@tiens.nguyen/gonext-local-worker 1.0.216 → 1.0.217

package/gonext-repl.mjs

@@ -965,9 +965,19 @@ async function runAgentTurn(history) {
       jobRunning = job.jobStatus === "running";
       const text = typeof job.resultText === "string" ? job.resultText : "";
       if (job.jobStatus === "completed") {
-        // Drop the transient status line and any partial reasoning tail. If this was a
-        // plain reply, its text is ALREADY fully visible (streamed live above) — the
-        // caller must not print it again; `shown` tells it so.
+        // The job can flip to "completed" between polls with authoritative resultText that
+        // the live stream hasn't caught up to yet (a debounced last chunk, or the final
+        // PATCH landing before its chunk flush was polled). resultText is the source of
+        // truth, so flush any not-yet-shown suffix BEFORE returning — otherwise a plain
+        // reply, whose live stream is its ONLY output, renders truncated mid-word. Gated
+        // to plainReplyFlow: the agent path reprints its answer from resultText via the
+        // caller (shown=false), so a live flush there is unnecessary (and would compound
+        // the separate agent double-print). This runs for the completed poll only; the
+        // running-branch consume below handles every earlier poll.
+        if (plainReplyFlow && text.length > shownChars) {
+          consume(text.slice(shownChars));
+          shownChars = text.length;
+        }
         clearStatus();
         return { text: answerFrom(text), shown: answerShownLive };
       }

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@tiens.nguyen/gonext-local-worker",
-  "version": "1.0.216",
+  "version": "1.0.217",
   "description": "Polls GoNext cloud API for async local LLM jobs and runs them against Ollama/OpenAI-compatible servers on this Mac",
   "type": "module",
   "license": "MIT",