shuiyu67
b577b0ff21
方案 A（去重）: - mcp_to_openai_tools 末尾按 function.name 去重，消除 cache+builtin 叠加 造成的同名重复（memory/scheduler/lsp__* 等），FC 工具 42→36，去掉 约 14% 的垃圾 schema token - get_liugin_usage_prompts 的 plugins 段由逐条 name:desc 改为工具名清单， 消除与 FC tools 字段的双重曝光，保留无 FC 模型的回退清单 方案 B（分层裁剪）: - 新增 prune_tools(fc_tools, user_input): 常驻 ALWAYS(tool_search/memory/ scheduler) + 外部桥接(mcp__)默认保留 + 按关键词信号追加对应工具 (代码/文件->code_editor+file_manager+lsp__*; git/shell/浏览器/搜索/子agent/ 工程化各按信号) - cli_core 两处 FC 构建点(TUI/CLI)接入 prune_tools(current_input) 实测(AICLI): 纯聊天 FC token ~4310->285(降93%)，写代码 ~973(降77%) 测试: 新增 tests/test_prune_tools.py (7 passed)，全量 324 passed 无回归
450 lines
16 KiB
Python
450 lines
16 KiB
Python
"""
|
||
Function Calling 工具格式转换
|
||
将 MCP 插件定义转为 OpenAI FC 格式，收集所有可用工具
|
||
"""
|
||
import os
|
||
import json
|
||
import threading
|
||
from typing import List, Dict, Optional
|
||
|
||
|
||
def mcp_to_openai_tools(plugin_manager) -> List[Dict]:
|
||
"""
|
||
从插件管理器收集所有工具定义，转为 OpenAI function calling 格式
|
||
|
||
Returns:
|
||
[{"type": "function", "function": {"name": ..., "description": ..., "parameters": ...}}, ...]
|
||
"""
|
||
tools = []
|
||
|
||
# 从插件管理器获取工具
|
||
if not plugin_manager:
|
||
return tools
|
||
|
||
# 优先使用统一工具管理器的 MCP 定义（含外部 MCP client 桥接的工具，schema 保真）
|
||
if hasattr(plugin_manager, 'get_mcp_tools'):
|
||
try:
|
||
for mcp_def in plugin_manager.get_mcp_tools():
|
||
fc_def = mcp_to_fc(mcp_def)
|
||
if fc_def:
|
||
tools.append(fc_def)
|
||
except Exception:
|
||
pass
|
||
# 统一工具管理器（兜底，旧实现未暴露 get_mcp_tools 时）
|
||
elif hasattr(plugin_manager, 'plugins'):
|
||
for plugin in plugin_manager.plugins:
|
||
fc_def = _extract_fc_def(plugin)
|
||
if fc_def:
|
||
tools.append(fc_def)
|
||
# 普通插件管理器
|
||
elif hasattr(plugin_manager, 'tools'):
|
||
for tool_info in plugin_manager.tools:
|
||
fc_def = _tool_info_to_fc(tool_info)
|
||
if fc_def:
|
||
tools.append(fc_def)
|
||
|
||
# 追加内置工具（memory 等）
|
||
builtin_tools = _get_builtin_tools()
|
||
tools.extend(builtin_tools)
|
||
|
||
# ── 去重：同名工具只保留首个 ──
|
||
# 来源：get_mcp_tools() 的缓存里已含 memory/scheduler/lsp 等，
|
||
# 下方 _get_builtin_tools() 又追加了一遍，导致同名重复（约 17% 的 FC token 是垃圾）。
|
||
# 按 function.name 去重，消除双重曝光，对真实工具无影响。
|
||
seen = set()
|
||
deduped = []
|
||
for t in tools:
|
||
fname = (t.get("function") or {}).get("name", "")
|
||
if fname in seen:
|
||
continue
|
||
seen.add(fname)
|
||
deduped.append(t)
|
||
return deduped
|
||
|
||
|
||
# ══════════════════════════════════════════════════════════════════
|
||
# 动态工具裁剪（方案 B：按场景分层）
|
||
#
|
||
# 把 40+ 的 FC 工具全量发给模型，每轮要重发数千 token schema，且
|
||
# 模型要在大列表里选对工具，决策面大、易选错、成本高。
|
||
# 这里按「当前用户输入」的信号，只下发本轮相关的工具子集：
|
||
# - ALWAYS_KEEP：始终保留（含强制 tool_search 规则依赖、记忆、定时）
|
||
# - 外部桥接工具（含 "mcp__"）默认保留（用户显式配置，误杀代价高）
|
||
# - 命中关键词信号才追加对应工具（代码/文件/git/shell/浏览器/搜索/子agent/工程化）
|
||
# 纯聊天时模型只背 ~10 个核心工具而非 40+，FC token 可降约 70%+。
|
||
# ══════════════════════════════════════════════════════════════════
|
||
|
||
# 始终保留的工具（与系统提示里的强制规则/记忆机制强绑定）
|
||
ALWAYS_KEEP = {
|
||
"tool_search", # 系统提示的「强制工具查询规则」依赖它
|
||
"memory", # 持久化记忆
|
||
"scheduler", # 定时任务
|
||
}
|
||
|
||
# 信号：关键词集合 → 追加「以此前缀开头」的工具
|
||
_SIGNAL_RULES = [
|
||
# 代码 / 文件操作：Python 等源码信号命中时，连 pylsp 的 lsp__* 一起给
|
||
(
|
||
{"写", "创建", "新建", "修改", "编辑", "代码", "脚本", "函数", "bug", "修复",
|
||
"调试", "重构", "文件", "报错", "语法", "实现", "module", "def ", "import ",
|
||
"class ", "print(", ".py", ".js", ".ts", ".java", ".go", ".cpp", ".c", ".rs"},
|
||
["code_editor", "code_search", "file_manager", "lsp__", "syntax_check", "check"],
|
||
),
|
||
# 版本控制
|
||
(
|
||
{"git", "提交", "commit", "分支", "branch", "版本", "diff", "stash", "merge", "rebase"},
|
||
["git_tools"],
|
||
),
|
||
# Shell 命令
|
||
(
|
||
{"命令", "shell", "终端", "执行", "运行", "cmd", "bash", "powershell", "terminal"},
|
||
["cmd_executor"],
|
||
),
|
||
# 浏览器
|
||
(
|
||
{"浏览器", "网页", "点击", "browser", "打开网页", "截图", "scroll", "页面"},
|
||
["browser_auto"],
|
||
),
|
||
# 联网 / 搜索 / HTTP
|
||
(
|
||
{"搜索", "联网", "http", "请求", "url", "curl", "api", "查一下", "上网", "fetch"},
|
||
["network_tools", "ai_search"],
|
||
),
|
||
# 子 Agent 并行
|
||
(
|
||
{"子agent", "并行", "subagent", "sub_agent", "多智能体"},
|
||
["sub_agent"],
|
||
),
|
||
# 工程化自动化
|
||
(
|
||
{"测试", "test", "lint", "build", "自动化", "工程化", "ci", "部署", "deploy"},
|
||
["auto_engineer"],
|
||
),
|
||
]
|
||
|
||
|
||
def prune_tools(fc_tools: List[Dict], user_input: str) -> List[Dict]:
|
||
"""
|
||
按当前用户输入动态裁剪 FC 工具子集。
|
||
|
||
返回保留下来的工具列表。无输入或空列表时原样返回（不裁剪）。
|
||
规则见模块顶部 _SIGNAL_RULES / ALWAYS_KEEP 注释。
|
||
"""
|
||
if not fc_tools:
|
||
return fc_tools
|
||
|
||
text = (user_input or "").lower()
|
||
|
||
keep_prefixes = set()
|
||
for keywords, prefixes in _SIGNAL_RULES:
|
||
if any(kw in text for kw in keywords):
|
||
keep_prefixes.update(prefixes)
|
||
|
||
out = []
|
||
for t in fc_tools:
|
||
name = (t.get("function") or {}).get("name", "")
|
||
# 1) 常驻工具
|
||
if name in ALWAYS_KEEP:
|
||
out.append(t)
|
||
continue
|
||
# 2) 外部桥接工具（用户显式配置）默认保留，避免误杀
|
||
if "mcp__" in name:
|
||
out.append(t)
|
||
continue
|
||
# 3) 命中当前轮信号 → 追加（前缀匹配，如 lsp__diagnostics 命中 "lsp__"）
|
||
if any(name.startswith(p) for p in keep_prefixes):
|
||
out.append(t)
|
||
continue
|
||
# 4) 其余（未命中信号的低频工具）本轮裁剪掉
|
||
|
||
return out
|
||
|
||
|
||
def _extract_fc_def(plugin) -> Optional[Dict]:
|
||
"""从插件实例提取 FC 定义"""
|
||
if not hasattr(plugin, 'get_mcp_definition'):
|
||
return None
|
||
|
||
try:
|
||
mcp_def = plugin.get_mcp_definition()
|
||
return mcp_to_fc(mcp_def)
|
||
except Exception:
|
||
return None
|
||
|
||
|
||
def _tool_info_to_fc(tool_info: dict) -> Optional[Dict]:
|
||
"""从 tool_info 字典转为 FC 定义"""
|
||
name = tool_info.get('name', '')
|
||
desc = tool_info.get('description', '')
|
||
if not name:
|
||
return None
|
||
|
||
return {
|
||
"type": "function",
|
||
"function": {
|
||
"name": name,
|
||
"description": desc,
|
||
"parameters": {
|
||
"type": "object",
|
||
"properties": {
|
||
"args": {
|
||
"type": "string",
|
||
"description": f"操作参数。格式参见 {name} 工具说明"
|
||
}
|
||
},
|
||
"required": ["args"]
|
||
}
|
||
}
|
||
}
|
||
|
||
|
||
def mcp_to_fc(mcp_def: dict) -> Optional[Dict]:
|
||
"""
|
||
MCP 定义 → OpenAI FC 定义
|
||
|
||
MCP 格式: {"name": ..., "description": ..., "inputSchema": {...}}
|
||
OpenAI FC: {"type": "function", "function": {"name": ..., "description": ..., "parameters": {...}}}
|
||
"""
|
||
name = mcp_def.get('name', '')
|
||
desc = mcp_def.get('description', '')
|
||
schema = mcp_def.get('inputSchema', mcp_def.get('parameters', {}))
|
||
|
||
if not name:
|
||
return None
|
||
|
||
# 如果 schema 为空，生成默认的 args 参数
|
||
if not schema or not schema.get('properties'):
|
||
schema = {
|
||
"type": "object",
|
||
"properties": {
|
||
"args": {
|
||
"type": "string",
|
||
"description": f"操作参数字符串，格式参见 {name} 工具说明"
|
||
}
|
||
},
|
||
"required": ["args"]
|
||
}
|
||
|
||
return {
|
||
"type": "function",
|
||
"function": {
|
||
"name": name,
|
||
"description": desc,
|
||
"parameters": schema
|
||
}
|
||
}
|
||
|
||
|
||
def fc_result_to_message(tool_call_id: str, tool_name: str, result: str) -> Dict:
|
||
"""
|
||
工具执行结果 → OpenAI 消息格式
|
||
|
||
Returns:
|
||
{"role": "tool", "tool_call_id": ..., "content": ...}
|
||
注意: 官方 OpenAI 的 tool 消息不含 name 字段（严格服务端如讯飞会拒），故不写入。
|
||
"""
|
||
return {
|
||
"role": "tool",
|
||
"tool_call_id": tool_call_id,
|
||
"content": result[:10000] if result else "" # 截断过长结果
|
||
}
|
||
|
||
|
||
def parse_fc_response(response_data: dict) -> Optional[List[Dict]]:
|
||
"""
|
||
解析 OpenAI FC 响应，提取 tool_calls
|
||
|
||
Returns:
|
||
[{"id": "call_xxx", "name": "tool_name", "arguments": {...}}, ...]
|
||
或 None（如果不是 FC 响应）
|
||
"""
|
||
choices = response_data.get("choices", [])
|
||
if not choices:
|
||
return None
|
||
|
||
message = choices[0].get("message", {})
|
||
tool_calls_raw = message.get("tool_calls")
|
||
|
||
if not tool_calls_raw:
|
||
return None
|
||
|
||
tool_calls = []
|
||
for tc in tool_calls_raw:
|
||
func = tc.get("function", {})
|
||
name = func.get("name", "")
|
||
args_str = func.get("arguments", "{}")
|
||
|
||
# 解析 arguments JSON
|
||
try:
|
||
if isinstance(args_str, str):
|
||
args = json.loads(args_str)
|
||
else:
|
||
args = args_str
|
||
except json.JSONDecodeError:
|
||
args = {"raw": args_str}
|
||
|
||
# 统一为小狸内部格式: {"action": "use_tool", "tool": name, "args": "..." }
|
||
# 如果 args 里有 args 字段，直接用；否则把整个 args 序列化
|
||
if "args" in args and isinstance(args["args"], str):
|
||
final_args = args["args"]
|
||
elif "args" in args:
|
||
final_args = json.dumps(args["args"], ensure_ascii=False)
|
||
else:
|
||
# 把所有参数拼成字符串
|
||
parts = []
|
||
for k, v in args.items():
|
||
if isinstance(v, str):
|
||
parts.append(f"{v}")
|
||
else:
|
||
parts.append(json.dumps(v, ensure_ascii=False))
|
||
final_args = " ".join(parts) if parts else ""
|
||
|
||
tool_calls.append({
|
||
"id": tc.get("id", f"call_{name}"),
|
||
"name": name,
|
||
"arguments": args,
|
||
# 小狸内部格式
|
||
"action": "use_tool",
|
||
"tool": name,
|
||
"args": final_args,
|
||
})
|
||
|
||
return tool_calls
|
||
|
||
|
||
def _get_builtin_tools() -> List[Dict]:
|
||
"""内置工具定义（不依赖插件系统）"""
|
||
return [
|
||
{
|
||
"type": "function",
|
||
"function": {
|
||
"name": "memory",
|
||
"description": "持久化记忆系统。写日记、记录长期记忆、搜索历史记忆和聊天记录。",
|
||
"parameters": {
|
||
"type": "object",
|
||
"properties": {
|
||
"action": {
|
||
"type": "string",
|
||
"enum": ["write", "diary", "remember", "search", "context",
|
||
"diary_list", "diary_read", "chat_list", "chat_search"],
|
||
"description": "操作类型"
|
||
},
|
||
"content": {
|
||
"type": "string",
|
||
"description": "内容或关键词"
|
||
}
|
||
},
|
||
"required": ["action"]
|
||
}
|
||
}
|
||
},
|
||
{
|
||
"type": "function",
|
||
"function": {
|
||
"name": "scheduler",
|
||
"description": "定时任务工具，设定定时提醒。支持相对时间(30m/2h)、今日时间(14:30)、绝对时间、重复任务(r:前缀)。",
|
||
"parameters": {
|
||
"type": "object",
|
||
"properties": {
|
||
"action": {
|
||
"type": "string",
|
||
"enum": ["add", "list", "delete", "clear"],
|
||
"description": "操作类型"
|
||
},
|
||
"time": {
|
||
"type": "string",
|
||
"description": "时间: 30m / 14:30 / 2026-04-28 10:00 / r:30m"
|
||
},
|
||
"message": {
|
||
"type": "string",
|
||
