| 12345678910111213141516171819202122232425262728293031323334353637383940414243444546474849505152535455565758596061626364656667686970717273747576777879808182838485868788899091929394959697989910010110210310410510610710810911011111211311411511611711811912012112212312412512612712812913013113213313413513613713813914014114214314414514614714814915015115215315415515615715815916016116216316416516616716816917017117217317417517617717817918018118218318418518618718818919019119219319419519619719819920020120220320420520620720820921021121221321421521621721821922022122222322422522622722822923023123223323423523623723823924024124224324424524624724824925025125225325425525625725825926026126226326426526626726826927027127227327427527627727827928028128228328428528628728828929029129229329429529629729829930030130230330430530630730830931031131231331431531631731831932032132232332432532632732832933033133233333433533633733833934034134234334434534634734834935035135235335435535635735835936036136236336436536636736836937037137237337437537637737837938038138238338438538638738838939039139239339439539639739839940040140240340440540640740840941041141241341441541641741841942042142242342442542642742842943043143243343443543643743843944044144244344444544644744844945045145245345445545645745845946046146246346446546646746846947047147247347447547647747847948048148248348448548648748848949049149249349449549649749849950050150250350450550650750850951051151251351451551651751851952052152252352452552652752852953053153253353453553653753853954054154254354454554654754854955055155255355455555655755855956056156256356456556656756856957057157257357457557657757857958058158258358458558658758858959059159259359459559659759859960060160260360460560660760860961061161261361461561661761861962062162262362462562662762862963063163263363463563663763863964064164264364464564664764864965065165265365465565665765865966066166266366466566666766866967067167267367467567667767867968068168268368468568668768868969069169269369469569669769869970070170270370470570670770870971071171271371471571671771871972072172272372472572672772872973073173273373473573673773873974074174274374474574674774874975075175275375475575675775875976076176276376476576676776876977077177277377477577677777877978078178278378478578678778878979079179279379479579679779879980080180280380480580680780880981081181281381481581681781881982082182282382482582682782882983083183283383483583683783883984084184284384484584684784884985085185285385485585685785885986086186286386486586686786886987087187287387487587687787887988088188288388488588688788888989089189289389489589689789889990090190290390490590690790890991091191291391491591691791891992092192292392492592692792892993093193293393493593693793893994094194294394494594694794894995095195295395495595695795895996096196296396496596696796896997097197297397497597697797897998098198298398498598698798898999099199299399499599699799899910001001100210031004100510061007100810091010101110121013101410151016101710181019102010211022102310241025102610271028102910301031103210331034103510361037103810391040104110421043104410451046104710481049105010511052105310541055105610571058105910601061106210631064106510661067106810691070107110721073107410751076107710781079108010811082108310841085108610871088108910901091109210931094109510961097109810991100110111021103110411051106110711081109111011111112111311141115111611171118111911201121112211231124112511261127112811291130113111321133113411351136113711381139114011411142114311441145114611471148114911501151115211531154115511561157115811591160116111621163116411651166116711681169117011711172117311741175117611771178117911801181118211831184118511861187118811891190119111921193119411951196119711981199120012011202120312041205120612071208120912101211121212131214121512161217121812191220122112221223122412251226122712281229123012311232123312341235123612371238123912401241124212431244124512461247124812491250125112521253125412551256125712581259126012611262126312641265126612671268126912701271127212731274 |
- """Chat orchestration without HTTP response objects (main.py split stage 4)."""
- from __future__ import annotations
- import json
- import re
- import sqlite3
- from dataclasses import dataclass, field
- from datetime import date, datetime
- from zoneinfo import ZoneInfo
- from hello_agents.core.exceptions import HelloAgentsException
- from starlette.concurrency import run_in_threadpool
- from fithealth_agent.context_budget import AGENT_INPUT_MAX_CHARS, ContextInputError
- from fithealth_agent.health_safety import (
- CAUTION, URGENT, RiskFinding, acute_pain_finding, emergency_reply,
- merge_findings, risk_directive, screen_health_risk, urgent_plan_block_note,
- )
- from fithealth_agent.info_store import (
- MemoryStoreDegradedError, fact_is_active, fact_retention_class,
- normalize_memory_facts, parse_weekly_schedule, resolve_confirmed_memory_facts,
- )
- from fithealth_agent.muscle_map import resolve_muscles_for_exercise
- from fithealth_agent.muscle_recovery import (
- MuscleRecoverySnapshot, build_recovery_snapshot,
- soreness_reply_needs_clarification,
- )
- from fithealth_agent.plan_workflow import PlanWorkflowState, state_after_generation
- from fithealth_agent.runtime import deps
- from fithealth_agent.runtime.deps import logger
- from fithealth_agent.runtime.responses import context_error_response
- from fithealth_agent.domain.intent_rules import (
- _is_safety_bypass_request, current_instruction_override, is_profile_query,
- is_training_record_query, is_training_related, navigation_only_message,
- )
- from fithealth_agent.domain.memory_view import (
- _memory_confirmation_decision,
- active_temporary_health_facts as _active_temporary_health_facts,
- confirmed_memory_profile as _confirmed_memory_profile,
- confirmed_weekly_schedule as _confirmed_weekly_schedule,
- format_cross_session_memories as _format_cross_session_memories,
- temporary_constraint_for_message, youtube_channels_to_avoid as _youtube_channels_to_avoid,
- )
- from fithealth_agent.domain.plan_context import (
- confirmed_weekly_schedule_entry, daily_schedule_constraint as _daily_schedule_constraint,
- format_plan_context, resolve_plan_context as _resolve_plan_context,
- scheduled_plan_for_message, scheduled_weekly_entry_for_message,
- )
- from fithealth_agent.domain.plan_validation import (
- extract_plan_subject, infer_plan_title, is_generic_training_subject,
- looks_like_complete_training_plan, most_recent_complete_training_plan,
- plan_card_title, plan_validation_fact_usage, validate_generated_training_plan,
- )
- from fithealth_agent.domain.profile_rules import (
- equipment_change_preview, merge_profile_updates_with_existing, onboarding_reply,
- profile_summary, validate_profile_tool_updates,
- )
- from fithealth_agent.domain.recovery_view import (
- _recovery_context_payload, _subject_recovery_regions,
- explicitly_requested_recovery_regions,
- )
- from fithealth_agent.domain.record_view import (
- _nutrition_record_date, _requested_record_date,
- _training_record_items,
- )
- @dataclass
- class ChatTurn:
- payload: dict[str, object]
- message: str
- history: list[dict]
- source: str
- profile: dict[str, object] = field(default_factory=dict)
- memories: list[dict] = field(default_factory=list)
- plan_context: dict[str, object] = field(default_factory=dict)
- artifacts: list[dict[str, object]] = field(default_factory=list)
- debug: dict[str, object] = field(default_factory=dict)
- @classmethod
- def from_payload(cls, payload: dict[str, object]) -> "ChatTurn":
- return cls(
- payload=payload,
- message=str(payload["message"]),
- history=list(payload["history"]),
- source=str(payload["source"]),
- )
- @dataclass(frozen=True)
- class ChatResult:
- body: dict[str, object]
- status_code: int = 200
- def stage_request_context(turn: ChatTurn) -> ChatTurn:
- """Initialize store-backed state after the route has validated the request."""
