atlasbot: improve metric parsing and cluster intent

services/comms/scripts/atlasbot/bot.py

@@ -190,6 +190,8 @@ _INSIGHT_HINT_WORDS = {
     "surprising",
     "weird",
     "odd",
+    "unusual",
+    "outlier",
     "fun",
     "cool",
     "unique",
@@ -540,6 +542,13 @@ def _detect_operation(q: str) -> str | None:
 
 def _detect_metric(q: str) -> str | None:
     tokens = set(_tokens(q))
+    expanded: set[str] = set(tokens)
+    for token in list(tokens):
+        for part in re.split(r"[-_]", token):
+            part = part.strip()
+            if len(part) >= 2:
+                expanded.add(part)
+    tokens = expanded
     for metric, phrases in METRIC_HINTS.items():
         for phrase in phrases:
             if " " in phrase:
@@ -1271,6 +1280,19 @@ def snapshot_metric_answer(
         pending = metrics.get("pods_pending")
         failed = metrics.get("pods_failed")
         succeeded = metrics.get("pods_succeeded")
+        status_terms = ("running", "pending", "failed", "succeeded", "completed")
+        if sum(1 for term in status_terms if term in q) > 1:
+            parts = []
+            if running is not None:
+                parts.append(f"running {running:.0f}")
+            if pending is not None:
+                parts.append(f"pending {pending:.0f}")
+            if failed is not None:
+                parts.append(f"failed {failed:.0f}")
+            if succeeded is not None:
+                parts.append(f"succeeded {succeeded:.0f}")
+            if parts:
+                return _format_confidence(f"Pods: {', '.join(parts)}.", "high")
         if "pending" in q and pending is not None:
             return _format_confidence(f"Pending pods: {pending:.0f}.", "high")
         if "failed" in q and failed is not None:
@@ -1345,7 +1367,17 @@ def structured_answer(
         if hw_line:
             return _format_confidence(hw_line, "high")
 
-    if op == "top" and metric is None:
+    if entity == "node" and op == "status" and metric is None:
+        summary = _nodes_summary_line(inventory, snapshot)
+        if summary:
+            return _format_confidence(summary, "high")
+
+    if entity == "node" and metric is None and any(word in q for word in ("hardware", "architecture", "class", "mix")):
+        hw_line = _hardware_mix_line(inventory)
+        if hw_line:
+            return _format_confidence(hw_line, "medium")
+
+    if op == "top" and metric is None and not any(word in q for word in ("hardware", "architecture", "class")):
         metric = "cpu"
 
     # Metrics-first when a metric or top operation is requested.
@@ -2807,8 +2839,11 @@ class _AtlasbotHandler(BaseHTTPRequestHandler):
             inventory=inventory,
             workloads=workloads,
         )
-        cluster_query = _is_cluster_query(cleaned, inventory=inventory, workloads=workloads) or (
+        cluster_query = (
-            _is_subjective_query(cleaned) and history_cluster
+            _is_cluster_query(cleaned, inventory=inventory, workloads=workloads)
+            or history_cluster
+            or _knowledge_intent(cleaned)
+            or _is_subjective_query(cleaned)
         )
         context = ""
         if cluster_query:
@@ -3347,8 +3382,11 @@ def sync_loop(token: str, room_id: str):
                     inventory=inventory,
                     workloads=workloads,
                 )
-                cluster_query = _is_cluster_query(cleaned_body, inventory=inventory, workloads=workloads) or (
+                cluster_query = (
-                    _is_subjective_query(cleaned_body) and history_cluster
+                    _is_cluster_query(cleaned_body, inventory=inventory, workloads=workloads)
+                    or history_cluster
+                    or _knowledge_intent(cleaned_body)
+                    or _is_subjective_query(cleaned_body)
                 )
                 context = ""
                 if cluster_query: