Skip to content

Agent Zero AI framework

2.1 KB
Raw Blame
import os
from typing import Any

from helpers import responses_tools
from helpers.extension import Extension, extensible
from helpers import files, subagents, tool_policy
from helpers.print_style import PrintStyle
from agent import Agent, LoopData


TOOL_KWARGS_KEY = "_tool_prompt_kwargs"


class ToolsPrompt(Extension):

    async def execute(
        self,
        system_prompt: list[str] = [],
        loop_data: LoopData = LoopData(),
        **kwargs: Any,
    ):
        if not self.agent:
            return
        prompt = await build_prompt(self.agent)
        system_prompt.append(prompt)
        responses_tools.register_prompt(self.agent, loop_data, "tools", prompt)


@extensible
async def build_prompt(agent: Agent) -> str:
    # collect tool files from all prompt directories
    prompt_dirs = subagents.get_paths(agent, "prompts")
    tool_files = files.get_unique_filenames_in_dirs(
        prompt_dirs, "agent.system.tool.*.md"
    )

    # per-file kwargs registered by plugin config extensions (e.g. _09_text_editor_config)
    all_tool_kwargs: dict[str, dict[str, Any]] = agent.get_data(TOOL_KWARGS_KEY) or {}

    tool_prompts: list[tuple[str, str]] = []
    for tool_file in tool_files:
        try:
            basename = os.path.basename(tool_file)
            extra = all_tool_kwargs.get(basename, {})
            tool = agent.read_prompt(basename, **extra)
            tool_prompts.append((basename, tool))
        except Exception as e:
            PrintStyle().error(f"Error loading tool '{tool_file}': {e}")

    tools = [
        tool
        for tool in tool_policy.filter_tool_prompts(agent, tool_prompts)
        if tool
    ]
    tools_str = "\n\n".join(tools)
    prompt = agent.read_prompt("agent.system.tools.md", tools=tools_str)

    # vision support
    from plugins._model_config.helpers.model_config import (
        get_chat_model_config,
        get_vision_model_config,
    )

    chat_cfg = get_chat_model_config(agent)
    if get_vision_model_config(agent) or chat_cfg.get("vision", False):
        prompt += "\n\n" + agent.read_prompt("agent.system.tools_vision.md")

    return prompt