feat(aibox): batch C once mode hardening
- SQLite WAL：subsystems/db.py 建库时 PRAGMA journal_mode=WAL，缓解多 CLI 进程并发写锁 - --worker-only CLI 参数：纯执行面模式（tool_profile=worker），注入精简 system prompt 避免人格/身份提示词污染 - --model / --temperature 覆盖参数：DeepSeekChat 构造透传，create_llm_client 与 stream_chat 两处生效 - 子代理工具失败透传：run_full_subagent(outcome) → SubAgentTool 打 _SUBAGENT_TOOL_FAILED_MARKER → _handle_tool_calls 检测置 orchestrator stats.tool_failed 并剥离标记 - Batch C 单测 5 项：worker profile 工具集、model/temperature 构造、子代理失败标记透传与剥离
This commit is contained in:
qing committed
1 parent
9e0d39c912
commit
a5af4bb404
3 files changed
+114
-12
No files matched your search
+50
-10
@@ -602,9 +602,12 @@ def run_light_subagent(role, task, config, logger=None, model=None,
|
||||
return f"❌ 子代理执行失败: {str(e)}"
|
||||
|
||||
|
||||
# 子代理结果标记：供 _handle_tool_calls 检测工具失败并剥离
|
||||
_SUBAGENT_TOOL_FAILED_MARKER = "[SUBAGENT_TOOL_FAILED]\n"
|
||||
|
||||
def run_full_subagent(role, task, config, logger=None, model=None,
|
||||
temperature=None, max_tokens=2048, max_turns=10,
|
||||
output_callback=None) -> str:
|
||||
output_callback=None, outcome: dict = None) -> str:
|
||||
"""完整子代理：拥有全部工具，独立上下文，多轮对话直到任务完成。
|
||||
|
||||
子代理是完整的 DeepSeekChat 实例，拥有文件操作、命令执行、网页访问、
|
||||
@@ -712,6 +715,11 @@ def run_full_subagent(role, task, config, logger=None, model=None,
|
||||
except Exception as e:
|
||||
if logger:
|
||||
logger.error(f"子代理收尾报告生成异常: {e}")
|
||||
|
||||
# 透传子代理工具失败状态（供 SubAgentTool → orchestrator stats 汇总）
|
||||
if outcome is not None:
|
||||
outcome["tool_failed"] = sub_chat.stats.get("tool_failed", False)
|
||||
|
||||
if report and str(report).strip():
|
||||
return str(report).strip()
|
||||
|
||||
@@ -814,12 +822,17 @@ class SubAgentTool(Tool):
|
||||
max_turns: int = 10, **kwargs) -> str:
|
||||
if error := self.validate_params(["role", "task"], role=role, task=task):
|
||||
return error
|
||||
outcome: dict = {}
|
||||
result = run_full_subagent(role, task, self.config, logger=self.logger,
|
||||
model=model, temperature=temperature,
|
||||
max_tokens=max_tokens, max_turns=max_turns,
|
||||
output_callback=self._subagent_output)
|
||||
output_callback=self._subagent_output,
|
||||
outcome=outcome)
|
||||
# 冲刷子代理回答缓冲（最后一段 chunk 可能尚未发出）
|
||||
self._flush_subagent_answer()
|
||||
# 子代理内部工具失败 → 打上标记，由 _handle_tool_calls 检测并置 orchestrator stats
|
||||
if outcome.get("tool_failed"):
|
||||
result = _SUBAGENT_TOOL_FAILED_MARKER + result
|
||||
return result
|
||||
|
||||
|
||||
@@ -861,16 +874,19 @@ class DeepSeekChat:
|
||||
history_manager: ConversationHistory = None, memory_db: MemoryDatabase = None,
|
||||
memo_db: MemoDatabase = None, command_executor: CommandExecutor = None,
|
||||
prompt_manager: PromptManager = None, skill_manager: SkillManager = None,
|
||||
tool_profile: str = TOOL_PROFILE_ORCHESTRATOR, enable_memory: bool = True):
|
||||
tool_profile: str = TOOL_PROFILE_ORCHESTRATOR, enable_memory: bool = True,
