aria-code 0.167.0__py3-none-win_amd64.whl

aria_code/agents/team.py

@@ -0,0 +1,500 @@
+"""
+agents/team.py — 多 Agent 并行执行与结果汇总
+=============================================
+/team AAPL                          → 运行默认4个内置 agent
+/team AAPL --agents macro,technical → 只运行指定 agent
+/team AAPL --agents macro,my_agent  → 内置 + 自定义混合
+"""
+
+from __future__ import annotations
+
+import asyncio
+import logging
+import time
+from dataclasses import dataclass, field
+from typing import Any, Callable, Dict, List, Optional
+
+from .base import BaseAgent, AgentResult
+from .registry import get_registry
+from .signal_scheme import SignalScheme, FINANCIAL_SCHEME
+
+logger = logging.getLogger(__name__)
+
+# 默认 team 构成（金融场景；其他领域调用方应显式传 agents= 而不是依赖这个）
+DEFAULT_TEAM = ["macro", "fundamental", "technical", "risk"]
+
+
+@dataclass
+class TeamResult:
+    symbol:       str
+    agents_run:   List[str]
+    results:      List[AgentResult]
+    synthesis:    str = ""             # 综合结论（synthesis agent 输出）
+    final_signal: str = "HOLD"        # 多数表决
+    confidence:   float = 0.0
+    elapsed_sec:  float = 0.0
+    error:        Optional[str] = None
+
+
+class AgentTeam:
+    """
+    并行运行多个 Agent，汇总结果。
+
+    用法:
+        team   = AgentTeam(llm_provider=provider, data_router=router)
+        result = await team.run("NVDA", agents=["macro","technical","risk"])
+    """
+
+    def __init__(
+        self,
+        llm_provider=None,
+        data_router=None,
+        on_token: Optional[Callable[[str], None]] = None,
+        on_thought: Optional[Callable[[str], None]] = None,
+        on_tool_start: Optional[Callable[[str, Dict], None]] = None,
+        on_tool_end: Optional[Callable[[str, Any], None]] = None,
+        on_agent_done: Optional[Callable[[str, AgentResult], None]] = None,
+        on_synthesis_start: Optional[Callable[[List["AgentResult"]], None]] = None,
+        timeout_per_agent: float = 60.0,
+        synthesis_timeout: float | None = None,
+        lang: str = "zh",
+        signal_scheme: SignalScheme = FINANCIAL_SCHEME,
+    ):
+        self.llm                = llm_provider
+        self.data               = data_router
+        self.on_token           = on_token
+        self.on_thought         = on_thought
+        self.on_tool_start      = on_tool_start
+        self.on_tool_end        = on_tool_end
+        self.on_agent_done      = on_agent_done
+        self.on_synthesis_start = on_synthesis_start
+        self.timeout            = timeout_per_agent
+        self.synthesis_timeout  = (
+            float(synthesis_timeout)
+            if synthesis_timeout is not None
+            else min(float(timeout_per_agent), 30.0)
+        )
+        self.lang               = lang
+        # 领域可插拔的信号词汇表/表决规则；默认金融词汇，对现有调用方零行为
+        # 变化。realty/sports 等领域应传各自的 SignalScheme（见 signal_scheme.py），
+        # 而不是让自己的 agent 借用金融的 BUY/SELL 语义表达完全不同的含义。
+        self.signal_scheme      = signal_scheme
+
+    def _build_agent(self, name: str) -> Optional[BaseAgent]:
+        registry = get_registry()
+        cls = registry.get(name)
+        if not cls:
+            logger.warning(f"未知 Agent: {name}，跳过")
+            return None
+        return cls(
+            llm_provider=self.llm,
+            data_router=self.data,
+            on_token=self.on_token,
+            on_thought=self.on_thought,
+            on_tool_start=self.on_tool_start,
+            on_tool_end=self.on_tool_end,
+            lang=self.lang,
+        )
+
+    @staticmethod
+    def _available_default_team() -> List[str]:
+        """Return only default agents that are registered in this installation."""
+        registry = get_registry()
+        return [name for name in DEFAULT_TEAM if registry.get(name)]
+
+    def _emit_agent_done(self, name: str, result: AgentResult) -> None:
+        """Notify UI adapters without allowing rendering errors to fail agents."""