- turn.profile = deps.profile_store.get_profile()
- turn.debug["external_models_enabled"] = deps.external_model_settings_store.get()[
- "external_models_enabled"
- ]
- return turn
- def stage_health_safety(turn: ChatTurn) -> ChatTurn:
- """Named boundary for the health/soreness phase executed by the workflow."""
- return turn
- def stage_local_navigation(turn: ChatTurn) -> ChatTurn:
- """Named boundary for local record/profile navigation short circuits."""
- return turn
- def stage_intent_routing(turn: ChatTurn) -> ChatTurn:
- """Named boundary for constrained chat intent routing."""
- return turn
- def stage_plan_context(turn: ChatTurn) -> ChatTurn:
- """Named boundary for plan-context resolution and agent input assembly."""
- return turn
- def stage_postprocess(turn: ChatTurn) -> ChatTurn:
- """Named boundary for artifacts, validation and response metadata."""
- return turn
- _DURABLE_MEMORY_CUE = re.compile(
- r"(?:以后|今后|从今往后|长期|一直|每周|每星期|周[一二三四五六日天]|"
- r"不喜欢|喜欢|讨厌|避免|不要|不做|不练|我有|我是|慢性|骨盆(?:前|后)倾|"
- r"下交叉|臀肌无力|髂腰肌|核心无力|腘绳肌|竖脊肌|腰肌劳损|体态矫正|久坐)"
- )
- def format_cross_session_memories(memories: list[dict] | None) -> str:
- return _format_cross_session_memories(
- memories, resolve_facts_fn=resolve_confirmed_memory_facts
- )
- def youtube_channels_to_avoid(
- memories: list[dict] | None, current_message: str = ""
- ) -> list[str]:
- return _youtube_channels_to_avoid(
- memories,
- current_message,
- resolve_facts_fn=resolve_confirmed_memory_facts,
- )
- def confirmed_memory_profile(memories: list[dict] | None) -> str:
- return _confirmed_memory_profile(
- memories, resolve_facts_fn=resolve_confirmed_memory_facts
- )
- def confirmed_weekly_schedule(memories: list[dict] | None) -> dict[str, str]:
- return _confirmed_weekly_schedule(
- memories,
- resolve_facts_fn=resolve_confirmed_memory_facts,
- parse_schedule_fn=parse_weekly_schedule,
- )
- def active_temporary_health_facts(
- memories: list[dict] | None,
- ) -> list[dict[str, object]]:
- return _active_temporary_health_facts(
- memories, resolve_facts_fn=resolve_confirmed_memory_facts
- )
- def format_response(response: object) -> str:
- if isinstance(response, str):
- return response
- if isinstance(response, dict):
- return json.dumps(response, ensure_ascii=False, indent=2)
- return str(response)
- def resolve_plan_context(
- message: str,
- profile: dict[str, object],
- memories: list[dict] | None,
- recovery: MuscleRecoverySnapshot | None = None,
- routed_intent: dict[str, object] | None = None,
- soreness_reports: list[object] | None = None,
- ) -> dict[str, object]:
- return _resolve_plan_context(
- message=message,
- profile=profile,
- memories=memories,
- recovery=recovery,
- routed_intent=routed_intent,
- soreness_reports=soreness_reports,
- current_instruction_override_fn=current_instruction_override,
- is_safety_bypass_request_fn=_is_safety_bypass_request,
- recovery_context_payload_fn=_recovery_context_payload,
- subject_recovery_regions_fn=_subject_recovery_regions,
- explicitly_requested_recovery_regions_fn=explicitly_requested_recovery_regions,
- scheduled_weekly_entry_for_message_fn=scheduled_weekly_entry_for_message,
- )
- def daily_schedule_constraint(message: str, memories: list[dict] | None) -> str:
- return _daily_schedule_constraint(
- message,
- memories,
- current_instruction_override_fn=current_instruction_override,
- scheduled_weekly_entry_for_message_fn=scheduled_weekly_entry_for_message,
- )
- def _plan_muscle_resolver(name: str) -> list:
- return resolve_muscles_for_exercise(
- name,
- allow_external_models=deps.external_model_settings_store.get()["external_models_enabled"],
- )
- def build_agent_input(
- message: str,
- profile: dict[str, object],
- history: list[dict] = None,
- memories: list[dict] | None = None,
- risk: RiskFinding | None = None,
- plan_context: dict[str, object] | None = None,
- ) -> str:
- current_time = __import__("datetime").datetime.now(__import__("zoneinfo").ZoneInfo("Asia/Shanghai"))
- current_time_context = (
- "[Current time]\n"
- + f"Current date: {current_time.date().isoformat()}\n"
- + f"Current time: {current_time.isoformat(timespec='seconds')} (Asia/Shanghai)\n"
- + "Resolve today/now/this week from this anchor unless the user explicitly provides a date.\n\n"
- )
- history_str = ""
- if history:
- history_str = "【近期对话上下文】\n"
- for msg in history:
- role = "用户" if msg.get("role") == "user" else "助手"
- text = msg.get("text", "")
- history_str += f"{role}:{text}\n"
- history_str += "\n"
- goal = str(profile.get("goal") or "未设置").strip() or "未设置"
- plan_context = dict(plan_context or resolve_plan_context(message, profile, memories))
- plan_context["goal"] = goal
- result = (
- # 健康风险提示放在最前:它的优先级高于主目标、周计划和任何偏好,
- # 位置靠前也能降低被后续上下文冲淡的概率。
- f"{current_time_context}"
- f"{risk_directive(risk) if risk is not None else ''}"
- "【用户档案(稳定事实)】\n"
- f"{profile_summary(profile)}\n\n"
- # 优先级说明只此一段:原先这里另有一段【训练设计优先级】,与
- # format_plan_context 的一行版、以及从未被调用的 training_priority_context
- # 三处并存(BUG-03/BUG-04)。现在全部合并进 format_plan_context。
- f"{format_plan_context(plan_context)}"
- f"{daily_schedule_constraint(message, memories) if plan_context.get('decision') == 'follow_schedule' else ''}"
- f"{format_cross_session_memories(memories)}"
- f"{history_str}"
- f"【最新用户消息】:{message}"
- )
- if len(result) > AGENT_INPUT_MAX_CHARS:
- raise ContextInputError(
- "CONTEXT_BUDGET_EXCEEDED",
- "整理后的上下文仍然过长,请缩短当前消息或训练计划。",
- status_code=413,
- )
- return result
- def _current_recovery_snapshot(garmin_recovery_hours: float) -> MuscleRecoverySnapshot:
- now = datetime.now(ZoneInfo("Asia/Shanghai"))
- deps.soreness_store.cleanup_expired(now=now)
