LZH-YS1998
5aa57e69ee
Project 3333 forensics: env_engineer sent a blocking question to the CTO, got a full reply 96s later, and still deadlocked the whole run — the park half (`_park_for_blocking_comms`) had no wired consumer, so WAITING_FOR_PEER work items could never be released. Unpark path (root fix): - The dispatcher loop now calls `_try_unpark_blocking_comms` each tick for parked, non-in-flight tasks; blocking replies land as durable inbox files, so the check is read-only until all replies are present. - `_try_unpark_blocking_comms` accepts orphaned waits (peer_wait stripped by the legacy resolver while the work item stayed parked) and falls back to the park predicate itself: an empty unresolved blocking outbox releases the task, anything pending keeps it parked. - `resolve_task_peer_wait` no longer touches comms_blocking waits (it flipped task.status without the work-item phase and stripped the peer_wait evidence); `_resume_peer_checkpoint` re-enters the company runtime for comms/orphaned waits and lets the dispatcher converge. File tools (defaults changed at their declaration sites, honoring the "empty tools = everything, explicit list = exactly that" contract): - corporate builtin groups gain file_write/file_edit for coordination, QA, and data-acquisition roles. - all shipped org YAML role tool lists gain the missing file_write/file_edit entries. - coordination turn modes no longer strip file_write/file_edit at runtime — in-context content (briefs, matrices) must be persistable instead of getting trapped in blocking DM hand-offs. Also includes the pending office_ui ws_handler change from the working tree. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
523 lines
20 KiB
Python
523 lines
20 KiB
Python
from __future__ import annotations
|
|
|
|
import unittest
|
|
from types import SimpleNamespace
|
|
from unittest.mock import AsyncMock
|
|
|
|
from opc.core.config import OPCConfig
|
|
from opc.core.models import AgentInfo, AgentStatus, ExecutionMode, Task, TaskResult, TaskStatus
|
|
from opc.layer2_organization.org_engine import TASK_MODE_GENERAL_ROLE_ID
|
|
from opc.layer3_agent.native_agent import NativeAgent
|
|
from opc.layer4_tools.registry import ToolDefinition, ToolRegistry
|
|
|
|
|
|
class _StubContextAssembler:
|
|
async def build_system_context(self, task: Task, role_id: str) -> str:
|
|
_ = (task, role_id)
|
|
return ""
|
|
|
|
def build_task_brief(self, task: Task) -> str:
|
|
return task.description or task.title
|
|
|
|
|
|
class _StubPreferences:
|
|
def build_preference_context(self, project_id: str) -> str:
|
|
_ = project_id
|
|
return ""
|
|
|
|
def summarize_autonomy_preferences(self, project_id: str) -> str:
|
|
_ = project_id
|
|
return ""
|
|
|
|
|
|
class _StubSkills:
|
|
def build_skills_summary(
|
|
self,
|
|
project_id: str | None = None,
|
|
*,
|
|
execution_mode: str | None = None,
|
|
role_id: str | None = None,
|
|
user_facing: bool = False,
|
|
final_decider_role_id: str | None = None,
|
|
) -> str:
|
|
_ = (project_id, execution_mode, role_id, user_facing, final_decider_role_id)
|
|
return ""
|
|
|
|
|
|
class _StubMemory:
|
|
history_compactor = None
|
|
|
|
async def build_session_history_tail_messages(
|
|
self,
|
|
session_id: str,
|
|
*,
|
|
include_latest_user_turn: bool = True,
|
|
) -> list[dict[str, str]]:
|
|
_ = (session_id, include_latest_user_turn)
|
|
return []
|
|
|
|
|
|
class _TrackingMemory(_StubMemory):
|
|
def __init__(self) -> None:
|
|
self.history_calls = 0
|
|
|
|
async def build_session_history_tail_messages(
|
|
self,
|
|
session_id: str,
|
|
*,
|
|
include_latest_user_turn: bool = True,
|
|
) -> list[dict[str, str]]:
|
|
self.history_calls += 1
|
|
return [{"role": "assistant", "content": f"history:{session_id}:{include_latest_user_turn}"}]
|
|
|
|
|
|
class _StubConfig:
|
|
max_tokens = 1024
|
|
|
|
|
|
class _StubLLM:
|
|
def __init__(self) -> None:
|
|
self.config = _StubConfig()
|
|
self.seen_tool_names: list[list[str]] = []
|
|
|
|