+        if not self.on_agent_done:
+            return
+        try:
+            self.on_agent_done(name, result)
+        except Exception as exc:
+            logger.debug("[%s] completion callback failed: %s", name, exc)
+
+    @staticmethod
+    def _fallback_data_is_usable(agent_name: str, data: Dict[str, Any]) -> bool:
+        if agent_name == "technical":
+            return bool(data.get("quote", {}).get("price") and data.get("history"))
+        if agent_name == "fundamental":
+            fundamentals = data.get("fundamentals") or {}
+            return any(
+                fundamentals.get(key) not in (None, "", 0)
+                for key in ("pe_ttm", "pe_ratio", "pb", "pb_ratio", "roe", "revenue_growth")
+            )
+        if agent_name == "risk":
+            return bool(data.get("risk_metrics"))
+        return False
+
+    async def _deterministic_fallback(
+        self,
+        agent: BaseAgent,
+        symbol: str,
+        data: Optional[Dict[str, Any]],
+        reason: str,
+    ) -> Optional[AgentResult]:
+        if not data or not self._fallback_data_is_usable(agent.name, data):
+            return None
+        try:
+            fallback_agent = agent.__class__(
+                llm_provider=None,
+                data_router=None,
+                on_token=None,
+                lang=self.lang,
+            )
+            result = await fallback_agent.analyze(symbol, data)
+            result.degraded = True
+            result.confidence = min(float(result.confidence or 0), 0.45)
+            result.provenance = list(dict.fromkeys([
+                *result.provenance,
+                "prefetched_market_bundle",
+                "deterministic_template",
+            ]))
+            result.limitations = list(dict.fromkeys([
+                *result.limitations,
+                f"LLM agent unavailable: {reason}",
+            ]))
+            result.data_used = dict(result.data_used or {})
+            result.data_used["fallback_reason"] = reason
+            return result
+        except Exception as exc:
+            logger.warning("[%s] deterministic fallback failed: %s", agent.name, exc)
+            return None
+
+    async def _run_one(
+        self,
+        agent: BaseAgent,
+        symbol: str,
+        prefetched_data: Optional[Dict[str, Any]] = None,
+    ) -> AgentResult:
+        try:
+            agent._current_data = prefetched_data or {}
+            operation = (
+                agent.analyze(symbol, prefetched_data)
+                if prefetched_data is not None
+                else agent.run(symbol)
+            )
+            result = await asyncio.wait_for(
+                operation, timeout=self.timeout
+            )
+            self._emit_agent_done(agent.name, result)
+            return result
+        except asyncio.TimeoutError:
+            logger.warning(f"[{agent.name}] 超时 ({self.timeout}s)")
+            fallback = await self._deterministic_fallback(
+                agent, symbol, prefetched_data, "timeout"
+            )
+            if fallback is not None:
+                self._emit_agent_done(agent.name, fallback)
+                return fallback
+            _timeout_result = AgentResult(
+                agent=agent.name, symbol=symbol,
+                analysis="", confidence=0.0, error="timeout",
+            )
+            # Still emit the leaf so the streaming tree shows ⎿ ⏺ <agent> 超时
+            self._emit_agent_done(agent.name, _timeout_result)
+            return _timeout_result
+        except Exception as exc:
+            reason = f"{type(exc).__name__}: {exc}"
+            logger.warning("[%s] failed: %s", agent.name, reason)
+            fallback = await self._deterministic_fallback(
+                agent, symbol, prefetched_data, reason
+            )
+            if fallback is not None:
+                self._emit_agent_done(agent.name, fallback)
+                return fallback
+            failed = AgentResult(
+                agent=agent.name,
+                symbol=symbol,
+                analysis="",
+                confidence=0.0,
+                error=reason,
+            )
+            self._emit_agent_done(agent.name, failed)
+            return failed
+
+    async def run(
+        self,
+        symbol: str,
+        agents: Optional[List[str]] = None,
+        market_context: Optional[Dict[str, Any]] = None,
+        agent_data: Optional[Dict[str, Dict[str, Any]]] = None,
+    ) -> TeamResult:
+        """并行运行所有 agent，等待全部完成后汇总。"""
+        names_to_run = agents
+        t0 = time.time()
+
+        if not names_to_run or names_to_run == ["auto"]:
+            supervisor = self._build_agent("supervisor")
+            if supervisor:
+                try:
+                    sup_result = await asyncio.wait_for(
+                        supervisor.analyze(symbol, market_context or {}),
+                        timeout=self.timeout
+                    )
+                    import json
+                    resolved = json.loads(sup_result.analysis)
+                    names_to_run = (
+                        resolved
+                        if isinstance(resolved, list) and resolved
+                        else self._available_default_team()
+                    )
+                    logger.info(f"[supervisor] dynamically selected agents: {names_to_run}")
+                except Exception as e:
+                    logger.warning(f"[supervisor] failed to resolve dynamic agents: {e}, falling back to default.")