- return build_recovery_snapshot(
- deps.daily_record_store.list_records(),
- now=now,
- soreness=deps.soreness_store.list_reports(active_only=True, now=now),
- garmin_recovery_hours=garmin_recovery_hours,
- lookback_days=7,
- allow_external_models=deps.external_model_settings_store.get()["external_models_enabled"],
- )
- def _soreness_report_payload(report: object) -> dict[str, object]:
- now = datetime.now(ZoneInfo("Asia/Shanghai"))
- return {
- "id": report.id,
- "region": report.region,
- "muscle_ids": list(report.muscle_ids),
- "level": report.level,
- "reported_at": report.reported_at.isoformat(),
- "expires_at": report.expires_at.isoformat(),
- "evidence": report.evidence,
- "expired": bool(getattr(report, "expired", False)) or report.expires_at <= now,
- }
- def _soreness_acknowledgement(reports: list[object]) -> str:
- labels = {
- "recovered": "已恢复",
- "sore": "酸痛,将相应减量",
- "painful": "疼痛,已升级为安全信号",
- }
- details = [f"{report.region}{labels.get(report.level, report.level)}" for report in reports]
- return "已记录:" + ";".join(details) + "。记录将在 72 小时后自动过期,可在数据管理中修改或删除。"
- def _is_only_soreness_feedback(message: str) -> bool:
- normalized = re.sub(r"\s+", "", str(message or ""))
- stripped = normalized
- from fithealth_agent.muscle_recovery import REGION_ALIASES
- for aliases in REGION_ALIASES.values():
- for alias in aliases:
- stripped = stripped.replace(alias, "")
- stripped = re.sub(r"(?:有点|一点|稍微|轻微|很|非常|比较|还|又|酸痛|酸|疼痛|疼|痛|不酸|不疼|不痛|正常|好了|恢复了)", "", stripped)
- if re.search(r"怎么办|怎么处理|如何|建议|要不要|需要吗|能不能|可以吗", normalized):
- return False
- if len(re.sub(r"[,。!?、;;::吗呢吧呀\?]", "", stripped)) < 4:
- return True
- return not bool(re.search(
- r"(?:训练计划|帮我|怎么练|如何练|能不能练|可以练|继续练|照常练|就要练|坚持练|"
- r"安排|推荐动作|生成计划|查看|查询|保存|修改|删除|怎么办|如何处理|建议|本周|这周)",
- re.sub(r"\s+", "", message),
- ))
- def _immediate_memory_candidate(message: str, *, allow_external_models: bool) -> dict[str, object] | None:
- """Persist explicit durable facts as a pending candidate during chat.
- Logout remains a compatibility sweep, but durable facts must not depend on a
- browser close event. This path intentionally uses the same information
- router and InfoStore normalisation as logout; it never auto-confirms facts.
- """
- if not allow_external_models or not _DURABLE_MEMORY_CUE.search(str(message or "")):
- return None
- decision = deps.route_information(
- [{"role": "user", "text": str(message)}],
- allow_external_models=True,
- )
- if not decision.get("save"):
- return None
- facts = [
- fact
- for fact in normalize_memory_facts(decision.get("facts"))
- if fact_retention_class(fact) in {"long_term", "medium"}
- ]
- if not facts:
- return None
- # Retries and later logout sweeps must not create another pending card for
- # the same active fact. Different weekly schedules remain distinct here;
- # version conflict handling still applies when the user confirms them.
- deps.info_store.cleanup_expired()
- existing = {
- (str(fact.get("namespace")), str(fact.get("key")), str(fact.get("value")).casefold())
- for memory in deps.info_store.get_all()
- if isinstance(memory, dict)
- for fact in normalize_memory_facts(memory.get("facts"))
- if fact.get("status") in {"active", "rejected"} and fact_is_active(fact)
- }
- new_facts = [
- fact for fact in facts
- if (str(fact.get("namespace")), str(fact.get("key")), str(fact.get("value")).casefold())
- not in existing
- ]
- if not new_facts:
- return None
- memory_type = str(decision.get("type") or "training_feedback")
- entry = deps.info_store.add_entry(
- summary=str(decision.get("summary") or "")[:200],
- metadata={
- "reason": decision.get("reason", ""),
- "pipeline_stage": decision.get("pipeline_stage", "level3"),
- "capture": "chat_turn",
- },
- expires_at=None,
- memory_type=memory_type,
- importance=int(decision.get("importance", 1)),
- user_confirmed=False,
- facts=new_facts,
- )
- return {
- "entry_id": entry["id"],
- "summary": entry["summary"],
- "facts": [
- {"namespace": fact["namespace"], "key": fact["key"], "value": fact["value"]}
- for fact in entry["facts"]
- ],
- }
- def _date_range_memory_candidate(constraint: dict[str, object]) -> dict[str, object] | None:
- if constraint.get("scope") != "date_range":
- return None
- fact = normalize_memory_facts([{
- "namespace": "health",
- "key": "injury_or_constraint",
- "status": "active",
- "evidence": str(constraint.get("value") or "")[:160],
- **constraint,
- }])
- if not fact:
- return None
- target = fact[0]
- identity = (target["namespace"], target["key"], str(target["value"]).casefold())
- for memory in deps.info_store.get_all():
- for existing in normalize_memory_facts(memory.get("facts") if isinstance(memory, dict) else None):
- if (
- existing.get("status") in {"active", "rejected"}
- and fact_is_active(existing)
- and (existing.get("namespace"), existing.get("key"), str(existing.get("value")).casefold()) == identity
- ):
- return None
- entry = deps.info_store.add_entry(
- summary=f"临时训练限制:{target['value']}",
- metadata={"capture": "chat_date_range"},
- expires_at=None,
- memory_type="constraint",
- importance=5,
- user_confirmed=False,
- facts=fact,
- )
- return {
- "entry_id": entry["id"],
- "summary": entry["summary"],
- "facts": [{
- "namespace": target["namespace"], "key": target["key"], "value": target["value"],
- "scope": target["scope"], "valid_from": target.get("valid_from"),
- "valid_until": target.get("valid_until"),
- }],
- }
- def _acute_injury_memory_candidate(message: str, finding: RiskFinding) -> dict[str, object] | None:
- """Record an urgent, user-stated injury as a short-lived pending fact.
