feat: add raven stub, state tracking, sam-research + searxng monitoring

- RAVEN_URL env var: sends down/recovery alerts when Raven is live (silent no-op until then)
- service-states.json persists prev state for change detection
- config.yaml: adds sam-research + searxng services, adds sam/citadel/varys agents

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

config.yaml

@@ -8,6 +8,10 @@ services:
     url: http://hodor-gateway:8200/health
   - name: citadel-mcp
     url: http://citadel-mcp:8300/sse
+  - name: sam-research
+    url: http://sam-research:8500/health
+  - name: searxng
+    url: http://searxng:8080
 
   # Core services
   - name: open-webui
@@ -34,3 +38,6 @@ services:
 # One-shot agents to watchdog (reads /opt/agent-os/logs/<name>/last-run.json).
 agents:
   - name: bran-changelog
+  - name: sam-research
+  - name: citadel-mcp
+  - name: varys-monitor

main.py

@@ -13,9 +13,36 @@ logger = logging.getLogger("varys")
 
 SITES_DIR = Path(os.getenv("SITES_DIR", "/opt/sites"))
 AGENT_OS_DIR = Path(os.getenv("AGENT_OS_DIR", "/opt/agent-os"))
+RAVEN_URL = os.getenv("RAVEN_URL", "")
 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 {}
+    try:
+        return json.loads(path.read_text())
+    except Exception:
+        return {}
+
+
+def _save_states(services: 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}))
+
+
+def _notify_raven(message: str, severity: str):
+    if not RAVEN_URL:
+        return
+    try:
+        httpx.post(f"{RAVEN_URL}/notify", json={"message": message, "severity": severity, "source": "varys"}, timeout=5)
+        logger.info(f"raven notified: {message}")
+    except Exception as e:
+        logger.warning(f"raven notify failed (raven not live yet?): {e}")
+
+
 def check_service(name: str, url: str) -> dict:
     start = time.monotonic()
     try:
@@ -151,6 +178,15 @@ def main():
         logger.info(f"agent {agent['name']}: {result['status']}")
         agents.append(result)
 
+    prev_states = _load_prev_states()
+    for s in services:
+        prev = prev_states.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)
+
     out_dir = SITES_DIR / "varys"
     out_dir.mkdir(parents=True, exist_ok=True)
     (out_dir / "index.html").write_text(render_html(services, agents))

run.sh

@@ -12,4 +12,5 @@ docker run --rm \
   --volume /opt/agent-os:/opt/agent-os \
   --env SITES_DIR=/opt/sites \
   --env AGENT_OS_DIR=/opt/agent-os \
+  --env RAVEN_URL= \
   varys-monitor