diff --git a/ci/publish-release.sh b/ci/publish-release.sh
index 5ae3649..5531dc7 100755
--- a/ci/publish-release.sh
+++ b/ci/publish-release.sh
@@ -65,6 +65,35 @@ tested_wc=$(ci_plugin_header 'WC tested up to')
 body=$(printf '%s\n\n---\n\nTested up to: %s\nWC tested up to: %s\n' \
   "$notes" "${tested_wp:-unknown}" "${tested_wc:-unknown}")
 
+# Read one top-level field out of a JSON response.
+#
+# Not sed. A greedy `.*"id":([0-9]+)` matches the LAST id in the payload, and a
+# Gitea release carries both an embedded "author" object and, once anything is
+# attached, an "assets" array — each with an id of its own. Measured on this
+# repo: the four published releases have ids 416/417/419/421, and the greedy
+# pattern returns 1/2/3/4, which are the asset ids.
+#
+# Those four releases were nonetheless published correctly, and only by luck:
+# the author id on this instance is **-1**, and the minus sign stops `[0-9]+`
+# from matching, so on a create response — which has no assets yet — the last
+# numeric id happened to be the release's. A real author id, or a re-run
+# against a release that already has its zip, and it picks the wrong number.
+#
+# Found by the session porting this into subscriptions, where it picked the
+# author id and the upload 404'd. python3 is already a required tool here.
+json_field() { # field  (JSON on stdin)
+  python3 -c '
+import json, sys
+try:
+    data = json.load( sys.stdin )
+except Exception:
+    sys.exit( 0 )
+value = data.get( sys.argv[1] )
+if value is not None:
+    print( value )
+' "$1"
+}
+
 api() { # method path [data]
   curl -fsS -X "$1" \
     -H "Authorization: token $GH_TOKEN" \
@@ -94,20 +123,42 @@ print(json.dumps({
   # A re-run of a release must not fail on "already exists".
   existing=$(curl -fsS -H "Authorization: token $GH_TOKEN" \
     "$GITHUB_API_URL/repos/$owner/$repo/releases/tags/$TAG" 2>/dev/null || true)
-  id=$(printf '%s' "$existing" | sed -n -E 's/.*"id":[[:space:]]*([0-9]+).*/\1/p' | head -1)
+  id=$(printf '%s' "$existing" | json_field id)
 
   if [ -z "$id" ]; then
     created=$(api POST "$GITHUB_API_URL/repos/$owner/$repo/releases" "$payload")
-    id=$(printf '%s' "$created" | sed -n -E 's/.*"id":[[:space:]]*([0-9]+).*/\1/p' | head -1)
+    id=$(printf '%s' "$created" | json_field id)
   fi
   [ -n "$id" ] || { echo "Could not create or find the release for $TAG." >&2; exit 1; }
   echo "  release id $id"
 
-  curl -fsS -X POST \
-    -H "Authorization: token $GH_TOKEN" \
-    -F "attachment=@$zip" \
-    "$GITHUB_API_URL/repos/$owner/$repo/releases/$id/assets?name=$PLUGIN_SLUG.zip" >/dev/null
-  echo "  attached $PLUGIN_SLUG.zip"
+  # Before uploading anything to it, confirm the id really is this tag's. The
+  # failure being guarded against is not a 404 — that is loud — but an id that
+  # exists and belongs to something else, which would attach this plugin's zip
+  # to an unrelated release with every check still green.
+  release=$(curl -fsS -H "Authorization: token $GH_TOKEN" \
+    "$GITHUB_API_URL/repos/$owner/$repo/releases/$id")
+  actual=$(printf '%s' "$release" | json_field tag_name)
+  if [ "$actual" != "$TAG" ]; then
+    echo "Release id $id is tagged '$actual', not '$TAG'. Refusing to upload the zip to it." >&2
+    exit 1
+  fi
+
+  # A re-run must not fail on the half that already succeeded: the upload
+  # rejects a duplicate asset name.
+  if printf '%s' "$release" | python3 -c '
+import json, sys
+data = json.load( sys.stdin )
+sys.exit( 0 if any( a.get( "name" ) == sys.argv[1] for a in data.get( "assets", [] ) ) else 1 )
+' "$PLUGIN_SLUG.zip"; then
+    echo "  $PLUGIN_SLUG.zip is already attached — leaving it alone"
+  else
+    curl -fsS -X POST \
+      -H "Authorization: token $GH_TOKEN" \
+      -F "attachment=@$zip" \
+      "$GITHUB_API_URL/repos/$owner/$repo/releases/$id/assets?name=$PLUGIN_SLUG.zip" >/dev/null
+    echo "  attached $PLUGIN_SLUG.zip"
+  fi
 
   release_api="$GITHUB_API_URL/repos/$owner/$repo/releases/latest"
 fi
diff --git a/docs/REUSE.md b/docs/REUSE.md
index bc93110..648df72 100644
--- a/docs/REUSE.md
+++ b/docs/REUSE.md
@@ -139,6 +139,15 @@ and a release with no zip attached both produce a page that looks perfect to
 anyone logged in, and an update that no shop is ever offered. A check that
 sends a token is the false positive that hides exactly that.
 
+**Do not parse a forge's JSON with `sed`.** A greedy `.*"id":([0-9]+)` matches
+the *last* id in the payload, and a release object carries an embedded
+`author` and, once anything is attached, an `assets` array — each with an id of
+its own. `ci/publish-release.sh` uses a four-line `json_field()` over `python3`
+instead. Measured here: the greedy pattern returns the asset ids 1/2/3/4 for
+releases whose real ids are 416/417/419/421. Four releases were published
+correctly anyway, purely because this instance's author id is **-1** and the
+minus sign stops `[0-9]+` from matching. That is not a safety margin.
+
 **A forge's own download URL is not the address your users can reach.**
 `browser_download_url` is built from the forge's configured root URL;
 `gitea.nolimit.dk` reports `http://57.129.89.117:25001/...` there. The updater