def get_tool_definitions(self, tools):
|
|
return tools
|
|
|
|
def get_context_window(self):
|
|
return 8192
|
|
|
|
def count_input_tokens(self, messages, tools=None):
|
|
_ = (messages, tools)
|
|
return 100
|
|
|
|
def is_context_overflow_error(self, error):
|
|
_ = error
|
|
return False
|
|
|
|
def prepare_user_message_content(self, user_message: str, attachment_refs=None):
|
|
_ = attachment_refs
|
|
return user_message
|
|
|
|
async def simple_chat(self, prompt: str, system: str | None = None, task_type: str | None = None) -> str:
|
|
_ = (prompt, system, task_type)
|
|
return "summary"
|
|
|
|
async def chat(self, messages, tools=None):
|
|
_ = messages
|
|
self.seen_tool_names.append([tool.get("name", "") for tool in tools or []])
|
|
return {"content": "done", "tool_calls": [], "usage": {}}
|
|
|
|
|
|
def _make_role(
|
|
tools: list[str] | None = None,
|
|
*,
|
|
role_id: str = TASK_MODE_GENERAL_ROLE_ID,
|
|
can_spawn: list[str] | None = None,
|
|
role_type: str = "worker",
|
|
prompt_refs: list[str] | None = None,
|
|
) -> AgentInfo:
|
|
return AgentInfo(
|
|
role_id=role_id,
|
|
name="Task Generalist",
|
|
responsibility="Primary session agent for task mode.",
|
|
status=AgentStatus.IDLE,
|
|
can_spawn=list(can_spawn or []),
|
|
tools=list(tools or []),
|
|
prompt_refs=list(prompt_refs or []),
|
|
runtime_policy={"role_type": role_type},
|
|
)
|
|
|
|
|
|
def _make_native_agent(role: AgentInfo) -> NativeAgent:
|
|
return _make_native_agent_with_memory(role, _StubMemory())
|
|
|
|
|
|
def _make_native_agent_with_memory(
|
|
role: AgentInfo,
|
|
memory: _StubMemory,
|
|
tool_registry: ToolRegistry | None = None,
|
|
) -> NativeAgent:
|
|
return NativeAgent(
|
|
role=role,
|
|
llm=_StubLLM(),
|
|
tool_registry=tool_registry or ToolRegistry(),
|
|
context_assembler=_StubContextAssembler(),
|
|
memory=memory,
|
|
preferences=_StubPreferences(),
|
|
skills=_StubSkills(),
|
|
event_bus=SimpleNamespace(publish=AsyncMock()),
|
|
config=OPCConfig(),
|
|
)
|
|
|
|
|
|
async def _noop_tool(**kwargs):
|
|
_ = kwargs
|
|
return {"ok": True}
|
|
|
|
|
|
def _registry_with_tools(names: list[str]) -> ToolRegistry:
|
|
registry = ToolRegistry()
|
|
for name in names:
|
|
registry.register(ToolDefinition(name=name, description=name, parameters={}, func=_noop_tool))
|
|
return registry
|
|
|
|
|
|
class TaskModeNativeAgentTests(unittest.IsolatedAsyncioTestCase):
|
|
async def test_context_messages_skip_session_history_with_explicit_flag(self) -> None:
|
|
memory = _TrackingMemory()
|
|
agent = _make_native_agent_with_memory(_make_role(), memory)
|
|
task = Task(
|
|
title="Internal assessment",
|
|
description="Return JSON only.",
|
|
session_id="sess-internal",
|
|
project_id="proj1",
|
|
metadata={"mode": "task", "execution_mode": ExecutionMode.TASK_MODE.value},
|
|
context_snapshot={"skip_session_history": True},
|
|
)
|
|
|
|
context_messages = await agent._build_context_messages(task)
|
|
|
|
self.assertEqual(memory.history_calls, 0)
|
|
self.assertFalse(any("history:sess-internal" in str(message.get("content", "")) for message in context_messages))
|
|
|
|
async def test_legacy_boolean_runtime_resume_skips_history_without_crashing(self) -> None:
|
|
memory = _TrackingMemory()
|
|
agent = _make_native_agent_with_memory(_make_role(), memory)
|
|
task = Task(
|
|
title="Legacy internal assessment",
|
|
description="Return JSON only.",
|
|
session_id="sess-legacy",
|
|
project_id="proj1",
|
|
metadata={"mode": "task", "execution_mode": ExecutionMode.TASK_MODE.value},
|
|
context_snapshot={"runtime_resume": True},
|
|
)
|
|
|
|
context_messages = await agent._build_context_messages(task)
|
|
|
|
self.assertEqual(memory.history_calls, 0)
|
|
self.assertFalse(any("history:sess-legacy" in str(message.get("content", "")) for message in context_messages))
|
|
|
|
async def test_task_mode_prompt_uses_unified_capability_contract(self) -> None:
|
|
agent = _make_native_agent(_make_role(prompt_refs=["Legacy task-mode persona text."]))
