🦋 New version release

CHANGELOG.md

@@ -1,5 +1,15 @@
 # Changelog
 
+## 0.13.0
+
+### Minor Changes
+
+- 92b31ad: Add an opt-in `showToolActivity` option that renders the agent's tool activity (grep / read_file / bash / knowledge_search…) as inline chips interleaved with its prose, mirroring the smooth daemon SPA's `blocks` model.
+
+  - Defaults to **`false`** — a customer-facing support widget stays byte-for-byte unchanged (prose bubble only). Enable via the `show-tool-activity` HTML attribute or `showToolActivity: true` in the config.
+  - When enabled, an assistant turn that invokes a tool renders as an ordered strip of prose bubbles + tool chips (running… / done ✓ / error), in the order the model produced them. Tool activity is read from `state.rawResponse.toolCall` / `state.rawResponse.toolResult` (the correct nested path — reading `state.toolResult` leaves chips stuck "running…").
+  - Tool name/args are rendered via `textContent` (never `innerHTML`), so a tool payload can't inject markup. New public `MessageBlock` / `ToolCall` types.
+
 ## 0.12.0
 
 ### Minor Changes

package.json

@@ -1,6 +1,6 @@
 {
     "name": "@smooai/chat-widget",
-    "version": "0.12.0",
+    "version": "0.13.0",
     "description": "Embeddable AI chat as a framework-light web component — the Aurora Glass design, streaming replies, grounded sources, and per-brand theming. Speaks the smooth-operator WebSocket protocol.",
     "license": "MIT",
     "author": "SmooAI",