"description": "任务描述"
|
||
},
|
||
"task_id": {
|
||
"type": "integer",
|
||
"description": "任务ID (用于 delete)"
|
||
}
|
||
},
|
||
"required": ["action"]
|
||
}
|
||
}
|
||
}
|
||
]
|
||
|
||
|
||
# ══════════════════════════════════════════════════════════════════
|
||
# FC 能力探测缓存
|
||
#
|
||
# 部分渠道/模型（如讯飞星火经 one-api 转发的 xophunyuan* 系列）
|
||
# 不支持 function calling，请求体里只要带上非空 tools 字段就会直接
|
||
# 返回 500 RequestParamsError:Invalid Params。
|
||
# 引擎捕获到这种错误后会去掉 tools 降级重试；本缓存把"该模型不支持 FC"
|
||
# 这件事记下来，后续请求直接跳过 tools，避免每次冷启动都白撞一次 500。
|
||
# ══════════════════════════════════════════════════════════════════
|
||
|
||
_FC_CACHE_FILE = os.path.join(
|
||
os.path.dirname(os.path.dirname(os.path.abspath(__file__))), ".fc_support.json"
|
||
)
|
||
_fc_cache = None
|
||
_fc_lock = threading.Lock()
|
||
|
||
|
||
def _fc_key(base_url, model) -> str:
|
||
"""缓存键：base_url + model（同一渠道下不同模型能力可能不同）"""
|
||
return f"{(base_url or '').rstrip('/')}::{model or ''}"
|
||
|
||
|
||
def _load_fc_cache() -> dict:
|
||
global _fc_cache
|
||
if _fc_cache is not None:
|
||
return _fc_cache
|
||
try:
|
||
with open(_FC_CACHE_FILE, 'r', encoding='utf-8') as f:
|
||
data = json.load(f)
|
||
_fc_cache = data if isinstance(data, dict) else {}
|
||
except (FileNotFoundError, json.JSONDecodeError, OSError, ValueError):
|
||
_fc_cache = {}
|
||
return _fc_cache
|
||
|
||
|
||
def _save_fc_cache(cache: dict):
|
||
try:
|
||
with open(_FC_CACHE_FILE, 'w', encoding='utf-8') as f:
|
||
json.dump(cache, f, ensure_ascii=False, indent=2)
|
||
except OSError:
|
||
pass
|
||
|
||
|
||
def supports_fc(base_url, model) -> bool:
|
||
"""该 (base_url, model) 是否支持 FC。未探测过时默认 True（乐观，先试一次）"""
|
||
return _load_fc_cache().get(_fc_key(base_url, model)) is not False
|
||
|
||
|
||
def mark_fc_unsupported(base_url, model) -> bool:
|
||
"""标记该 (base_url, model) 不支持 FC 并持久化。首次标记返回 True"""
|
||
with _fc_lock:
|
||
cache = _load_fc_cache()
|
||
key = _fc_key(base_url, model)
|
||
if cache.get(key) is False:
|
||
return False
|
||
cache[key] = False
|
||
_save_fc_cache(cache)
|
||
return True
|
||
|
||
|
||
def reset_fc_cache(base_url=None, model=None) -> int:
|
||
"""清除 FC 探测缓存。不传参数则全清，返回清除的条目数"""
|
||
with _fc_lock:
|
||
cache = _load_fc_cache()
|
||
if base_url is None and model is None:
|
||
count = len(cache)
|
||
cache.clear()
|
||
else:
|
||
count = 1 if cache.pop(_fc_key(base_url, model), None) is not None else 0
|
||
_save_fc_cache(cache)
|
||
return count
|
||
|
||
|
||
def get_fc_cache() -> dict:
|
||
"""返回 FC 探测缓存副本（供 /fc 命令展示）"""
|
||
return dict(_load_fc_cache())
|