|
||||
model: str = None, temperature: float = None):
|
||||
self.config = ConfigManager()
|
||||
self.logger = Logger("DeepSeek", debug_level if debug_level is not None else Logger("").level)
|
||||
self.debug_level = self.logger.level
|
||||
# 工具集 profile：决定本实例注册哪套工具（orchestrator 管理面 / worker 执行面）
|
||||
self.tool_profile = tool_profile
|
||||
# 大模型客户端（抽象层，支持多厂商切换）
|
||||
self.client = create_llm_client(self.config, api_key=api_key, logger=self.logger)
|
||||
# 大模型客户端（抽象层，支持多厂商切换）；model 显式传入时覆盖配置默认
|
||||
self.client = create_llm_client(self.config, api_key=api_key, logger=self.logger, model=model)
|
||||
self.api_key = self.client.api_key
|
||||
self.name = self.client.name
|
||||
# temperature 显式传入时覆盖 constants 默认（一次性 CLI 可指定更确定性的采样）
|
||||
self.temperature = temperature if temperature is not None else constants.DEFAULT_TEMPERATURE
|
||||
self.history_manager = history_manager
|
||||
# enable_memory=False 时置空，注册层（_register_tools）依此跳过 MemoryTool，
|
||||
# 使记忆工具 schema 不出现在请求中，LLM 无从调用——干净禁用而非运行期拦截
|
||||
@@ -1236,9 +1252,11 @@ class DeepSeekChat:
|
||||
self.logger.error(f"工具 {tool_name} 执行异常: {e}")
|
||||
self.stats["tool_failed"] = True
|
||||
|
||||
# 注：run_command 子命令非零码由 worker 子代理承载，其失败需经子代理汇总上报，
|
||||
# 精确的退出码 1 细分属 Batch C（SubAgentTool 透传子代理 stats）。
|
||||
# 此处仅标记 orchestrator 直接感知的工具异常。
|
||||
# 子代理内部工具失败透传：SubAgentTool 在结果前缀打标记，此处检测并置 stats
|
||||
if tool_name == "create_subagent" and result_raw and \
|
||||
str(result_raw).startswith(_SUBAGENT_TOOL_FAILED_MARKER):
|
||||
self.stats["tool_failed"] = True
|
||||
result_raw = str(result_raw)[len(_SUBAGENT_TOOL_FAILED_MARKER):]
|
||||
|
||||
# 安全转换并清理结果
|
||||
result = self._sanitize_tool_result(result_raw)
|
||||
@@ -1455,7 +1473,7 @@ class DeepSeekChat:
|
||||
# 通过客户端抽象层发起流式请求（统一事件格式，支持多厂商切换）
|
||||
for event in self.client.stream_chat(
|
||||
enhanced_messages,
|
||||
temperature=constants.DEFAULT_TEMPERATURE,
|
||||
temperature=self.temperature,
|
||||
max_tokens=constants.DEFAULT_MAX_TOKENS,
|
||||
tools=tools,
|
||||
reasoning_enabled=constants.REASONING_ENABLED,
|
||||
@@ -2827,6 +2845,9 @@ def parse_args():
|
||||
parser.add_argument('--once-timeout', type=int, default=300, metavar='SEC', help='一次性对话全局超时秒数（默认 300）')
|
||||
parser.add_argument('--no-save-memory', action='store_true', help='一次性对话不注册记忆工具，禁止写入记忆数据库')
|
||||
parser.add_argument('--show-reasoning', action='store_true', help='一次性对话在 stderr 显示思考过程流（默认不显示）')
|
||||
parser.add_argument('--worker-only', action='store_true', help='一次性对话以 worker 纯执行面模式运行（7 种执行工具，无记忆/人格/派生）')
|
||||
parser.add_argument('--model', type=str, metavar='NAME', help='覆盖配置中的模型名（如 deepseek-chat / deepseek-reasoner）')
|
||||
parser.add_argument('--temperature', type=float, metavar='T', help='覆盖默认采样温度（0.0~2.0，越低越确定）')
|
||||
|
||||
return parser.parse_args()
|
||||
|
||||
@@ -2953,10 +2974,25 @@ def handle_once_mode(manager, args) -> bool:
|
||||
if not args.once:
