feat: add agent visibility APIs and harden runtime verification

Add Day 4 visibility endpoints and response models, strengthen collaboration/task verification behavior, and patch conversation schema startup migration for agent_state compatibility. Extend backend regression coverage for runtime schemas, verifier behavior, visibility APIs, router auth, and legacy conversation list loading.

backend/app/agents/state.py

@@ -3,8 +3,9 @@ from enum import Enum
 from typing import Annotated, Any, Literal, TypedDict
 
 from app.agents.schemas.event import AgentEvent
-from app.agents.schemas.task import AgentTask, TaskResult, VerificationStatus
-from langchain_core.messages import BaseMessage
+from app.agents.schemas.message import AgentMessage
+from app.agents.schemas.task import AgentTask, CollaborationBudget, InterruptRecord, RecoveryRecord, TaskResult, VerificationStatus
+from langchain_core.messages import AIMessage, BaseMessage, HumanMessage
 from langgraph.graph.message import add_messages
 
 
@@ -24,12 +25,27 @@ class ConversationTurn:
     model: str | None = None
 
 
+def turn_to_message(turn: ConversationTurn) -> BaseMessage:
+    if turn.role == "user":
+        return HumanMessage(content=turn.content)
+    return AIMessage(content=turn.content)
+
+
 class AgentState(TypedDict):
     messages: Annotated[list[BaseMessage], add_messages]
     user_id: str
     conversation_id: str
+    parent_conversation_id: str | None
+    thread_id: str | None
+    last_message_id: str | None
+    message_sequence: int
+    agent_id: str | None
+    parent_agent_id: str | None
+    root_agent_id: str | None
+    collaboration_depth: int
+    spawned_agent_ids: list[str]
 
-    execution_mode: Literal["direct", "delegated", "verified"]
+    execution_mode: Literal["direct", "collaboration", "delegated", "verified"]
     current_agent: str | None
     next_step: str | None
     active_agents: list[AgentRole]
@@ -38,11 +54,16 @@ class AgentState(TypedDict):
     sub_commander_trace: list[dict[str, Any]]
     agent_trace: list[str]
     event_trace: list[AgentEvent | dict[str, Any]]
+    message_trace: list[AgentMessage | dict[str, Any]]
 
     pending_tasks: list[dict[str, Any]]
     completed_tasks: list[dict[str, Any]]
     active_tasks: list[AgentTask | dict[str, Any]]
     task_results: list[TaskResult | dict[str, Any]]
+    task_hierarchy: dict[str, list[str]]
+    interrupted_tasks: list[InterruptRecord | dict[str, Any]]
+    recovery_trace: list[RecoveryRecord | dict[str, Any]]
+    recovery_points: list[dict[str, Any]]
     tool_calls: list[dict[str, Any]]
     last_tool_result: str | None
     action_results: list[dict[str, Any]]
@@ -54,7 +75,8 @@ class AgentState(TypedDict):
     verification_status: VerificationStatus | None
     verification_summary: str | None
     verification_evidence: list[dict[str, Any]]
-    budget_state: dict[str, Any] | None
+    budget_state: CollaborationBudget | dict[str, Any] | None
+    collaboration_budget_history: list[CollaborationBudget | dict[str, Any]]
 
     tool_strategy_used: str | None
     tool_round_count: int
@@ -102,6 +124,15 @@ def initial_state(user_id: str, conversation_id: str) -> AgentState:
         messages=[],
         user_id=user_id,
         conversation_id=conversation_id,
+        parent_conversation_id=None,
+        thread_id=None,
+        last_message_id=None,
+        message_sequence=0,
+        agent_id=AgentRole.MASTER.value,
+        parent_agent_id=None,
+        root_agent_id=AgentRole.MASTER.value,
+        collaboration_depth=0,
+        spawned_agent_ids=[],
         execution_mode="direct",
         current_agent=AgentRole.MASTER.value,
         next_step=None,
@@ -111,10 +142,15 @@ def initial_state(user_id: str, conversation_id: str) -> AgentState:
         sub_commander_trace=[],
         agent_trace=[AgentRole.MASTER.value],
         event_trace=[],
+        message_trace=[],
         pending_tasks=[],
         completed_tasks=[],
         active_tasks=[],
         task_results=[],
+        task_hierarchy={},
+        interrupted_tasks=[],
+        recovery_trace=[],
+        recovery_points=[],
         tool_calls=[],
         last_tool_result=None,
         action_results=[],
@@ -126,6 +162,7 @@ def initial_state(user_id: str, conversation_id: str) -> AgentState:
         verification_summary=None,
         verification_evidence=[],
         budget_state=None,
+        collaboration_budget_history=[],
         tool_strategy_used=None,
         tool_round_count=0,
         max_tool_rounds=2,