- This is deliberately local and deterministic. An acute safety statement
- must survive an LLM timeout, but remains unconfirmed and expires under the
- existing recovery-status TTL unless the user confirms it.
- """
- if finding.level != "urgent":
- return None
- from fithealth_agent.muscle_recovery import REGION_ALIASES
- regions = [
- region for region, aliases in REGION_ALIASES.items()
- if any(alias in str(message or "") for alias in aliases)
- ]
- concise_value = "、".join(dict.fromkeys(regions)) + (":" if regions else "") + finding.summary
- fact = normalize_memory_facts([{
- "namespace": "health",
- "key": "recovery_status",
- "value": concise_value[:120],
- "status": "active",
- "evidence": str(message or "").strip()[:160],
- }])
- if not fact:
- return None
- deps.info_store.cleanup_expired()
- candidate_fact = fact[0]
- identity = (
- str(candidate_fact["namespace"]),
- str(candidate_fact["key"]),
- str(candidate_fact["value"]).casefold(),
- )
- for memory in deps.info_store.get_all():
- if not isinstance(memory, dict):
- continue
- for existing in normalize_memory_facts(memory.get("facts")):
- existing_identity = (
- str(existing.get("namespace")), str(existing.get("key")),
- str(existing.get("value")).casefold(),
- )
- if existing.get("status") in {"active", "rejected"} and fact_is_active(existing) and existing_identity == identity:
- return None
- entry = deps.info_store.add_entry(
- summary=f"用户报告急性伤病信号:{finding.summary}",
- metadata={"capture": "chat_safety", "risk_level": finding.level, "risk_labels": list(finding.labels)},
- expires_at=None,
- memory_type="training_feedback",
- importance=5,
- user_confirmed=False,
- facts=fact,
- )
- return {
- "entry_id": entry["id"],
- "summary": entry["summary"],
- "facts": [{"namespace": "health", "key": "recovery_status", "value": candidate_fact["value"]}],
- }
- async def chat(payload: dict[str, object]) -> ChatResult:
- turn = stage_request_context(ChatTurn.from_payload(payload))
- turn = stage_health_safety(turn)
- turn = stage_intent_routing(turn)
- turn = stage_local_navigation(turn)
- turn = stage_plan_context(turn)
- turn = stage_postprocess(turn)
- payload = turn.payload
- message = turn.message
- history = turn.history
- profile = turn.profile
- source = turn.source
- plan_context = payload["plan_context"]
- garmin_recovery_hours = payload["garmin_recovery_hours"]
- external_models_enabled = bool(turn.debug["external_models_enabled"])
- asked_regions = payload["soreness_prompt_regions"]
- pending_memory_entry_ids = payload["pending_memory_entry_ids"]
- saved_soreness: list[object] = []
- soreness_ack = ""
- memory_candidates: list[dict[str, object]] = []
- def _chat_response(
- reply: str,
- *,
- artifact: dict[str, object] | None = None,
- source: str | None = None,
- status_code: int = 200,
- include_soreness_ack: bool = True,
- **extra: object,
- ) -> ChatResult:
- final_reply = str(reply or "")
- if include_soreness_ack and soreness_ack:
- final_reply = soreness_ack + ("\n\n" + final_reply if final_reply else "")
- if memory_candidates:
- candidate_labels = [
- str(candidate.get("summary") or "检测到一条长期偏好或限制")
- for candidate in memory_candidates
- ]
- final_reply += (
- ("\n\n" if final_reply else "")
- + "检测到待确认的跨会话记忆:"
- + ";".join(candidate_labels)
- + "。要加入跨会话记忆吗?"