|
|
task = Task(
|
|
title="Inspect OpenReview assignments",
|
|
description="Open the site and inspect my review assignments.",
|
|
project_id="proj1",
|
|
metadata={"mode": "task", "execution_mode": ExecutionMode.TASK_MODE.value},
|
|
)
|
|
|
|
prompt = await agent._build_system_prompt(task)
|
|
context_messages = await agent._build_context_messages(task)
|
|
context_prompt = "\n\n".join(str(item.get("content", "")) for item in context_messages)
|
|
|
|
self.assertEqual(task.metadata["runtime_prompt_profile"], "unified")
|
|
self.assertIn("You are Task Generalist, an OpenOPC task execution agent.", prompt)
|
|
self.assertNotIn("One-Person Company system", prompt)
|
|
self.assertIn("## Core Operating Principles", prompt)
|
|
self.assertNotIn("## Core Beliefs", prompt)
|
|
self.assertIn("## Native Working Contract", prompt)
|
|
self.assertIn("not as prompt profiles selected by metadata", prompt)
|
|
self.assertIn("## Native Self-Verification Contract", prompt)
|
|
self.assertNotIn("## Task-Mode Orchestration", prompt)
|
|
self.assertIn("## Task-Mode Orchestration", context_prompt)
|
|
self.assertIn("primary task-mode execution agent", context_prompt)
|
|
self.assertIn("recruiting flow, employee persona, or staff assignment", context_prompt)
|
|
self.assertIn("agent_spawn", context_prompt)
|
|
self.assertIn("## Runtime Artifact", context_prompt)
|
|
self.assertIn("Tool Strategy", context_prompt)
|
|
self.assertIn("## Task-Mode Orchestration", str(context_messages[0].get("content", "")))
|
|
self.assertIn("Tool Strategy", str(context_messages[-1].get("content", "")))
|
|
self.assertIn("Include a short verification status", prompt)
|
|
self.assertNotIn("## Role Operating Instructions", prompt)
|
|
self.assertNotIn("Legacy task-mode persona text.", context_prompt)
|
|
self.assertNotIn("## Planning Principles", prompt)
|
|
self.assertNotIn("## Plan Profile", prompt)
|
|
|
|
async def test_plan_subagent_uses_unified_prompt_without_task_mode_orchestration(self) -> None:
|
|
agent = _make_native_agent(_make_role(["file_read", "todo_write"]))
|
|
task = Task(
|
|
title="Draft the migration plan",
|
|
description="Inspect the runtime modules and propose a migration sequence.",
|
|
project_id="proj1",
|
|
metadata={"_subagent_mode": "plan"},
|
|
)
|
|
|
|
prompt = await agent._build_system_prompt(task)
|
|
context_messages = await agent._build_context_messages(task)
|
|
context_prompt = "\n\n".join(str(item.get("content", "")) for item in context_messages)
|
|
|
|
self.assertEqual(task.metadata["runtime_prompt_profile"], "unified")
|
|
self.assertIn("## Native Working Contract", prompt)
|
|
self.assertIn("For planning, produce decision-complete steps", prompt)
|
|
self.assertNotIn("## Task Tracking", prompt)
|
|
self.assertIn("Tool Strategy", context_prompt)
|
|
self.assertIn("Use the task ledger", context_prompt)
|
|
self.assertNotIn("## Plan Profile", prompt)
|
|
self.assertNotIn("## Planning Principles", prompt)
|
|
self.assertNotIn("## Task-Mode Orchestration", prompt)
|
|
|
|
async def test_company_review_work_item_uses_company_contract_and_verify_profile(self) -> None:
|
|
agent = _make_native_agent(_make_role(["file_read", "file_search", "list_dir"]))
|
|
task = Task(
|
|
title="QA Review",
|
|
description="Review the execution output with fresh eyes.",
|
|
project_id="proj1",
|
|
metadata={
|
|
"execution_mode": ExecutionMode.COMPANY_MODE.value,
|
|
"work_item_projection_title": "QA Review",
|
|
"work_item_turn_type": "review",
|
|
"work_item_orchestration_profile": "company_review_fresh_eyes",
|
|
},
|
|
)
|
|
|
|
prompt = await agent._build_system_prompt(task)
|
|
context_messages = await agent._build_context_messages(task)
|
|
context_prompt = "\n\n".join(str(item.get("content", "")) for item in context_messages)
|
|
|
|
self.assertEqual(task.metadata["runtime_prompt_profile"], "unified")
|
|
self.assertIn("## Native Working Contract", prompt)
|
|
self.assertNotIn("## Company Work Item Contract", prompt)
|
|
self.assertIn("## Company Work Item Contract", context_prompt)
|
|
self.assertIn("Work item turn type: `review`", context_prompt)
|
|
self.assertIn("## Company Work Item Turn: Review", context_prompt)
|
|
# Fix 4 reshaped the inline example from the compact
|
|
# ``"review_verdict":"approve|reject"`` alternation into two
|
|
# explicit Approve / Reject examples. Assert both shapes show up.
|
|
self.assertIn('"review_verdict":"approve"', context_prompt)
|
|
self.assertIn('"review_verdict":"reject"', context_prompt)
|
|
self.assertIn("blocking_issues", context_prompt)
|
|
self.assertNotIn("## Task-Mode Orchestration", prompt)
|
|
|
|
async def test_task_mode_strips_inherited_company_collaboration_tools(self) -> None:
|
|
agent = _make_native_agent(_make_role())
|
|
task = Task(
|
|
title="Task-mode child",
|
|
description="Continue the task without company collaboration.",
|
|
project_id="proj1",
|
|
metadata={
|
|
"mode": "task",
|
|
"execution_mode": ExecutionMode.TASK_MODE.value,
|
|
"_fork_allowed_tools": ["file_read", "send_dm"],
|
|
},
|
|
)
|
|
|
|
allowed = agent._resolve_allowed_tools(task)
|
|
context_messages = await agent._build_context_messages(task)
|
|
context_prompt = "\n\n".join(str(item.get("content", "")) for item in context_messages)
|
|
|
|
self.assertEqual(allowed, ["file_read"])
|
|
self.assertNotIn("company collaboration tools", context_prompt)
|
|
|
|
async def test_company_execute_work_item_does_not_expose_manager_tools(self) -> None:
|
|
agent = _make_native_agent(_make_role(["file_write", "file_edit"]))
|
|
task = Task(
|
|
title="Engineering Execution",
|
|
description="Implement the approved scope.",
|
|
project_id="proj1",
|
|
metadata={
|
|
"execution_mode": ExecutionMode.COMPANY_MODE.value,
|
|
"work_item_projection_id": "engineering_execution",
|
|
"work_item_turn_type": "execute",
|
|
},
|
|
)
|
|
|
|
allowed = agent._resolve_allowed_tools(task)
|
|
|
|
self.assertNotIn("delegate_work", allowed)
|
|
self.assertNotIn("manager_board_read", allowed)
|
|
self.assertNotIn("propose_runtime_replan", allowed)
|
|
|
|
async def test_company_empty_role_tools_allow_all_registered_general_tools(self) -> None:
|
|
