1; then
+	brew trust --tap "$tap_user/tap"
+fi
+
 echo "==> loading the rendered cask through brew"
 # `brew info --cask` evaluates every stanza, so an unknown one (the #639 bug)
 # fails here. Output is kept: a reviewer should see the artifact list change