- )
- body: dict[str, object] = {
- **extra,
- "reply": final_reply,
- "artifact": artifact,
- "soreness_saved": bool(saved_soreness),
- "soreness_reports": [
- _soreness_report_payload(item) for item in saved_soreness
- ],
- "asked_regions": asked_regions,
- }
- if source is not None:
- body["source"] = source
- if memory_candidates:
- body["memory_candidate"] = memory_candidates[0]
- body["memory_candidates"] = list(memory_candidates)
- return ChatResult(body=body, status_code=status_code)
- confirmation_decision = _memory_confirmation_decision(message)
- if pending_memory_entry_ids and confirmation_decision is not None:
- changed = 0
- try:
- for entry_id in pending_memory_entry_ids:
- result = (
- deps.info_store.confirm_entry(entry_id)
- if confirmation_decision
- else deps.info_store.reject_entry(entry_id)
- )
- if result is not None:
- changed += 1
- except ValueError as exc:
- return _chat_response(
- f"记忆确认失败:{exc}", source="memory_confirmation", status_code=409
- )
- if not changed:
- return _chat_response(
- "这些待确认记忆已不存在或已经处理。", source="memory_confirmation", status_code=404
- )
- return _chat_response(
- f"已{'加入' if confirmation_decision else '拒绝'} {changed} 条跨会话记忆。",
- source="memory_confirmation",
- memory_confirmation={"accepted": confirmation_decision, "count": changed},
- )
- user_health_statement = None
- if source == "chat":
- user_health_statement = await run_in_threadpool(deps.classify_user_health_statement, message)
- immediate_memory_candidate: dict[str, object] | None = None
- health_risk = None
- if source == "chat" and user_health_statement is not False:
- health_risk = screen_health_risk(message)
- recovery_snapshot = _current_recovery_snapshot(garmin_recovery_hours)
- temporary_constraint = temporary_constraint_for_message(message) if source == "chat" else None
- if temporary_constraint is not None and temporary_constraint.get("scope") == "date_range":
- try:
- date_range_candidate = _date_range_memory_candidate(temporary_constraint)
- if date_range_candidate is not None:
- memory_candidates.append(date_range_candidate)
- except (MemoryStoreDegradedError, OSError, ValueError):
- logger.exception("日期区间限制候选保存失败")
- soreness_reports = (
- deps.parse_soreness_reply(message, asked_regions)
- if source == "chat" and (bool(asked_regions) or user_health_statement is not False)
- else []
- )
- if (
- source == "chat"
- and not soreness_reports
- and not (health_risk is not None and health_risk.level in {"urgent", "emergency"})
- and soreness_reply_needs_clarification(message, asked_regions)
- ):
- return _chat_response(
- "你说了当前感受,但我刚才询问了多个部位。请明确是哪一个区域,例如“手臂有点酸”或“腿部正常”。",
- source="soreness_clarification",
- )
- if soreness_reports:
- try:
- saved_soreness = list(deps.soreness_store.add_reports(soreness_reports))
- except (OSError, ValueError) as exc:
- logger.exception("肌群酸痛反馈保存失败")
- return _chat_response(f"酸痛反馈未能保存:{exc}", status_code=503)
- soreness_ack = _soreness_acknowledgement(saved_soreness)
- recovery_snapshot = _current_recovery_snapshot(garmin_recovery_hours)
- active_soreness_reports = deps.soreness_store.list_reports(active_only=True)
- # AGENT-01:确定性健康风险筛查。
- # 放在意图路由之前有三个原因:(1) 急症不该等一次外部模型往返;
- # (2) 关闭联网模型或网络故障时同样要生效;(3) 这条判断不依赖任何
- # 已确认记忆——用户在**当前这条消息**里新出现的症状本来就不在记忆里。
- # 只筛查对话消息;上传的训练计划正文里出现症状词不算用户主诉。
- painful_regions = [report.region for report in soreness_reports if report.level == "painful"]
- if painful_regions:
- health_risk = merge_findings(health_risk, acute_pain_finding(painful_regions))
- if source == "chat" and health_risk is not None and health_risk.level == "urgent":
- try:
- acute_candidate = _acute_injury_memory_candidate(message, health_risk)
- if acute_candidate is not None:
- memory_candidates.append(acute_candidate)
- except (MemoryStoreDegradedError, OSError, ValueError):
- logger.exception("急性伤病候选保存失败")
- if health_risk is not None and health_risk.level == "emergency":
- # Emergency guidance stays visually undiluted. The structured flag still
- # tells the client that any soreness report from this turn was saved.
- return _chat_response(
- emergency_reply(health_risk),
- source="health_risk_block",
- include_soreness_ack=False,
- health_risk={"level": health_risk.level, "signals": list(health_risk.labels)},
- workflow_state=str(PlanWorkflowState.CONSTRAINT_CONFLICT),
- )
- if source == "chat" and (health_risk is None or health_risk.level in {CAUTION, URGENT}):
- try:
- durable_candidate = await run_in_threadpool(
- _immediate_memory_candidate,
- message,
- allow_external_models=external_models_enabled,
- )
- if durable_candidate is not None:
- memory_candidates.append(durable_candidate)
- except (MemoryStoreDegradedError, OSError, ValueError):
- logger.exception("即时长期记忆候选保存失败")
- immediate_memory_candidate = memory_candidates[0] if memory_candidates else None
- if soreness_reports and _is_only_soreness_feedback(message):
- deps.info_store.cleanup_expired()
- memories = deps.info_store.get_context_memories(n=5)
- resolved_plan_context = resolve_plan_context(
- message, profile, memories, recovery=recovery_snapshot,
- soreness_reports=active_soreness_reports,
- )
- reply = ""
- if painful_regions and health_risk is not None:
- reply = urgent_plan_block_note(health_risk)
- return _chat_response(
- reply,
- source="local_soreness_feedback",
- plan_context=resolved_plan_context,
- health_risk=(
- {"level": health_risk.level, "signals": list(health_risk.labels)}
- if health_risk is not None else None
- ),
- workflow_state=resolved_plan_context["workflow_state"],
- **({"memory_candidate": immediate_memory_candidate} if immediate_memory_candidate else {}),
- **({"memory_candidates": memory_candidates} if memory_candidates else {}),
- )
- if source == "uploaded_plan":
- updates, updated_fields = {}, []
- chat_intent = None
- else:
- chat_intent = await run_in_threadpool(
- deps.route_chat_intent,