+                    names_to_run = self._available_default_team()
+            else:
+                names_to_run = self._available_default_team()
+
+        # 过滤掉 synthesis 和 debate（各自在并行批次后单独运行）
+        regular = [n for n in names_to_run if n not in ("synthesis", "debate")]
+        agent_objects = [a for n in regular if (a := self._build_agent(n))]
+
+        if not agent_objects:
+            return TeamResult(
+                symbol=symbol, agents_run=[], results=[],
+                error="no_agents_available"
+            )
+
+        # 并行执行 — return_exceptions=True 确保单个 agent 异常不取消其余 agent
+        prefetched = agent_data or {}
+        tasks = [self._run_one(a, symbol, prefetched.get(a.name)) for a in agent_objects]
+        _raw    = await asyncio.gather(*tasks, return_exceptions=True)
+        results: List[AgentResult] = []
+        for _item, _agent in zip(_raw, agent_objects):
+            if isinstance(_item, BaseException):
+                logger.warning("[%s] 意外异常: %s", _agent.name, _item)
+                results.append(AgentResult(
+                    agent=_agent.name, symbol=symbol,
+                    analysis="", confidence=0.0,
+                    error=f"exception: {type(_item).__name__}: {_item}",
+                ))
+            else:
+                results.append(_item)
+
+        # DebateAgent — 显式请求 OR 信号冲突时自动触发
+        explicit_debate = "debate" in names_to_run
+        if explicit_debate or self.signal_scheme.needs_debate(results):
+            debate_agent = self._build_agent("debate")
+            if debate_agent:
+                debate_data = {
+                    "conflicting": [r.to_dict() for r in results if r.success],
+                    # 领域词汇表要一起传：DebateAgent 默认说金融的 BUY/HOLD/SELL，
+                    # 不告诉它当前领域用什么词，它产出的信号会被 vote() 过滤掉
+                    # （不在 scores 里），等于裁判意见完全不参与最终结论。
+                    "signal_scheme": self.signal_scheme,
+                }
+                try:
+                    debate_result = await asyncio.wait_for(
+                        debate_agent.analyze(symbol, debate_data),
+                        timeout=self.timeout,
+                    )
+                    results.append(debate_result)
+                    logger.info("[debate] %s 信号冲突已调解", symbol)
+                except Exception as e:
+                    logger.warning("[debate] 调解失败: %s", e)
+
+        # Fire on_synthesis_start callback so callers can print the agent table
+        # before synthesis begins streaming tokens.