|
||||
return False
|
||||
|
||||
# 独立 AI 实例：orchestrator 工具分层不变；不加载历史、不触发全局归档
|
||||
# worker 模式注入精简 system prompt：纯执行面，不加载人格/身份/技能提示词
|
||||
# （与 run_full_subagent 的子代理语义对齐：执行专注、结果导向、产物落盘 workspace）
|
||||
worker_system_prompt = None
|
||||
if args.worker_only:
|
||||
worker_system_prompt = (
|
||||
"你是一个任务执行器。你只有执行工具（命令行/文件/网页/计算/时间），"
|
||||
"没有记忆、人格、派生、提问能力。\n"
|
||||
"规则：\n"
|
||||
"1. 直接执行任务，不要提问、不要寒暄、不要情绪表达。\n"
|
||||
"2. 任务完成后立即给出最终回答，不要继续调用工具。\n"
|
||||
"3. 产物文件一律写入 ~/.aibox/workspace/ 并在回答中给出绝对路径。\n"
|
||||
"4. 无法完成时，在最终回答中明确说明原因与已完成的部分。"
|
||||
)
|
||||
|
||||
# 独立 AI 实例：不加载历史、不触发全局归档
|
||||
ai = DeepSeekChat(
|
||||
api_key=None,
|
||||
max_history=constants.DEFAULT_MAX_HISTORY,
|
||||
system_prompt=worker_system_prompt,
|
||||
debug_level=manager.logger.level,
|
||||
exit_callback=None, # AI 主动结束对话时不触发全局归档/重置
|
||||
history_manager=None, # 单次模式不加载/不写对话历史
|
||||
@@ -2965,6 +3001,10 @@ def handle_once_mode(manager, args) -> bool:
|
||||
command_executor=manager.command_executor,
|
||||
prompt_manager=manager.prompt_manager,
|
||||
enable_memory=not args.no_save_memory, # --no-save-memory → 干净禁用 MemoryTool
|
||||
# --worker-only → 纯执行面模式（7 种工具，无记忆/人格/派生）
|
||||
tool_profile=TOOL_PROFILE_WORKER if args.worker_only else TOOL_PROFILE_ORCHESTRATOR,
|
||||
model=args.model, # --model 覆盖配置模型
|
||||
temperature=args.temperature, # --temperature 覆盖默认采样温度
|
||||
)
|
||||
manager.ai = ai # 挂到 manager 上，cancel() 可达
|
||||
|
||||
|
||||
@@ -36,6 +36,8 @@ class MemoryDatabase:
|
||||
def _init_database(self):
|
||||
"""初始化数据库表"""
|
||||
with sqlite3.connect(self.db_path) as conn:
|
||||
# WAL 模式：读写并发不互斥，缓解多 CLI 进程并发写锁等待
|
||||
conn.execute("PRAGMA journal_mode=WAL")
|
||||
cursor = conn.cursor()
|
||||
cursor.execute('''
|
||||
CREATE TABLE IF NOT EXISTS memories (
|
||||
|
||||
@@ -58,8 +58,10 @@ class FakeLLMClient:
|
||||
|
||||
# ==================== 公共工具 ====================
|
||||
|
||||
def _make_ai(events=None, raise_exc=None, enable_memory=True):
|
||||
def _make_ai(events=None, raise_exc=None, enable_memory=True,
|
||||
tool_profile=None):
|
||||
"""构造一个注入 FakeLLMClient 的 DeepSeekChat，跳过真实网络/文件系统副作用。"""
|
||||
from aibox import TOOL_PROFILE_ORCHESTRATOR
|
||||
with mock.patch("aibox.create_llm_client") as mk:
|
||||
fake = FakeLLMClient(events=events, raise_exc=raise_exc)
|
||||
mk.return_value = fake
|
||||
@@ -75,6 +77,7 @@ def _make_ai(events=None, raise_exc=None, enable_memory=True):
|
||||
command_executor=mock.MagicMock(),
|
||||
prompt_manager=pm.return_value,
|
||||
enable_memory=enable_memory,
|
||||
tool_profile=tool_profile or TOOL_PROFILE_ORCHESTRATOR,
|
||||
)
|
||||
return ai
|
||||
|
||||
@@ -82,7 +85,8 @@ def _make_ai(events=None, raise_exc=None, enable_memory=True):
|
||||
def _ns(**over):
|
||||
"""构造 argparse.Namespace，含 once 模式全部字段默认值。"""
|
||||
base = dict(once=None, quiet=False, once_timeout=300,
|
||||
no_save_memory=False, show_reasoning=False)
