feat: alert on agent failure/recovery via Raven

Tracks agent statuses in service-states.json alongside services.
Sends critical alert when agent status changes to failure (includes
result/error message from last-run.json). Sends recovery alert on
failure → success transition.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

main.py

@@ -20,17 +20,24 @@ CONFIG_FILE = Path(__file__).parent / "config.yaml"
 def _load_prev_states() -> dict:
     path = AGENT_OS_DIR / "logs" / "varys-monitor" / "service-states.json"
     if not path.exists():
-        return {}
+        return {"services": {}, "agents": {}}
     try:
-        return json.loads(path.read_text())
+        data = json.loads(path.read_text())
+        # migrate old flat format (services only)
+        if "services" not in data:
+            return {"services": data, "agents": {}}
+        return data
     except Exception:
-        return {}
+        return {"services": {}, "agents": {}}
 
 
-def _save_states(services: list):
+def _save_states(services: list, agents: list):
     path = AGENT_OS_DIR / "logs" / "varys-monitor" / "service-states.json"
     path.parent.mkdir(parents=True, exist_ok=True)
-    path.write_text(json.dumps({s["name"]: s["status"] for s in services}))
+    path.write_text(json.dumps({
+        "services": {s["name"]: s["status"] for s in services},
+        "agents": {a["name"]: a["status"] for a in agents},
+    }))
 
 
 def _notify_raven(message: str, severity: str):
@@ -179,13 +186,25 @@ def main():
         agents.append(result)
 
     prev_states = _load_prev_states()
+    prev_services = prev_states.get("services", {})
+    prev_agents = prev_states.get("agents", {})
+
     for s in services:
-        prev = prev_states.get(s["name"])
+        prev = prev_services.get(s["name"])
         if prev and prev != "down" and s["status"] == "down":
             _notify_raven(f"{s['name']} is DOWN", "critical")
         elif prev == "down" and s["status"] not in ("down", "degraded"):
             _notify_raven(f"{s['name']} recovered (UP)", "info")
-    _save_states(services)
+
+    for a in agents:
+        prev = prev_agents.get(a["name"])
+        if a["status"] == "failure" and prev != "failure":
+            detail = a.get("result", "no details")
+            _notify_raven(f"{a['name']} failed: {detail}", "critical")
+        elif prev == "failure" and a["status"] == "success":
+            _notify_raven(f"{a['name']} recovered (success)", "info")
+
+    _save_states(services, agents)
 
     out_dir = SITES_DIR / "varys"
     out_dir.mkdir(parents=True, exist_ok=True)