diff --git a/runner/harness/lifecycle.py b/runner/harness/lifecycle.py
index c33941a..b18b85d 100644
--- a/runner/harness/lifecycle.py
+++ b/runner/harness/lifecycle.py
@@ -295,18 +295,25 @@ def previous_version(recipe: str) -> str | None:
     return vers[-2] if len(vers) >= 2 else None
 
 
-def _app_container(domain: str, service: str = "app") -> str:
-    """The running container id for <stack>_<service>."""
+def _app_container(domain: str, service: str = "app", timeout: int = 60) -> str:
+    """The running container id for <stack>_<service>, with a BOUNDED POLL for it to (re)appear.
+    A lifecycle op can briefly leave no running task — notably `abra app backup create`, where
+    backup-bot-two stops/cycles the app container, so a mutate exec right after backup hit an empty
+    `docker ps` and raised. Poll (no bare sleep) until the container is back or timeout."""
     name = f"{_stack_name(domain)}_{service}"
-    proc = subprocess.run(
-        ["docker", "ps", "--filter", f"name={name}", "--format", "{{.ID}}"],
-        capture_output=True,
-        text=True,
-    )
-    cid = proc.stdout.strip().split("\n")[0]
-    if not cid:
-        raise RuntimeError(f"no running container for {name}")
-    return cid
+    deadline = time.time() + timeout
+    while True:
+        proc = subprocess.run(
+            ["docker", "ps", "--filter", f"name={name}", "--format", "{{.ID}}"],
+            capture_output=True,
+            text=True,
+        )
+        cid = proc.stdout.strip().split("\n")[0]
+        if cid:
+            return cid
+        if time.time() >= deadline:
+            raise RuntimeError(f"no running container for {name} after {timeout}s")
+        time.sleep(3)
 
 
 def exec_in_app(domain: str, cmd: list[str], service: str = "app") -> str: