nanobot/nanobot/agent/subagent.py

578 lines
22 KiB
Python

"""Subagent manager for background task execution."""
import asyncio
import json
import time
import uuid
import warnings
from dataclasses import dataclass, field
from pathlib import Path
from typing import Any, Callable, TypedDict
from loguru import logger
from nanobot.agent.hook import AgentHook, AgentHookContext
from nanobot.agent.runner import AgentRunner, AgentRunResult, AgentRunSpec
from nanobot.agent.tools.base import ToolResult
from nanobot.agent.tools.context import (
RequestContext,
ToolContext,
bind_request_context,
reset_request_context,
)
from nanobot.agent.tools.exec_session import ExecSessionManager
from nanobot.agent.tools.file_state import FileStates
from nanobot.agent.tools.loader import ToolLoader
from nanobot.agent.tools.registry import ToolRegistry
from nanobot.bus.events import InboundMessage
from nanobot.bus.queue import MessageBus
from nanobot.config.schema import AgentDefaults, ToolsConfig
from nanobot.providers.base import LLMProvider
from nanobot.security.workspace_access import (
WorkspaceScope,
bind_workspace_scope,
reset_workspace_scope,
workspace_sandbox_status,
)
from nanobot.utils.llm_runtime import LLMRuntime
from nanobot.utils.prompt_templates import render_template
class _SubagentOrigin(TypedDict):
channel: str
chat_id: str
session_key: str | None
@dataclass(slots=True)
class SubagentStatus:
"""Real-time status of a running subagent."""
task_id: str
label: str
task_description: str
started_at: float # time.monotonic()
phase: str = "initializing" # initializing | awaiting_tools | tools_completed | final_response | done | error
iteration: int = 0
tool_events: list[dict[str, str]] = field(default_factory=list)
usage: dict[str, int] = field(default_factory=dict)
stop_reason: str | None = None
error: str | None = None
class _SubagentHook(AgentHook):
"""Hook for subagent execution — logs tool calls and updates status."""
def __init__(self, task_id: str, status: SubagentStatus | None = None) -> None:
super().__init__()
self._task_id = task_id
self._status = status
async def before_execute_tools(self, context: AgentHookContext) -> None:
for tool_call in context.tool_calls:
args_str = json.dumps(tool_call.arguments, ensure_ascii=False)
logger.debug(
"Subagent [{}] executing: {} with arguments: {}",
self._task_id, tool_call.name, args_str,
)
async def after_iteration(self, context: AgentHookContext) -> None:
if self._status is None:
return
self._status.iteration = context.iteration
self._status.tool_events = list(context.tool_events)
self._status.usage = dict(context.usage)
if context.error:
self._status.error = str(context.error)
class SubagentManager:
"""Manages background subagent execution."""
def __init__(
self,
provider: LLMProvider | None = None,
workspace: Path | None = None,
bus: MessageBus | None = None,
max_tool_result_chars: int | None = None,
model: str | None = None,
tools_config: ToolsConfig | None = None,
restrict_to_workspace: bool = False,
disabled_skills: list[str] | None = None,
max_iterations: int | None = None,
max_concurrent_subagents: int | None = None,
fail_on_tool_error: bool | None = None,
llm_wall_timeout_for_session: Callable[[str | None], float | None] | None = None,
):
if workspace is None:
raise TypeError("SubagentManager.__init__() missing required argument: 'workspace'")
if bus is None:
raise TypeError("SubagentManager.__init__() missing required argument: 'bus'")
if max_tool_result_chars is None:
raise TypeError(
"SubagentManager.__init__() missing required argument: 'max_tool_result_chars'"
)
if model is not None and provider is None:
raise TypeError("SubagentManager model compatibility argument requires provider")
defaults = AgentDefaults()
self._compat_runtime: LLMRuntime | None = None
if provider is not None:
warnings.warn(
"SubagentManager provider/model constructor arguments are deprecated; "
"pass runtime=... to spawn() instead",
DeprecationWarning,
stacklevel=2,
)
self._compat_runtime = LLMRuntime.capture(
provider,
model or provider.get_default_model(),
context_window_tokens=defaults.context_window_tokens,
)
self.workspace = workspace
self.bus = bus
self.tools_config = tools_config or ToolsConfig()
self.max_tool_result_chars = max_tool_result_chars
self.restrict_to_workspace = restrict_to_workspace
self.disabled_skills = set(disabled_skills or [])
self.max_iterations = (
max_iterations
if max_iterations is not None
else defaults.max_tool_iterations
)
self.max_concurrent_subagents = (
max_concurrent_subagents
if max_concurrent_subagents is not None
else defaults.max_concurrent_subagents
)
self.fail_on_tool_error = (
fail_on_tool_error
if fail_on_tool_error is not None
else defaults.fail_on_tool_error
)
self.runner = AgentRunner()
self._exec_session_manager = ExecSessionManager()
self._llm_wall_timeout_for_session = llm_wall_timeout_for_session
self._running_tasks: dict[str, asyncio.Task[str]] = {}
self._task_statuses: dict[str, SubagentStatus] = {}
self._session_tasks: dict[str, set[str]] = {} # session_key -> {task_id, ...}
def set_provider(self, provider: LLMProvider, model: str) -> None:
"""Update the deprecated runtime source used by legacy ``spawn`` calls."""