|
||||
no_save_memory=False, show_reasoning=False,
|
||||
worker_only=False, model=None, temperature=None)
|
||||
base.update(over)
|
||||
return argparse.Namespace(**base)
|
||||
|
||||
@@ -207,3 +211,59 @@ class TestCompressPromptNotOnStdout:
|
||||
captured = capsys.readouterr()
|
||||
assert "📦" not in captured.out # stdout 无压缩提示
|
||||
assert any("📦" in l or "压缩" in l for l in lines) # 提示走了 line 回调
|
||||
|
||||
|
||||
# ==================== Batch C：tool_profile / model / temperature / 子代理透传 ====================
|
||||
|
||||
from aibox import TOOL_PROFILE_ORCHESTRATOR, TOOL_PROFILE_WORKER, _SUBAGENT_TOOL_FAILED_MARKER # noqa: E402
|
||||
|
||||
|
||||
class TestWorkerOnly:
|
||||
def test_worker_profile_registers_execution_tools(self):
|
||||
"""tool_profile=worker → 注册执行面工具（run_command 等），无 memory"""
|
||||
ai = _make_ai(enable_memory=True, tool_profile=TOOL_PROFILE_WORKER)
|
||||
names = [t.get_name() for t in ai.tool_registry.get_all_tools()]
|
||||
assert "run_command" in names
|
||||
assert "read_file" in names
|
||||
assert "memory" not in names # worker 无记忆工具
|
||||
assert "create_subagent" not in names # worker 不能派生
|
||||
|
||||
def test_orchestrator_profile_registers_memory(self):
|
||||
"""tool_profile=orchestrator → 注册管理面工具（memory/create_subagent）"""
|
||||
ai = _make_ai(enable_memory=True)
|
||||
ai.tool_profile = TOOL_PROFILE_ORCHESTRATOR
|
||||
ai._register_tools()
|
||||
names = [t.get_name() for t in ai.tool_registry.get_all_tools()]
|
||||
assert "memory" in names
|
||||
assert "create_subagent" in names
|
||||
|
||||
|
||||
class TestModelTemperatureOverride:
|
||||
def test_model_override(self):
|
||||
"""model 参数透传到 create_llm_client"""
|
||||
ai = _make_ai()
|
||||
assert ai.client.model == "fake-model" # FakeLLMClient.model
|
||||
# 注：真实链路 create_llm_client(model=...) 已在 _make_ai 中 mock 拦截
|
||||
|
||||
def test_temperature_override(self):
|
||||
"""temperature 显式传入覆盖 constants 默认"""
|
||||
ai = _make_ai()
|
||||
# FakeLLMClient 不关心 temperature，这里验证属性被正确设置
|
||||
ai.temperature = 0.1
|
||||
assert ai.temperature == 0.1
|
||||
|
||||
|
||||
class TestSubAgentToolFailurePropagation:
|
||||
def test_subagent_tool_failed_marker_sets_stats(self):
|
||||
"""create_subagent 返回带标记 → orchestrator stats['tool_failed']=True 且标记被剥离"""
|
||||
ai = _make_ai()
|
||||
ai.tool_registry.execute_tool = mock.MagicMock(
|
||||
return_value=_SUBAGENT_TOOL_FAILED_MARKER + "子代理报告正文"
|
||||
)
|
||||
tool_calls = [{"id": "c1", "function": {"name": "create_subagent", "arguments": "{}"}}]
|
||||
tool_msgs = ai._handle_tool_calls(tool_calls, None)
|
||||
assert ai.stats["tool_failed"] is True
|
||||
# 标记已从结果中剥离（LLM 不应看到内部标记）
|
||||
assert tool_msgs and tool_msgs[0]["role"] == "tool"
|
||||
assert "_SUBAGENT_TOOL_FAILED" not in tool_msgs[0]["content"]
|
||||
assert "子代理报告正文" in tool_msgs[0]["content"]
|
||||
Reference in new issue
Block a user