+        if self.on_synthesis_start:
+            try:
+                self.on_synthesis_start(list(results))
+            except Exception:
+                pass
+
+        final_signal, confidence = self.signal_scheme.vote(results)
+
+        # synthesis — 把 agent 结果打包进 data，直接调 analyze() 而非 run()
+        synthesis_text = ""
+        successful_results = [result for result in results if result.success]
+        should_synthesize = bool(successful_results) and (
+            "synthesis" in names_to_run or len(successful_results) >= 2
+        )
+        if should_synthesize:
+            synth_cls = get_registry().get("synthesis")
+            if synth_cls:
+                synth_agent = synth_cls(
+                    llm_provider=self.llm,
+                    data_router=self.data,
+                    on_token=self.on_token,
+                    lang=self.lang,   # without it the agent answered in Chinese for an English /team
+                )
+                synth_data = {
+                    "agent_results": [r.to_dict() for r in results],
+                    "consensus_signal": final_signal,
+                    "consensus_confidence": confidence,
+                }
+                if market_context:
+                    synth_data.update(market_context)
+                try:
+                    synth_result = await asyncio.wait_for(
+                        synth_agent.analyze(symbol, synth_data),
+                        timeout=self.synthesis_timeout,
+                    )
+                    synthesis_text = synth_result.analysis
+                except Exception as e:
+                    logger.warning(f"[synthesis] 失败: {e}")
+                    synthesis_text = _template_synthesis(results, self.signal_scheme, lang=self.lang)
+            else:
+                synthesis_text = _template_synthesis(results, self.signal_scheme, lang=self.lang)
+        else:
+            synthesis_text = _template_synthesis(results, self.signal_scheme, lang=self.lang)
+
+        return TeamResult(
+            symbol       = symbol,
+            agents_run   = [a.name for a in agent_objects],
+            results      = list(results),
+            synthesis    = synthesis_text,
+            final_signal = final_signal,
+            confidence   = confidence,
+            elapsed_sec  = round(time.time() - t0, 1),
+        )
+
+    async def run_sequential(
+        self,
+        symbol: str,
+        agents: List[str],
+        market_context: Optional[Dict[str, Any]] = None,
+        agent_data: Optional[Dict[str, Dict[str, Any]]] = None,
+    ) -> TeamResult:
+        """
+        DAG / Pipeline 顺序工作流。
+        按给定顺序依次执行 Agent，上一个 Agent 的核心结论（key_points）会自动
+        注入到下一个 Agent 的上下文中。适合需要因果推理的场景（如 Macro -> Fundamental）。
+        """
+        t0 = time.time()
+        results: List[AgentResult] = []
+        accumulated_context = dict(market_context or {})
+
+        agent_objects = [a for n in agents if (a := self._build_agent(n))]
+        if not agent_objects:
+            return TeamResult(symbol=symbol, agents_run=[], results=[], error="no_agents_available")
+
+        prefetched = agent_data or {}
+
+        for agent in agent_objects:
+            # 注入前面累积的上下文（如果有 prefetched_data 则混合进去）
+            current_data = dict(prefetched.get(agent.name, {}))
+            current_data["upstream_context"] = accumulated_context.get("upstream_insights", [])
+
+            result = await self._run_one(agent, symbol, current_data)
+            results.append(result)
+
+            # 将当前结论提取，传递给下一个 Agent
+            if result.success and result.key_points:
+                insights = accumulated_context.get("upstream_insights", [])
+                insights.extend([f"[{agent.name.upper()}]: {pt}" for pt in result.key_points])
+                accumulated_context["upstream_insights"] = insights
+
+        # 执行最终的信号表决与综合
+        final_signal, confidence = self.signal_scheme.vote(results)
+        synthesis_text = _template_synthesis(results, self.signal_scheme, lang=self.lang)
+
+        return TeamResult(
+            symbol       = symbol,
+            agents_run   = [a.name for a in agent_objects],
+            results      = list(results),
+            synthesis    = synthesis_text,
+            final_signal = final_signal,
+            confidence   = confidence,
+            elapsed_sec  = round(time.time() - t0, 1),
+        )
+
+
+# ── 独立函数（兼容旧 financial_agents.py 调用方式）──────────────────────────
+
+async def run_team(
+    symbol: str,