warnings.warn(
"SubagentManager.set_provider() is deprecated; pass runtime=... to spawn() instead",
DeprecationWarning,
stacklevel=2,
)
context_window_tokens = (
self._compat_runtime.context_window_tokens
if self._compat_runtime is not None
else AgentDefaults().context_window_tokens
)
self._compat_runtime = LLMRuntime.capture(
provider,
model,
context_window_tokens=context_window_tokens,
)
def _compat_spawn_runtime(self) -> LLMRuntime:
runtime = self._compat_runtime
if runtime is None:
raise TypeError(
"SubagentManager.spawn() missing required keyword-only argument: 'runtime'"
)
warnings.warn(
"SubagentManager.spawn() without runtime is deprecated; pass runtime=... explicitly",
DeprecationWarning,
stacklevel=3,
)
return LLMRuntime.capture(
runtime.provider,
runtime.model,
context_window_tokens=runtime.context_window_tokens,
)
def _subagent_tools_config(self) -> ToolsConfig:
"""Build a ToolsConfig scoped for subagent use."""
return ToolsConfig(
exec=self.tools_config.exec,
web=self.tools_config.web,
file=self.tools_config.file,
restrict_to_workspace=self.restrict_to_workspace,
)
def _build_tools(
self,
workspace: Path | None = None,
tools_config: ToolsConfig | None = None,
) -> ToolRegistry:
"""Build an isolated subagent tool registry via ToolLoader."""
root = self.workspace if workspace is None else workspace
registry = ToolRegistry()
cfg = tools_config if tools_config is not None else self._subagent_tools_config()
ctx = ToolContext(
config=cfg,
workspace=str(root.resolve()),
exec_session_manager=self._exec_session_manager,
file_state_store=FileStates(),
workspace_sandbox=workspace_sandbox_status(
restrict_to_workspace=cfg.restrict_to_workspace,
workspace=root,
),
)
ToolLoader().load(ctx, registry, scope="subagent")
return registry
async def spawn(
self,
task: str,
label: str | None = None,
origin_channel: str = "cli",
origin_chat_id: str = "direct",
session_key: str | None = None,
origin_message_id: str | None = None,
temperature: float | None = None,
workspace_scope: WorkspaceScope | None = None,
*,
runtime: LLMRuntime | None = None,
) -> str:
"""Spawn a subagent to execute a task in the background."""
if runtime is None:
runtime = self._compat_spawn_runtime()
if temperature is not None:
runtime = runtime.with_generation_overrides(temperature=temperature)
task_id = str(uuid.uuid4())[:8]
display_label = label or task[:30] + ("..." if len(task) > 30 else "")
origin: _SubagentOrigin = {
"channel": origin_channel,
"chat_id": origin_chat_id,
"session_key": session_key,
}
status = SubagentStatus(
task_id=task_id,
label=display_label,
task_description=task,
started_at=time.monotonic(),
)
self._task_statuses[task_id] = status
bg_task = asyncio.create_task(
self._run_subagent(
task_id,
task,
display_label,
origin,
status,
runtime,
origin_message_id,
workspace_scope,
)
)
self._running_tasks[task_id] = bg_task
if session_key:
self._session_tasks.setdefault(session_key, set()).add(task_id)
def _cleanup(_: asyncio.Task[str]) -> None:
self._running_tasks.pop(task_id, None)
self._task_statuses.pop(task_id, None)
if session_key and (ids := self._session_tasks.get(session_key)):
ids.discard(task_id)
if not ids:
del self._session_tasks[session_key]
bg_task.add_done_callback(_cleanup)
logger.info("Spawned subagent [{}]: {}", task_id, display_label)
return f"Subagent [{display_label}] started (id: {task_id}). I'll notify you when it completes."