- message,
- allow_external_models=external_models_enabled,
- )
- updates, updated_fields = validate_profile_tool_updates(
- chat_intent.profile_updates, message
- )
- profile_update_artifact = (
- {
- "type": "profile_update",
- "updates": updates,
- "fields": updated_fields,
- **(
- {"equipment_diff": equipment_change_preview(profile, updates)}
- if equipment_change_preview(profile, updates) is not None else {}
- ),
- }
- if updates
- else None
- )
- pending_view_artifact: dict[str, object] | None = None
- pending_reply_prefix = ""
- pending_view_source: str | None = None
- # BUG-02:外部意图路由不可用时(关闭联网模型、缺 API key、超时、网络抖动),
- # route_chat_intent 会返回一个所有位都为 False 的空 ChatIntent,于是
- # "查看训练记录"这类**纯本地读操作**一路掉到下面的 503——而欢迎语里
- # 明明写着可以这么输入。这里用已有的确定性规则兜底,让本地能力在离线
- # 状态下同样可用。
- wants_training_records = (
- chat_intent is not None and chat_intent.view_training_records
- ) or (source == "chat" and is_training_record_query(message))
- if wants_training_records:
- records = _training_record_items(deps.daily_record_store.list_records())
- default_date = (
- (chat_intent.training_records_date if chat_intent is not None else "")
- or _requested_record_date(message)
- or (
- records[0]["date"]
- if records
- else datetime.now(ZoneInfo("Asia/Shanghai")).date().isoformat()
- )
- )
- navigation_only = navigation_only_message(message)
- pending_view_artifact = {
- "type": "training_records",
- "default_date": default_date,
- }
- pending_reply_prefix = "已打开本地训练记录。选择日期后,可在右侧下拉框查看当天的每一条训练。"
- pending_view_source = "local_training_records"
- if navigation_only or not external_models_enabled:
- offline_note = (
- "\n\n外部模型当前已关闭,因此暂时无法继续回答同一条消息中的其他问题。"
- if not navigation_only and not external_models_enabled else ""
- )
- return _chat_response(
- pending_reply_prefix + offline_note,
- artifact=pending_view_artifact,
- source=pending_view_source,
- )
- wants_nutrition_records = (
- chat_intent is not None and chat_intent.view_nutrition_records
- ) or (
- source == "chat"
- and any(word in message for word in ("查看饮食", "查看营养", "营养数据", "饮食数据", "摄入数据"))
- )
- if wants_nutrition_records:
- default_date = (
- (chat_intent.nutrition_records_date if chat_intent is not None else "")
- or _nutrition_record_date(message)
- or datetime.now(ZoneInfo("Asia/Shanghai")).date().isoformat()
- )
- navigation_only = navigation_only_message(message)
- pending_view_artifact = pending_view_artifact or {
- "type": "nutrition_records",
- "default_date": default_date,
- }
- nutrition_prefix = "已打开本地营养记录。选择日期后,可在右侧下拉框查看当天的不同饮食组。"
- pending_reply_prefix = "\n\n".join(filter(None, (pending_reply_prefix, nutrition_prefix)))
- pending_view_source = pending_view_source or "local_nutrition_records"
- if navigation_only or not external_models_enabled:
- offline_note = (
- "\n\n外部模型当前已关闭,因此暂时无法继续回答同一条消息中的其他问题。"
- if not navigation_only and not external_models_enabled else ""
- )
- return _chat_response(
- pending_reply_prefix + offline_note,
- artifact=pending_view_artifact,
- source=pending_view_source,
- )
- if chat_intent is not None and chat_intent.save_existing_training_plan:
- # BUG-05:优先用服务端缓存的完整正文。回捞聊天历史只作为进程重启
- # 后的兜底——history 已被上下文预算截断过,可能是残片。
- # 用 find() 而不是 latest():一个会话里可能生成过多份计划,用户说
- # "保存那份腿的计划"时无条件取最后一份会静默存错。
- draft = deps.plan_draft_cache.find(
- subject=chat_intent.saved_plan_subject or "",
- suggested_date=chat_intent.saved_plan_date or "",
- )
- content = draft.content if draft is not None else most_recent_complete_training_plan(history)
- if not content:
- return _chat_response(
- "没有找到可保存的完整训练计划。\n\n"
- "可能的原因:还没有生成过计划;或者页面刷新过、服务重启过,"
- "导致上一份计划的正文已经不在。请让我重新生成一份,"
- "再点击计划下方的“保存计划”。",
- source="local_plan_save",
- status_code=400,
- )
- subject = (
- (draft.subject if draft is not None else "")
- or chat_intent.saved_plan_subject
- or extract_plan_subject(content)
- )
- title = plan_card_title(
- (draft.title if draft is not None else "")
- or infer_plan_title(content, subject),
- subject,
- )
- plan_date = (
- chat_intent.saved_plan_date
- or (draft.suggested_date if draft is not None else "")
- or datetime.now(ZoneInfo("Asia/Shanghai")).date().isoformat()
- )
- deps.info_store.cleanup_expired()
- memories = deps.info_store.get_context_memories(n=5)
- expected_subject = subject
- try:
- saved_day = date.fromisoformat(plan_date)
- scheduled_entry = confirmed_weekly_schedule_entry(deps.info_store.get_all(), saved_day)
- if (
- (not subject or is_generic_training_subject(subject))
- and scheduled_entry is not None
- and scheduled_entry.get("type") != "rest"
- ):
- expected_subject = str(scheduled_entry.get("subject") or subject)
- except ValueError:
- pass
- safety_constraints = [
- str(fact.get("value") or "")
- for fact in resolve_confirmed_memory_facts(memories)
- if fact.get("status", "active") == "active"
- and fact.get("user_confirmed", True)
- and fact.get("namespace") == "health"
- and fact.get("key") in {"injury_or_constraint", "recovery_status"}
- ]
- plan_validation = validate_generated_training_plan(
- content,
- memories,
- expected_subject,
- recovery=recovery_snapshot,
- safety_constraints=safety_constraints,
- resolver=_plan_muscle_resolver,
- check_subject=False,
- )
- saved_goal_alignment = await run_in_threadpool(
- deps.validate_plan_goal_alignment,
- content,
- user_request=message,
- fallback_subject=expected_subject,
- weekly_subject=(
- str(scheduled_entry.get("subject") or "")
- if "scheduled_entry" in locals() and scheduled_entry is not None
- else ""
- ),
- schedule_decision="override_today" if subject and not is_generic_training_subject(subject) else "follow_schedule",
- allow_external_models=external_models_enabled,
- )
- if not saved_goal_alignment.get("passed"):
- missing = "、".join(saved_goal_alignment.get("missing_subjects") or [expected_subject])
- plan_validation.append(f"计划未覆盖用户要求的训练目标:{missing}")
- if plan_validation:
- return _chat_response(
