@tiens.nguyen/gonext-local-worker 1.0.86 → 1.0.88

package/gonext_agent_chat.py

@@ -330,6 +330,10 @@ def run_agent_chat(cfg):
         _emit({"type": "final", "text": "[No user message found in history]"})
         return
     task_text = (messages[last_user_idx].get("content") or "").strip()
+    # Routing must look at the CURRENT message alone — not the history-laden blob
+    # below. Otherwise the keyword router matches URLs/"api"/"GET" from prior turns
+    # and fires the agent on trivial replies like "thanks" or "good".
+    latest_user_text = task_text
 
     # Walk prior turns newest-first, keeping condensed lines until the budget is
     # spent, then restore chronological (oldest→newest) order.
@@ -358,12 +362,14 @@ def run_agent_chat(cfg):
 
     _log(
         f"history: {len(history_lines)} prior turn(s), {used} chars "
-        f"(budget {HISTORY_CHAR_BUDGET})"
+        f"(budget {HISTORY_CHAR_BUDGET}) — exact turns sent to the agent below:"
     )
-    _log(f"task={task_text[:120]!r}")
+    for j, ln in enumerate(history_lines):
+        _log(f"  history[{j}]: {ln[:240]}")
+    _log(f"current task (latest user message): {task_text.rsplit('Current task: ', 1)[-1][:240]!r}")
 
     # Route: ask the model if this task needs HTTP tool use.
-    needs_agent = _route(task_text, agent_base_url, agent_api_key, agent_model_id)
+    needs_agent = _route(latest_user_text, agent_base_url, agent_api_key, agent_model_id)
 
     if not needs_agent:
         _log("router: plain chat (no HTTP needed)")

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@tiens.nguyen/gonext-local-worker",
-  "version": "1.0.86",
+  "version": "1.0.88",
   "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",