async def run_inline(
self,
task: str,
label: str | None = None,
origin_channel: str = "cli",
origin_chat_id: str = "direct",
session_key: str | None = None,
origin_message_id: str | None = None,
temperature: float | None = None,
workspace_scope: WorkspaceScope | None = None,
*,
runtime: LLMRuntime | None = None,
) -> str:
"""Run a subagent synchronously and return its result to the caller."""
if runtime is None:
runtime = self._compat_spawn_runtime()
if temperature is not None:
runtime = runtime.with_generation_overrides(temperature=temperature)
task_id = str(uuid.uuid4())[:8]
display_label = label or task[:30] + ("..." if len(task) > 30 else "")
origin: _SubagentOrigin = {
"channel": origin_channel,
"chat_id": origin_chat_id,
"session_key": session_key,
}
status = SubagentStatus(
task_id=task_id,
label=display_label,
task_description=task,
started_at=time.monotonic(),
)
self._task_statuses[task_id] = status
logger.info("Running inline subagent [{}]: {}", task_id, display_label)
inline_task = asyncio.create_task(
self._run_subagent(
task_id,
task,
display_label,
origin,
status,
runtime,
origin_message_id,
workspace_scope,
announce=False,
)
)
self._running_tasks[task_id] = inline_task
if session_key:
self._session_tasks.setdefault(session_key, set()).add(task_id)
try:
result = await inline_task
if status.phase == "error" or status.stop_reason in {"error", "tool_error"}:
return ToolResult.error(result)
return result
finally:
self._running_tasks.pop(task_id, None)
self._task_statuses.pop(task_id, None)
if session_key and (ids := self._session_tasks.get(session_key)):
ids.discard(task_id)
if not ids:
del self._session_tasks[session_key]
async def _run_subagent(
self,
task_id: str,
task: str,
label: str,
origin: _SubagentOrigin,
status: SubagentStatus,
runtime: LLMRuntime,
origin_message_id: str | None = None,
workspace_scope: WorkspaceScope | None = None,
*,
announce: bool = True,
) -> str:
"""Execute the subagent task and announce the result."""
logger.info("Subagent [{}] starting task: {}", task_id, label)
async def _on_checkpoint(payload: dict[str, Any]) -> None:
status.phase = payload.get("phase", status.phase)
status.iteration = payload.get("iteration", status.iteration)
try:
root = workspace_scope.project_path if workspace_scope is not None else self.workspace
cfg = None
if workspace_scope is not None:
cfg = self._subagent_tools_config()
cfg.restrict_to_workspace = workspace_scope.restrict_to_workspace
# Construct from the agent workspace; the bound scope below supplies the project cwd.
tools = self._build_tools(tools_config=cfg)
system_prompt = self._build_subagent_prompt(workspace=root)
messages: list[dict[str, Any]] = [
{"role": "system", "content": system_prompt},
{"role": "user", "content": task},
]
sess_key = origin.get("session_key")
llm_timeout = (
self._llm_wall_timeout_for_session(sess_key)
if self._llm_wall_timeout_for_session
else None
)
request_token = bind_request_context(RequestContext(
channel=origin["channel"],
chat_id=origin["chat_id"],
message_id=origin_message_id,
session_key=sess_key,
runtime=runtime,
))
token = bind_workspace_scope(workspace_scope) if workspace_scope is not None else None
try:
result = await self.runner.run(AgentRunSpec(
initial_messages=messages,
tools=tools,
runtime=runtime,
max_iterations=self.max_iterations,
max_tool_result_chars=self.max_tool_result_chars,
hook=_SubagentHook(task_id, status),
max_iterations_message="Task completed but no final response was generated.",
finalize_on_max_iterations=False,
error_message=None,
fail_on_tool_error=self.fail_on_tool_error,
checkpoint_callback=_on_checkpoint,
session_key=sess_key,
workspace=root,
llm_timeout_s=llm_timeout,
))
finally:
if token is not None:
reset_workspace_scope(token)
reset_request_context(request_token)
status.phase = "done"
status.stop_reason = result.stop_reason
if result.stop_reason == "tool_error":
status.tool_events = list(result.tool_events)
final_result = self._format_partial_progress(result)
final_status = "error"
elif result.stop_reason == "error":
final_result = result.error or "Error: subagent execution failed."
