mirror of
https://github.com/HKUDS/nanobot.git
synced 2026-05-21 09:02:32 +00:00
14ee7cb121
The earlier commits picked up a large amount of Black-style reformatting
(multi-line frozenset / keyword-arg wrapping / docstring blanks / removed
parens) on top of the actual guard fix. @chengyongru flagged it; the
first pass reverted some but not all.
This restores nanobot/providers/base.py, runner.py, heartbeat/service.py,
and utils/evaluator.py to origin/main and reapplies only the guard logic:
- base.py: add should_execute_tools property
- runner.py / heartbeat/service.py / utils/evaluator.py: route through it
+ log a warning when has_tool_calls but finish_reason is anomalous
Net diff vs main is now +87/-4 (was +211/-102) — roughly 30 lines of real
logic, which is what the PR is actually about.
Behavior unchanged from previous HEAD; full suite still 2014 passed.
Made-with: Cursor
90 lines
3.0 KiB
Python
90 lines
3.0 KiB
Python
"""Post-run evaluation for background tasks (heartbeat & cron).
|
|
|
|
After the agent executes a background task, this module makes a lightweight
|
|
LLM call to decide whether the result warrants notifying the user.
|
|
"""
|
|
|
|
from __future__ import annotations
|
|
|
|
from typing import TYPE_CHECKING
|
|
|
|
from loguru import logger
|
|
|
|
from nanobot.utils.prompt_templates import render_template
|
|
|
|
if TYPE_CHECKING:
|
|
from nanobot.providers.base import LLMProvider
|
|
|
|
_EVALUATE_TOOL = [
|
|
{
|
|
"type": "function",
|
|
"function": {
|
|
"name": "evaluate_notification",
|
|
"description": "Decide whether the user should be notified about this background task result.",
|
|
"parameters": {
|
|
"type": "object",
|
|
"properties": {
|
|
"should_notify": {
|
|
"type": "boolean",
|
|
"description": "true = result contains actionable/important info the user should see; false = routine or empty, safe to suppress",
|
|
},
|
|
"reason": {
|
|
"type": "string",
|
|
"description": "One-sentence reason for the decision",
|
|
},
|
|
},
|
|
"required": ["should_notify"],
|
|
},
|
|
},
|
|
}
|
|
]
|
|
|
|
async def evaluate_response(
|
|
response: str,
|
|
task_context: str,
|
|
provider: LLMProvider,
|
|
model: str,
|
|
) -> bool:
|
|
"""Decide whether a background-task result should be delivered to the user.
|
|
|
|
Uses a lightweight tool-call LLM request (same pattern as heartbeat
|
|
``_decide()``). Falls back to ``True`` (notify) on any failure so
|
|
that important messages are never silently dropped.
|
|
"""
|
|
try:
|
|
llm_response = await provider.chat_with_retry(
|
|
messages=[
|
|
{"role": "system", "content": render_template("agent/evaluator.md", part="system")},
|
|
{"role": "user", "content": render_template(
|
|
"agent/evaluator.md",
|
|
part="user",
|
|
task_context=task_context,
|
|
response=response,
|
|
)},
|
|
],
|
|
tools=_EVALUATE_TOOL,
|
|
model=model,
|
|
max_tokens=256,
|
|
temperature=0.0,
|
|
)
|
|
|
|
if not llm_response.should_execute_tools:
|
|
if llm_response.has_tool_calls:
|
|
logger.warning(
|
|
"evaluate_response: ignoring tool calls under finish_reason='{}', defaulting to notify",
|
|
llm_response.finish_reason,
|
|
)
|
|
else:
|
|
logger.warning("evaluate_response: no tool call returned, defaulting to notify")
|
|
return True
|
|
|
|
args = llm_response.tool_calls[0].arguments
|
|
should_notify = args.get("should_notify", True)
|
|
reason = args.get("reason", "")
|
|
logger.info("evaluate_response: should_notify={}, reason={}", should_notify, reason)
|
|
return bool(should_notify)
|
|
|
|
except Exception:
|
|
logger.exception("evaluate_response failed, defaulting to notify")
|
|
return True
|