registry = _registry_with_tools([
|
|
"file_read",
|
|
"shell_exec",
|
|
"request_user_input",
|
|
"delegate_work",
|
|
"manager_board_read",
|
|
"send_dm",
|
|
"inbox",
|
|
])
|
|
agent = _make_native_agent_with_memory(
|
|
_make_role([], role_id="analyst"),
|
|
_StubMemory(),
|
|
registry,
|
|
)
|
|
task = Task(
|
|
title="Analysis",
|
|
description="Analyze with native tools.",
|
|
project_id="proj1",
|
|
metadata={
|
|
"execution_mode": ExecutionMode.COMPANY_MODE.value,
|
|
"work_item_projection_id": "analysis_execution",
|
|
"work_item_turn_type": "execute",
|
|
},
|
|
)
|
|
|
|
allowed = agent._resolve_allowed_tools(task)
|
|
|
|
self.assertIn("file_read", allowed)
|
|
self.assertIn("shell_exec", allowed)
|
|
self.assertIn("request_user_input", allowed)
|
|
self.assertIn("send_dm", allowed)
|
|
self.assertIn("inbox", allowed)
|
|
self.assertNotIn("delegate_work", allowed)
|
|
self.assertNotIn("manager_board_read", allowed)
|
|
|
|
async def test_company_explicit_role_tools_restrict_general_tools_only(self) -> None:
|
|
registry = _registry_with_tools([
|
|
"file_read",
|
|
"shell_exec",
|
|
"request_user_input",
|
|
"send_dm",
|
|
"inbox",
|
|
])
|
|
agent = _make_native_agent_with_memory(
|
|
_make_role(["file_read"], role_id="analyst"),
|
|
_StubMemory(),
|
|
registry,
|
|
)
|
|
task = Task(
|
|
title="Analysis",
|
|
description="Analyze with native tools.",
|
|
project_id="proj1",
|
|
metadata={
|
|
"execution_mode": ExecutionMode.COMPANY_MODE.value,
|
|
"work_item_projection_id": "analysis_execution",
|
|
"work_item_turn_type": "execute",
|
|
},
|
|
)
|
|
|
|
allowed = agent._resolve_allowed_tools(task)
|
|
|
|
self.assertIn("file_read", allowed)
|
|
self.assertIn("send_dm", allowed)
|
|
self.assertIn("inbox", allowed)
|
|
self.assertNotIn("shell_exec", allowed)
|
|
self.assertNotIn("request_user_input", allowed)
|
|
|
|
async def test_company_manager_work_item_exposes_delegate_and_manager_board_but_not_replan(self) -> None:
|
|
agent = _make_native_agent(_make_role(["file_read"], role_id="cto"))
|
|
task = Task(
|
|
title="CTO Delegation",
|
|
description="Break the work into subordinate tasks.",
|
|
project_id="proj1",
|
|
metadata={
|
|
"execution_mode": ExecutionMode.COMPANY_MODE.value,
|
|
"work_item_projection_id": "cto_delegation",
|
|
"work_item_turn_type": "plan",
|
|
"managed_team_id": "team::cto",
|
|
"delegation_seat_id": "seat::team::ceo::cto",
|
|
},
|
|
)
|
|
|
|
allowed = agent._resolve_allowed_tools(task)
|
|
context_messages = await agent._build_context_messages(task)
|
|
context_prompt = "\n\n".join(str(item.get("content", "")) for item in context_messages)
|
|
|
|
self.assertIn("delegate_work", allowed)
|
|
self.assertIn("manager_board_read", allowed)
|
|
self.assertNotIn("manager_board_update", allowed)
|
|
self.assertNotIn("manager_board_release", allowed)
|
|
self.assertNotIn("manager_board_rollup", allowed)
|
|
self.assertNotIn("propose_runtime_replan", allowed)
|
|
self.assertNotIn("read_inbox", allowed)
|
|
self.assertIn("company collaboration tools", context_prompt)
|
|
|
|
