fix(reward-memory): use effective Agent automation in prompts and settings

loopx/capabilities/reward_memory/configuration.py

@@ -56,6 +56,8 @@ def reward_memory_goal_configuration_summary(
         and receipt.get("exact_readback_verified") is True
     )
     effective_verified_agents: list[str] = []
+    configured_automatic_ingest = False
+    configured_automatic_recall = False
     for agent_id in recorded_verified_agents:
         try:
             status, _ = resolve_goal_reward_memory_experiment(
@@ -65,6 +67,14 @@ def reward_memory_goal_configuration_summary(
             continue
         if status.get("available") is True:
             effective_verified_agents.append(agent_id)
+            configured_automatic_ingest = (
+                configured_automatic_ingest
+                or status.get("automatic_ingest") is True
+            )
+            configured_automatic_recall = (
+                configured_automatic_recall
+                or status.get("automatic_recall") is True
+            )
     effective_available = bool(effective_verified_agents)
     return {
         "enabled": policy["enabled"],
@@ -75,17 +85,46 @@ def reward_memory_goal_configuration_summary(
         "experimental": policy["experimental"],
         "config_pointer_registered": bool(policy["config_path"]),
         "binding_revision": binding_revision,
-        "automatic_ingest": effective_available
-        and policy["automation"].get("automatic_ingest") is True,
-        "automatic_recall": effective_available
-        and policy["automation"].get("automatic_recall") is True,
+        "automatic_ingest": policy["automation"].get("automatic_ingest") is True
+        and configured_automatic_ingest,
+        "automatic_recall": policy["automation"].get("automatic_recall") is True
+        and configured_automatic_recall,
         "automation_intent": dict(policy["automation_intent"]),
         "host_coverage": reward_memory_host_coverage(),
         "enabled_agents": list(policy["enabled_agents"]),
         "enablement_verified_agents": effective_verified_agents,
     }
 
 
+def reward_memory_automatic_ingest_for_agent(
+    goal: Mapping[str, Any], agent_id: str | None
+) -> bool:
+    """Return whether one Agent may receive automatic-ingest prompt guidance.
+
+    Desired policy is a cheap off switch. Active admission reuses the runtime
+    resolver so the prompt requires this exact Agent's current config binding
+    and verified enablement receipt. The resolver is read-only and never calls
+    the configured provider.
+    """
+
+    policy = reward_memory_goal_policy(goal)
+    if (
+        not policy["enabled"]
+        or policy["automation"].get("automatic_ingest") is not True
+        or not isinstance(agent_id, str)
+        or not agent_id.strip()
+    ):
+        return False
+    try:
+        status, _ = resolve_goal_reward_memory_experiment(
+            goal=goal, agent_id=agent_id
+        )
+    except ValueError:
+        # Legacy alias-only registrations and malformed identities fail closed.
+        return False
+    return status.get("available") is True and status.get("automatic_ingest") is True
+
+
 def plan_reward_memory_goal_configuration(
     *,
     goal: Mapping[str, Any],

loopx/cli_commands/support_control.py

@@ -15,7 +15,6 @@
     DEFAULT_CHAT_PORT,
     serve_chat,
 )
-from ..control_plane.reward_memory import reward_memory_goal_policy
 from ..control_plane.scheduler.execution_context import SchedulerRuntimeProfile
 from ..dashboard_launcher import launch_dashboard, replace_existing_loopx_chat
 from ..execution_profile import execution_profile_turn_granularity
@@ -313,13 +312,6 @@ def handle_support_control_command(
                 if isinstance(registry_goal, dict)
                 else None
             )
-            reward_memory_policy = reward_memory_goal_policy(
-                registry_goal if isinstance(registry_goal, dict) else {}
-            )
-            reward_memory_enabled = bool(
-                reward_memory_policy["enabled"]
-                and reward_memory_policy["automation"].get("automatic_ingest") is True
-            )
             agent_profile = None
             if args.agent_id:
                 effective_agent_id = require_registered_agent_id(
@@ -331,6 +323,16 @@ def handle_support_control_command(
                 agent_profile = agent_profile_from_registry(
                     agent_registry_path, args.goal_id, effective_agent_id
                 )
+            reward_memory_enabled = False
+            if effective_agent_id is not None:
+                from ..capabilities.reward_memory.configuration import (
+                    reward_memory_automatic_ingest_for_agent,
+                )
+
+                reward_memory_enabled = reward_memory_automatic_ingest_for_agent(
+                    registry_goal if isinstance(registry_goal, dict) else {},
+                    effective_agent_id,
+                )
             explicit_scheduler_fields = (
                 args.host_surface,
                 args.scheduler_owner,

loopx/semantics/project_registry_io_manifest_v1.json

@@ -2287,15 +2287,15 @@
     },
     {
       "site": "loopx/upgrade.py::<module>.build_upgrade_plan::codec_read:load_registry#1",
-      "line": 741,
+      "line": 744,
       "column": 16,
       "kind": "codec_read",
       "api": "load_registry",
       "classification": "codec_api"
     },
     {
       "site": "loopx/upgrade.py::<module>.build_upgrade_plan::codec_read:load_registry#2",
-      "line": 747,
+      "line": 750,
       "column": 24,
       "kind": "codec_read",
       "api": "load_registry",

loopx/upgrade.py

@@ -14,7 +14,6 @@
 )
 from .execution_profile import execution_profile_turn_granularity
 from .heartbeat_prompt import build_heartbeat_prompt
-from .control_plane.reward_memory import reward_memory_goal_policy
 from .history import load_registry
 from .paths import global_registry_path, resolve_runtime_root, select_default_runtime_root
 from .registry import registry_goals, resolve_state_file
@@ -85,11 +84,15 @@ def goal_heartbeat_prompt(
     repo = Path(str(goal.get("repo") or ".")).expanduser()
     state_file = resolve_state_file(repo, goal.get("state_file"))
     registered_agents = registered_agent_ids_for_goal(goal)
-    reward_memory_policy = reward_memory_goal_policy(goal)
-    reward_memory_enabled = bool(
-        reward_memory_policy["enabled"]
-        and reward_memory_policy["automation"].get("automatic_ingest") is True
-    )
+    reward_memory_enabled = False
+    if agent_id is not None:
+        from .capabilities.reward_memory.configuration import (
+            reward_memory_automatic_ingest_for_agent,
+        )
+
+        reward_memory_enabled = reward_memory_automatic_ingest_for_agent(
+            goal, agent_id
+        )
     return build_heartbeat_prompt(
         goal_id=goal_id,
         active_state=None,