- "该训练计划未通过已确认记忆的后端校验,因此不能保存。\n\n## 计划校验未通过\n"
- + "\n".join(f"- {item}" for item in plan_validation)
- + "\n请先修正以上问题,再保存训练计划。"
- "\n\n你还问了另一件事;处理完这项安全冲突后,我再继续回答。",
- source="local_plan_save",
- status_code=409,
- plan_validation={"passed": False, "violations": plan_validation},
- )
- return _chat_response(
- "已找到本次会话中最近生成的完整训练计划,请确认下方信息后保存。",
- artifact={
- "type": "training_plan",
- "subject": subject,
- "title": title,
- "content": content,
- "source": "agent_generated",
- "suggested_date": plan_date,
- # 带上 draft_id,让后续 /plans 也走"按 id 取服务端原文"这条路,
- # 而不是退回靠前端传值——否则这条路径只是碰巧内容完整才没出错。
- **({"draft_id": draft.id} if draft is not None else {}),
- },
- source="local_plan_save",
- )
- if (chat_intent is not None and chat_intent.view_profile) or (
- source == "chat" and is_profile_query(message)
- ):
- deps.info_store.cleanup_expired()
- navigation_only = navigation_only_message(message)
- profile_prefix = (
- "## 个人档案\n"
- + profile_summary(profile, include_birth_date=False)
- + "\n\n## 跨会话记忆\n"
- + confirmed_memory_profile(deps.info_store.get_all())
- + "\n\n待确认记忆不会用于后续对话;可在“数据管理 → 临时记忆”中查看、确认或删除。"
- )
- pending_reply_prefix = "\n\n".join(filter(None, (pending_reply_prefix, profile_prefix)))
- pending_view_source = pending_view_source or "local_profile"
- if navigation_only or not external_models_enabled:
- offline_note = (
- "\n\n外部模型当前已关闭,因此暂时无法继续回答同一条消息中的其他问题。"
- if not navigation_only and not external_models_enabled else ""
- )
- return _chat_response(
- pending_reply_prefix + offline_note,
- artifact=pending_view_artifact,
- source=pending_view_source,
- )
- projected = merge_profile_updates_with_existing(profile, updates) if updates else profile
- is_complete = deps.profile_store.is_complete(projected)
- if not is_complete:
- missing = deps.profile_store.missing_fields(projected)
- return _chat_response(
- onboarding_reply(projected, missing, updated_fields)
- + "\n\n你还问了另一件事;补完档案后,我再继续回答。",
- artifact=profile_update_artifact,
- source="onboarding",
- )
- if updated_fields:
- return _chat_response(
- "\u7528\u6237\u4fe1\u606f\u5df2\u5b8c\u5584\uff0c\u8bf7\u786e\u8ba4\u4e0b\u65b9\u53d8\u66f4\u540e\u7ee7\u7eed\u3002\n\n"
- + profile_summary(projected, include_birth_date=False),
- artifact=profile_update_artifact,
- source="profile_update",
- )
- try:
- weekly_reply = deps.build_current_week_reply(
- message,
- health_store=deps.health_store,
- daily_record_store=deps.daily_record_store,
- profile=profile,
- today=datetime.now(ZoneInfo("Asia/Shanghai")).date(),
- recovery=recovery_snapshot,
- soreness_reports=active_soreness_reports,
- )
- except (OSError, ValueError, sqlite3.Error):
- logger.exception("本周数据本地查询失败")
- return _chat_response("本周数据读取失败,请检查已导入数据是否完整。", status_code=500)
- if weekly_reply is not None:
- return _chat_response(weekly_reply, source="local_weekly_summary")
- if not external_models_enabled:
- # BUG-02:以前这里只丢一句"外部模型已关闭",用户不知道离线到底还能做什么。
- # 上面的本地分支(训练记录 / 个人档案 / 本周汇总)已经先行拦截,能走到
- # 这里说明确实需要模型;给出可直接照抄的本地指令,比空泛一句有用得多。
- local_commands = "「查看训练记录」「查看个人信息」「查看本周数据」"
- hint = (
- f"这类问题需要模型参与。不过你可以直接输入 {local_commands} 来查看已保存的本地数据;"
- "健康数据的导入、训练组编辑、备份与删除也都不受影响。"
- if is_training_related(message)
- else f"你仍可使用本地功能:输入 {local_commands},以及健康数据导入、训练编辑、备份和删除。"
- )
- return _chat_response(
- "外部模型已关闭,因此不会发送你的对话、档案或记忆。" + hint,
- source="external_models_disabled",
- status_code=503,
- )
- try:
- deps.info_store.cleanup_expired()
- memory_intent = (
- "create_training_plan"
- if chat_intent is not None and chat_intent.create_training_plan
- else "search_youtube_video" if "youtube" in message.casefold() or "视频" in message else ""
- )
- memories = deps.info_store.get_context_memories(n=5, intent=memory_intent)
- candidate_memories = deps.info_store.get_enforceable_memories() if hasattr(deps.info_store, "get_enforceable_memories") else None
- enforceable_memories = candidate_memories if isinstance(candidate_memories, list) else memories
- routed_plan_intent = (
- {
- "schedule_decision": chat_intent.schedule_decision,
- "subject": chat_intent.training_plan_subject,
- "requested_subjects": getattr(chat_intent, "requested_subjects", []),
- "temporary_health_constraints": list(dict.fromkeys(
- chat_intent.temporary_health_constraints
- + ([str(temporary_constraint["value"])] if temporary_constraint else [])
- )),
- "excluded_subjects": chat_intent.excluded_subjects,
- "needs_clarification": chat_intent.needs_clarification,
- }
- if chat_intent is not None and chat_intent.create_training_plan
- else None
- )
- resolved_plan_context = resolve_plan_context(
- message,
- profile,
- enforceable_memories,
- recovery=recovery_snapshot,
- routed_intent=routed_plan_intent,
- soreness_reports=active_soreness_reports,
- )
- scheduled_plan = (
- scheduled_plan_for_message(message, deps.info_store.get_all())
- if resolved_plan_context.get("decision") == "follow_schedule"
- else None
- )
- if scheduled_plan is not None:
- scheduled_date, scheduled_subject = scheduled_plan
- resolved_plan_context.update({
- "scheduled_subject": scheduled_subject,
- "scheduled_date": scheduled_date.isoformat(),
- "effective_subject": scheduled_subject,
- })
- if (
- chat_intent is not None
- and chat_intent.create_training_plan
- and resolved_plan_context.get("blocking_reasons")
- ):
- return _chat_response(
- "当前请求与已确认的安全限制冲突,不能按原要求生成计划。"
- "请先说明是否要修改或清除该安全限制。\n\n"
- "当前生效的限制:"
- + ";".join(resolved_plan_context.get("active_safety_constraints") or ["(未列出)"])
- + "\n\n你还问了另一件事;处理完这项安全冲突后,我再继续回答。",
- source="plan_context_safety_block",
- status_code=409,
- plan_context=resolved_plan_context,
- workflow_state=str(PlanWorkflowState.CONSTRAINT_CONFLICT),
- )
- if (
- chat_intent is not None
- and chat_intent.create_training_plan
- and resolved_plan_context.get("decision") == "rest_today"
- ):
- scheduled_subject = resolved_plan_context.get("scheduled_subject")
