v0.14.4: Interpreter wired in v2, tool_call convention, Haiku models, UI fix

- Wire Interpreter into v2 pipeline (after Thinker tool_output, before Output)
- Rename tool_exec -> tool_call everywhere (consistent convention across v1/v2)
- Switch Director v1+v2 to anthropic/claude-haiku-4.5 (was opus, reserved)
- Fix UI apply_machine_ops crash when states are strings instead of dicts
- Fix runtime_test.py async poll to match on message ID (prevent stale results)
- Add traceback to pipeline error logging

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

agent/api.py

@@ -40,7 +40,7 @@ def _broadcast_sse(event: dict):
     if _pipeline_result.get("status") == "running":
         node = event.get("node", "")
         evt = event.get("event", "")
-        if node and evt in ("thinking", "perceived", "decided", "streaming", "tool_exec", "interpreted", "updated"):
+        if node and evt in ("thinking", "perceived", "decided", "streaming", "tool_call", "interpreted", "updated"):
             _pipeline_result["stage"] = node
             _pipeline_result["event"] = evt
 
@@ -199,7 +199,8 @@ def register_routes(app):
                     "memorizer": runtime.memorizer.state,
                 }
             except Exception as e:
-                log.error(f"[api] pipeline error: {e}")
+                import traceback
+                log.error(f"[api] pipeline error: {e}\n{traceback.format_exc()}")
                 _pipeline_result = {
                     "status": "error",
                     "id": msg_id,

agent/nodes/director_v1.py

@@ -12,7 +12,7 @@ log = logging.getLogger("runtime")
 class DirectorNode(Node):
     name = "director"
     model = "google/gemini-2.0-flash-001"
-    plan_model = "anthropic/claude-opus-4"  # Smart model for investigation planning
+    plan_model = "anthropic/claude-haiku-4.5"  # Smart model for investigation planning
     max_context_tokens = 2000
 
     SYSTEM = """You are the Director node — the strategist of this cognitive runtime.

agent/nodes/director_v2.py

@@ -12,7 +12,7 @@ log = logging.getLogger("runtime")
 
 class DirectorV2Node(Node):
     name = "director_v2"
-    model = "anthropic/claude-opus-4"
+    model = "anthropic/claude-haiku-4.5"
     max_context_tokens = 4000
 
     SYSTEM = """You are the Director — the brain of this cognitive agent runtime.

agent/nodes/thinker_v2.py

@@ -79,7 +79,7 @@ Rules:
         for step in plan.tool_sequence:
             tool = step.get("tool", "")
             args = step.get("args", {})
-            await self.hud("tool_exec", tool=tool, args=args)
+            await self.hud("tool_call", tool=tool, args=args)
 
             if tool == "emit_actions":
                 actions.extend(args.get("actions", []))

agent/nodes/ui.py

@@ -34,6 +34,8 @@ class UINode(Node):
                 states_list = op_data.get("states", [])
                 states = {}
                 for s in states_list:
+                    if isinstance(s, str):
+                        s = {"name": s}
                     name = s.get("name", "")
                     if name:
                         states[name] = {

agent/runtime.py

@@ -219,6 +219,10 @@ class Runtime:
         if self.is_v2:
             plan = await self.director.decide(command, self.history, memory_context=mem_ctx)
             thought = await self.thinker.process(command, plan, self.history, memory_context=mem_ctx)
+            if self.interpreter and thought.tool_used and thought.tool_output:
+                interpreted = await self.interpreter.interpret(
+                    thought.tool_used, thought.tool_output, action_desc)
+                thought.response = interpreted.summary
         else:
             thought = await self.thinker.process(command, self.history, memory_context=mem_ctx)
 
@@ -335,9 +339,15 @@ class Runtime:
             return
 
         if self.is_v2:
-            # v2 flow: Director decides, Thinker executes
+            # v2 flow: Director decides, Thinker executes, Interpreter reads results
             plan = await self.director.decide(command, self.history, memory_context=mem_ctx)
             thought = await self.thinker.process(command, plan, self.history, memory_context=mem_ctx)
+            # Interpreter: factual summary of tool results (no hallucination)
+            if self.interpreter and thought.tool_used and thought.tool_output:
+                interpreted = await self.interpreter.interpret(
+                    thought.tool_used, thought.tool_output, text)
+                # Replace thinker's response with interpreter's factual summary
+                thought.response = interpreted.summary
         else:
             # v1 flow: optional Director pre-planning for complex requests
             is_complex = command.analysis.complexity == "complex"

runtime_test.py

@@ -153,6 +153,19 @@ class CogClient:
             body["dashboard"] = dashboard
         r = self.client.post(f"{API}/send", json=body, headers=HEADERS)
         d = r.json()
+        # Async send: poll for result, match on message ID
+        if d.get("status") == "queued":
+            msg_id = d.get("id", "")
+            for _ in range(120):
+                time.sleep(0.5)
+                pr = self.client.get(f"{API}/result", headers=HEADERS)
+                pd = pr.json()
+                if pd.get("id") == msg_id and pd.get("status") == "done":
+                    d = pd
+                    break
+                if pd.get("id") == msg_id and pd.get("status") == "error":
+                    d = pd
+                    break
         self.last_response = d.get("response", "")
         self.last_memo = d.get("memorizer", {})
         time.sleep(0.5)
@@ -477,13 +490,22 @@ def run_standalone(paths: list[Path] = None):
     if not paths:
         paths = sorted(Path("testcases").glob("*.md"))
 
+    # Count total steps across all testcases for frontend progress
+    all_tcs = [parse_testcase(p) for p in paths]
+    total_steps = sum(len(s["commands"]) for tc in all_tcs for s in tc["steps"])
+    first_suite = True
+
     all_results = {}
-    for path in paths:
+    for tc in all_tcs:
-        tc = parse_testcase(path)
+        path = tc["file"]
         print(f"\n{'='*60}")
         print(f"  {tc['name']}")
         print(f"{'='*60}")
-        _push_status("suite_start", suite=tc["name"])
+        if first_suite:
+            _push_status("suite_start", suite=tc["name"], count=total_steps)
+            first_suite = False
+        else:
+            _push_status("suite_start", suite=tc["name"])
 
         runner = CogTestRunner()
         results = runner.run(tc)

static/app.js

@@ -397,7 +397,6 @@ function graphAnimate(event, node) {
       if (node) pulseNode(node);
       break;
     case 'tool_call':
-    case 'tool_exec':
       pulseNode(node || 'thinker'); flashEdge('thinker', 'ui');
       break;
     case 'tool_result':

test_nodes/test_thinker_v2.py

@@ -183,8 +183,8 @@ async def test_emits_hud_per_tool():
 
     with patch("agent.nodes.thinker_v2.llm_call", side_effect=mock_llm):
         await node.process(cmd, plan, [], memory_context="")
-    tool_events = hud.find("tool_exec")
+    tool_events = hud.find("tool_call")
-    assert len(tool_events) >= 2, f"expected 2+ tool_exec events, got {len(tool_events)}"
+    assert len(tool_events) >= 2, f"expected 2+ tool_call events, got {len(tool_events)}"
 
 
 async def test_create_machine_tool():