final_status = "error"
else:
final_result = result.final_content or "Task completed but no final response was generated."
final_status = "ok"
logger.info("Subagent [{}] completed successfully", task_id)
if announce:
await self._announce_result(
task_id,
label,
task,
final_result,
origin,
final_status,
origin_message_id,
)
return final_result
except Exception as e:
status.phase = "error"
status.error = str(e)
logger.exception("Subagent [{}] failed", task_id)
final_result = f"Error: {e}"
if announce:
await self._announce_result(
task_id,
label,
task,
final_result,
origin,
"error",
origin_message_id,
)
return final_result
async def _announce_result(
self,
task_id: str,
label: str,
task: str,
result: str,
origin: _SubagentOrigin,
status: str,
origin_message_id: str | None = None,
) -> None:
"""Announce the subagent result to the main agent via the message bus."""
status_text = "completed successfully" if status == "ok" else "failed"
announce_content = render_template(
"agent/subagent_announce.md",
label=label,
status_text=status_text,
task=task,
result=result,
)
# Inject as system message to trigger main agent.
# Use session_key_override to align with the main agent's effective
# session key (which accounts for unified sessions) so the result is
# routed to the correct pending queue (mid-turn injection) instead of
# being dispatched as a competing independent task.
override = origin.get("session_key") or f"{origin['channel']}:{origin['chat_id']}"
metadata: dict[str, Any] = {
"injected_event": "subagent_result",
"subagent_task_id": task_id,
}
if origin_message_id:
metadata["origin_message_id"] = origin_message_id
msg = InboundMessage(
channel="system",
sender_id="subagent",
chat_id=f"{origin['channel']}:{origin['chat_id']}",
content=announce_content,
session_key_override=override,
metadata=metadata,
)
await self.bus.publish_inbound(msg)
logger.debug("Subagent [{}] announced result to {}:{}", task_id, origin['channel'], origin['chat_id'])
@staticmethod
def _format_partial_progress(result: AgentRunResult) -> str:
completed = [e for e in result.tool_events if e["status"] == "ok"]
failure = next((e for e in reversed(result.tool_events) if e["status"] == "error"), None)
lines: list[str] = []
if completed:
lines.append("Completed steps:")
for event in completed[-3:]:
lines.append(f"- {event['name']}: {event['detail']}")
if failure:
if lines:
lines.append("")
lines.append("Failure:")
lines.append(f"- {failure['name']}: {failure['detail']}")
if result.error and not failure:
if lines:
lines.append("")
lines.append("Failure:")
lines.append(f"- {result.error}")
return "\n".join(lines) or (result.error or "Error: subagent execution failed.")
def _build_subagent_prompt(self, workspace: Path | None = None) -> str:
"""Build a focused system prompt for the subagent."""
from nanobot.agent.skills import SkillsLoader
agent_workspace = self.workspace.expanduser().resolve()
project_workspace = workspace.expanduser().resolve() if workspace else agent_workspace
skills_summary = SkillsLoader(
self.workspace,
disabled_skills=self.disabled_skills,
).build_skills_summary()
return render_template(
"agent/subagent_system.md",
workspace=str(project_workspace),
agent_workspace=str(agent_workspace),
history_log=str(agent_workspace / "memory" / "history.jsonl"),
skills_summary=skills_summary or "",
)
async def cancel_by_session(self, session_key: str) -> int:
"""Cancel all subagents for the given session. Returns count cancelled."""
tasks = [self._running_tasks[tid] for tid in self._session_tasks.get(session_key, [])
if tid in self._running_tasks and not self._running_tasks[tid].done()]
for t in tasks:
t.cancel()
if tasks:
await asyncio.gather(*tasks, return_exceptions=True)
await self._exec_session_manager.terminate_by_owner(session_key)
return len(tasks)
async def close(self) -> None:
"""Cancel running subagents and close their shared exec sessions."""
tasks = [task for task in self._running_tasks.values() if not task.done()]
for task in tasks:
task.cancel()
if tasks:
await asyncio.gather(*tasks, return_exceptions=True)
await self._exec_session_manager.close_all()
def get_running_count(self) -> int:
"""Return the number of currently running subagents."""
return len(self._running_tasks)
def get_running_count_by_session(self, session_key: str) -> int:
"""Return the number of currently running subagents for a session."""
tids = self._session_tasks.get(session_key, set())
return sum(
1 for tid in tids
if tid in self._running_tasks and not self._running_tasks[tid].done()
)