async def test_multi_team_manager_turn_strips_execution_heavy_tools(self) -> None:
|
|
agent = _make_native_agent(
|
|
_make_role(["file_read", "shell_exec", "file_write", "web_search"], role_id="ceo", can_spawn=["cto"])
|
|
)
|
|
task = Task(
|
|
title="CEO Intake",
|
|
description="Coordinate the top-level delegation.",
|
|
project_id="proj1",
|
|
metadata={
|
|
"execution_mode": ExecutionMode.COMPANY_MODE.value,
|
|
"runtime_model": "multi_team_org",
|
|
"work_item_projection_id": "ceo_intake",
|
|
"work_item_turn_type": "intake",
|
|
"current_turn_mode": "dispatch_required",
|
|
"delegation_seat_id": "seat::team::ceo::ceo",
|
|
"managed_team_id": "team::ceo",
|
|
"runtime_topology": {
|
|
"seats": [
|
|
{
|
|
"seat_id": "seat::team::ceo::ceo",
|
|
"team_id": "team::ceo",
|
|
"role_id": "ceo",
|
|
"managed_team_id": "team::ceo",
|
|
},
|
|
{
|
|
"seat_id": "seat::team::ceo::cto",
|
|
"team_id": "team::ceo",
|
|
"role_id": "cto",
|
|
"manager_role_id": "ceo",
|
|
},
|
|
]
|
|
},
|
|
},
|
|
)
|
|
|
|
allowed = agent._resolve_allowed_tools(task)
|
|
|
|
self.assertIn("file_read", allowed)
|
|
self.assertIn("delegate_work", allowed)
|
|
self.assertIn("manager_board_read", allowed)
|
|
# file_write stays available: coordination turns must be able to
|
|
# persist in-context content instead of trapping it in DM hand-offs.
|
|
self.assertIn("file_write", allowed)
|
|
self.assertNotIn("shell_exec", allowed)
|
|
self.assertNotIn("web_search", allowed)
|
|
self.assertNotIn("agent_spawn", allowed)
|
|
self.assertNotIn("agent_wait", allowed)
|
|
self.assertNotIn("agent_send", allowed)
|
|
|
|
async def test_company_coordinator_work_item_hides_legacy_replan_and_routing_tools_in_multi_team(self) -> None:
|
|
agent = _make_native_agent(
|
|
_make_role(["file_read"], role_id="ceo", can_spawn=["cto", "cmo"], role_type="coordinator")
|
|
)
|
|
task = Task(
|
|
title="CEO Intake",
|
|
description="Coordinate the top-level delegation.",
|
|
project_id="proj1",
|
|
metadata={
|
|
"execution_mode": ExecutionMode.COMPANY_MODE.value,
|
|
"runtime_model": "multi_team_org",
|
|
"work_item_projection_id": "ceo_intake",
|
|
"work_item_turn_type": "intake",
|
|
"current_turn_mode": "dispatch_required",
|
|
"delegation_seat_id": "seat::team::ceo::ceo",
|
|
"managed_team_id": "team::ceo",
|
|
"runtime_topology": {
|
|
"seats": [
|
|
{
|
|
"seat_id": "seat::team::ceo::ceo",
|
|
"team_id": "team::ceo",
|
|
"role_id": "ceo",
|
|
"managed_team_id": "team::ceo",
|
|
},
|
|
{
|
|
"seat_id": "seat::team::ceo::cto",
|
|
"team_id": "team::ceo",
|
|
"role_id": "cto",
|
|
"manager_role_id": "ceo",
|
|
},
|
|
]
|
|
},
|
|
},
|
|
)
|
|
|
|
allowed = agent._resolve_allowed_tools(task)
|
|
|
|
self.assertIn("delegate_work", allowed)
|
|
self.assertNotIn("route_work", allowed)
|
|
self.assertNotIn("propose_task_adjustment", allowed)
|
|
self.assertNotIn("propose_runtime_replan", allowed)
|
|
self.assertNotIn("find_and_ask_expert", allowed)
|
|
self.assertNotIn("ask_peer_and_wait", allowed)
|
|
self.assertNotIn("read_inbox", allowed)
|
|
|
|
|
|
if __name__ == "__main__":
|
|
unittest.main()
|