Xingyao Wang ca424ec15d
[agent] Add LLM risk analyzer (#9349)
Signed-off-by: dependabot[bot] <support@github.com>
Co-authored-by: openhands <openhands@all-hands.dev>
Co-authored-by: Graham Neubig <neubig@gmail.com>
Co-authored-by: llamantino <213239228+llamantino@users.noreply.github.com>
Co-authored-by: mamoodi <mamoodiha@gmail.com>
Co-authored-by: Tim O'Farrell <tofarr@gmail.com>
Co-authored-by: Hiep Le <69354317+hieptl@users.noreply.github.com>
Co-authored-by: dependabot[bot] <49699333+dependabot[bot]@users.noreply.github.com>
Co-authored-by: Ryan H. Tran <descience.thh10@gmail.com>
Co-authored-by: Neeraj Panwar <49247372+npneeraj@users.noreply.github.com>
Co-authored-by: sp.wack <83104063+amanape@users.noreply.github.com>
Co-authored-by: Insop <1240382+insop@users.noreply.github.com>
Co-authored-by: test <test@test.com>
Co-authored-by: Engel Nyst <enyst@users.noreply.github.com>
Co-authored-by: Zhonghao Jiang <zhonghao.J@outlook.com>
Co-authored-by: Ray Myers <ray.myers@gmail.com>
2025-08-22 14:02:36 +00:00

152 lines
5.5 KiB
Python

import os
from dataclasses import dataclass, field
from itertools import islice
from jinja2 import Environment, FileSystemLoader, Template
from openhands.controller.state.state import State
from openhands.core.message import Message, TextContent
from openhands.events.observation.agent import MicroagentKnowledge
@dataclass
class RuntimeInfo:
date: str
available_hosts: dict[str, int] = field(default_factory=dict)
additional_agent_instructions: str = ''
custom_secrets_descriptions: dict[str, str] = field(default_factory=dict)
working_dir: str = ''
@dataclass
class RepositoryInfo:
"""Information about a GitHub repository that has been cloned."""
repo_name: str | None = None
repo_directory: str | None = None
branch_name: str | None = None
@dataclass
class ConversationInstructions:
"""Optional instructions the agent must follow throughout the conversation while addressing the user's initial task
Examples include
1. Resolver instructions: you're responding to GitHub issue #1234, make sure to open a PR when you are done
2. Slack instructions: make sure to check whether any of the context attached is relevant to the task <context_messages>
"""
content: str = ''
class PromptManager:
"""Manages prompt templates and includes information from the user's workspace micro-agents and global micro-agents.
This class is dedicated to loading and rendering prompts (system prompt, user prompt).
Attributes:
prompt_dir: Directory containing prompt templates.
"""
def __init__(
self,
prompt_dir: str,
system_prompt_filename: str = 'system_prompt.j2',
):
if prompt_dir is None:
raise ValueError('Prompt directory is not set')
self.prompt_dir: str = prompt_dir
self.env = Environment(loader=FileSystemLoader(prompt_dir))
self.system_template: Template = self._load_template(system_prompt_filename)
self.user_template: Template = self._load_template('user_prompt.j2')
self.additional_info_template: Template = self._load_template(
'additional_info.j2'
)
self.microagent_info_template: Template = self._load_template(
'microagent_info.j2'
)
def _load_template(self, template_name: str) -> Template:
"""Load a template from the prompt directory.
Args:
template_name: Full filename of the template to load, including the .j2 extension.
Returns:
The loaded Jinja2 template.
Raises:
FileNotFoundError: If the template file is not found.
"""
try:
return self.env.get_template(template_name)
except Exception:
template_path = os.path.join(self.prompt_dir, template_name)
raise FileNotFoundError(f'Prompt file {template_path} not found')
def get_system_message(self, **context) -> str:
from openhands.agenthub.codeact_agent.tools.prompt import refine_prompt
system_message = self.system_template.render(**context).strip()
return refine_prompt(system_message)
def get_example_user_message(self) -> str:
"""This is an initial user message that can be provided to the agent
before *actual* user instructions are provided.
It can be used to provide a demonstration of how the agent
should behave in order to solve the user's task. And it may
optionally contain some additional context about the user's task.
These additional context will convert the current generic agent
into a more specialized agent that is tailored to the user's task.
"""
return self.user_template.render().strip()
def build_workspace_context(
self,
repository_info: RepositoryInfo | None,
runtime_info: RuntimeInfo | None,
conversation_instructions: ConversationInstructions | None,
repo_instructions: str = '',
) -> str:
"""Renders the additional info template with the stored repository/runtime info."""
return self.additional_info_template.render(
repository_info=repository_info,
repository_instructions=repo_instructions,
runtime_info=runtime_info,
conversation_instructions=conversation_instructions,
).strip()
def build_microagent_info(
self,
triggered_agents: list[MicroagentKnowledge],
) -> str:
"""Renders the microagent info template with the triggered agents.
Args:
triggered_agents: A list of MicroagentKnowledge objects containing information
about triggered microagents.
"""
return self.microagent_info_template.render(
triggered_agents=triggered_agents
).strip()
def add_turns_left_reminder(self, messages: list[Message], state: State) -> None:
latest_user_message = next(
islice(
(
m
for m in reversed(messages)
if m.role == 'user'
and any(isinstance(c, TextContent) for c in m.content)
),
1,
),
None,
)
if latest_user_message:
reminder_text = f'\n\nENVIRONMENT REMINDER: You have {state.iteration_flag.max_value - state.iteration_flag.current_value} turns left to complete the task. When finished reply with <finish></finish>.'
latest_user_message.content.append(TextContent(text=reminder_text))