+    agents: Optional[List[str]] = None,
+    llm_provider=None,
+    data_router=None,
+    on_token: Optional[Callable] = None,
+    on_thought: Optional[Callable] = None,
+    on_tool_start: Optional[Callable] = None,
+    on_tool_end: Optional[Callable] = None,
+    on_agent_done: Optional[Callable] = None,
+    on_synthesis_start: Optional[Callable] = None,
+    lang: str = "zh",
+    market_context: Optional[Dict[str, Any]] = None,
+    agent_data: Optional[Dict[str, Dict[str, Any]]] = None,
+    timeout_per_agent: float = 60.0,
+    synthesis_timeout: float | None = None,
+    use_pipeline: bool = False,
+) -> TeamResult:
+    """
+    便捷函数，替代原 financial_agents.run_team_analysis()。
+
+    旧签名兼容:
+        result = await run_team_analysis("NVDA", ollama_url, model, on_token)
+    新签名:
+        result = await run_team("NVDA", llm_provider=provider, on_token=cb)
+    """
+    team = AgentTeam(
+        llm_provider=llm_provider,
+        data_router=data_router,
+        on_token=on_token,
+        on_thought=on_thought,
+        on_tool_start=on_tool_start,
+        on_tool_end=on_tool_end,
+        on_agent_done=on_agent_done,
+        on_synthesis_start=on_synthesis_start,
+        timeout_per_agent=timeout_per_agent,
+        synthesis_timeout=synthesis_timeout,
+        lang=lang,
+    )
+    if use_pipeline:
+        return await team.run_sequential(
+            symbol,
+            agents=agents,
+            market_context=market_context,
+            agent_data=agent_data,
+        )
+    else:
+        return await team.run(
+            symbol,
+            agents=agents,
+            market_context=market_context,
+            agent_data=agent_data,
+        )
+
+
+# ── 内部工具 ──────────────────────────────────────────────────────────────────
+
+def _needs_debate(results: List[AgentResult], scheme: SignalScheme = FINANCIAL_SCHEME) -> bool:
+    """当出现真实的正/负信号分歧时返回 True（默认金融 BUY/SELL 词汇，向后兼容）。"""
+    return scheme.needs_debate(results)
+
+
+def _vote_signal(results: List[AgentResult], scheme: SignalScheme = FINANCIAL_SCHEME) -> tuple:
+    """按置信度加权表决最终信号（默认金融词汇，向后兼容 agents/deep/pipeline.py 的直接调用）。"""
+    return scheme.vote(results)
+
+
+def _template_synthesis(results: List[AgentResult], scheme: SignalScheme = FINANCIAL_SCHEME,
+                        lang: str = "zh") -> str:
+    """无 synthesis agent 时的模板汇总 — the summary when no model wrote one, in the team's language."""
+    zh = str(lang).lower().startswith("zh")
+    T = (lambda a, b: a) if zh else (lambda a, b: b)
+    if not results:
+        return T("分析完成，无结果。", "Analysis finished with no results.")
+    lines = [T("## 团队分析汇总\n", "## Team summary\n")]
+    failed_count = sum(1 for r in results if not r.success)
+    if failed_count:
+        lines.append(T(f"> ⚠️ {failed_count}/{len(results)} 个 agent 未能完成分析"
+                       f"（超时或 LLM 不可用），以下结论仅基于成功的 agent。\n",
+                       f"> ⚠️ {failed_count} of {len(results)} analysts did not finish (timeout or no model); "
+                       "the conclusion uses only the ones that did.\n"))
+    for r in results:
+        if r.success:
+            lines.append(f"**{r.agent.upper()}** ({r.signal}, {T('置信度', 'confidence')} {r.confidence:.0%})")
+            for pt in (r.key_points or [])[:3]:
+                lines.append(f"  • {pt}")
+        else:
+            err_label = T("超时", "timed out") if r.error == "timeout" else (r.error or T("分析失败", "failed"))
+            lines.append(f"**{r.agent.upper()}** ⚠️ {err_label}")
+    signal, conf = scheme.vote(results)
+    lines.append(f"\n**{T('综合结论', 'Conclusion')}**: {signal}（{T('置信度', 'confidence')} {conf:.0%}）"
+                 if zh else f"\n**Conclusion**: {signal} (confidence {conf:.0%})")
+    if failed_count == len(results):
+        lines.append(T("\n> ⚠️ 所有 agent 均未成功，此结论仅为默认值，不具参考意义。请确认 LLM 服务正常后重试。",
+                       "\n> ⚠️ No analyst succeeded; this is a default, not a conclusion. Check the model and retry."))
+    return "\n".join(lines)

aria_code/agents/video/__init__.py

@@ -0,0 +1,16 @@
+"""video package — 视频导演、编剧与质检 Agent 网络。"""
+
+from .director import VideoDirectorAgent, DiTMotionPlanner, CharacterConsistencyManager
+from .qa_agent import VisualQAAgent, QACheckResult
+from .script_doctor import ScriptDoctorAgent, PolishedBeat, PolishedScriptResult
+
+__all__ = [
+    "VideoDirectorAgent",
+    "DiTMotionPlanner",
+    "CharacterConsistencyManager",
+    "VisualQAAgent",
+    "QACheckResult",
+    "ScriptDoctorAgent",
+    "PolishedBeat",
+    "PolishedScriptResult",
+]