- scheduled_note = f",原周计划科目为{scheduled_subject}" if scheduled_subject else ""
- rest_reply = "已将本次安排为休息" + scheduled_note + "。今天不生成训练计划;恢复后可再安排下一次训练。"
- return _chat_response(
- rest_reply,
- artifact=profile_update_artifact,
- source="plan_context_rest",
- plan_context=resolved_plan_context,
- workflow_state=str(resolved_plan_context["workflow_state"]),
- )
- if chat_intent is not None and chat_intent.create_training_plan and resolved_plan_context.get("clarification_required"):
- return _chat_response(
- "你已表示本次不按周计划训练,但还没有说明是休息还是更换训练科目。请明确选择后,我再生成计划。",
- source="plan_context_clarification",
- status_code=409,
- plan_context=resolved_plan_context,
- workflow_state=str(PlanWorkflowState.NEEDS_CLARIFICATION),
- )
- avoided_youtube_channels = youtube_channels_to_avoid(memories, message)
- agent = deps.create_fithealth_agent(
- avoid_youtube_channels=avoided_youtube_channels
- )
- try:
- agent_input = build_agent_input(message, profile, history, memories, risk=health_risk, plan_context=resolved_plan_context)
- except ContextInputError as exc:
- return context_error_response(exc)
- answer = format_response(await run_in_threadpool(agent.run, agent_input))
- if pending_reply_prefix:
- answer = pending_reply_prefix + "\n\n" + answer
- artifact = None
- if source == "uploaded_plan" and looks_like_complete_training_plan(answer):
- subject = extract_plan_subject(answer, str(plan_context.get("subject") or ""))
- artifact = {
- "type": "training_plan",
- "subject": subject,
- "title": infer_plan_title(answer, subject),
- "content": answer,
- "source": "uploaded_optimized",
- "suggested_date": str(plan_context.get("suggested_date") or ""),
- }
- elif (
- source == "chat"
- and chat_intent is not None
- and chat_intent.create_training_plan
- and looks_like_complete_training_plan(answer)
- ):
- if scheduled_plan is not None:
- scheduled_date, subject = scheduled_plan
- suggested_date = scheduled_date.isoformat()
- else:
- subject = chat_intent.training_plan_subject
- suggested_date = ""
- artifact = {
- "type": "training_plan",
- "subject": subject if subject.endswith(("训练", "计划")) else subject + "训练",
- "title": plan_card_title(chat_intent.training_plan_title, subject),
- "content": answer,
- "source": "agent_generated",
- "suggested_date": suggested_date,
- }
- elif scheduled_plan is not None and looks_like_complete_training_plan(answer):
- scheduled_date, scheduled_subject = scheduled_plan
- artifact = {
- "type": "training_plan",
- "subject": scheduled_subject,
- "title": infer_plan_title(answer, scheduled_subject),
- "content": answer,
- "source": "agent_generated",
- "suggested_date": scheduled_date.isoformat(),
- }
- plan_validation: list[str] = []
- first_plan_validation: list[str] = []
- plan_auto_correction: dict[str, object] = {
- "attempted": False,
- "passed": False,
- "first_violations": [],
- "final_violations": [],
- }
- # AGENT-01(urgent 档):急性损伤信号下确定性地拒绝产出可保存的计划。
- # 放在计划校验之前——无论模型写了什么,后端都不放行这个 artifact。
- if (
- artifact
- and artifact.get("type") == "training_plan"
- and health_risk is not None
- and health_risk.blocks_training_plan()
- ):
- artifact = None
- answer += "\n\n" + urgent_plan_block_note(health_risk)
- if artifact and artifact.get("type") == "training_plan":
- expected_plan_subject = str(
- resolved_plan_context.get("effective_subject")
- or artifact.get("subject")
- or ""
- )
- plan_validation = validate_generated_training_plan(
- answer,
- enforceable_memories,
- expected_plan_subject,
- recovery=recovery_snapshot,
- safety_constraints=list(resolved_plan_context.get("active_safety_constraints") or []),
- explicitly_requested=set((resolved_plan_context.get("muscle_recovery") or {}).get("explicitly_requested") or []),
- resolver=_plan_muscle_resolver,
- check_subject=False,
- )
- goal_alignment = await run_in_threadpool(
- deps.validate_plan_goal_alignment,
- answer,
- user_request=message,
- requested_subjects=(getattr(chat_intent, "requested_subjects", []) if chat_intent is not None else []),
- fallback_subject=expected_plan_subject,
- weekly_subject=str(resolved_plan_context.get("scheduled_subject") or ""),
- schedule_decision=str(resolved_plan_context.get("decision") or "follow_schedule"),
- allow_external_models=external_models_enabled,
- )
- if not goal_alignment.get("passed"):
- missing = "、".join(goal_alignment.get("missing_subjects") or [expected_plan_subject])
- plan_validation.append(f"计划未覆盖用户要求的训练目标:{missing}")
- if plan_validation:
- first_plan_validation = list(plan_validation)
- plan_auto_correction.update({
- "attempted": True,
- "first_violations": first_plan_validation,
- })
- correction_prompt = (
- "你刚才生成的训练计划未通过后端安全校验。请只输出一份完整、可直接使用的修正版训练计划,"
- "不要解释校验过程,也不要遗漏原请求的训练目标。必须逐项消除以下违规:\n"
- + "\n".join(f"- {item}" for item in first_plan_validation)
- + "\n\n原始用户请求:\n" + message[:4000]
- + "\n\n第一次计划:\n" + str(artifact.get("content") or answer)[:24000]
- )
- try:
- correction_agent = deps.create_fithealth_agent(
- avoid_youtube_channels=avoided_youtube_channels
- )
- corrected_answer = format_response(
- await run_in_threadpool(correction_agent.run, correction_prompt)
- )
- except Exception: # noqa: BLE001 - 自动修正失败应降级为明确的校验失败,而非整次 500
- logger.exception("训练计划自动修正调用失败")
- corrected_answer = ""
- corrected_artifact = dict(artifact)
- corrected_artifact["content"] = corrected_answer
- corrected_artifact["source"] = "agent_auto_corrected"
- if not corrected_answer:
- plan_validation = first_plan_validation + ["自动修正服务调用失败"]
- elif not looks_like_complete_training_plan(corrected_answer):
- plan_validation = ["自动修正结果不是完整训练计划"]
- else:
- plan_validation = validate_generated_training_plan(
- corrected_answer,
- enforceable_memories,
- expected_plan_subject,
- recovery=recovery_snapshot,
- safety_constraints=list(resolved_plan_context.get("active_safety_constraints") or []),
- explicitly_requested=set((resolved_plan_context.get("muscle_recovery") or {}).get("explicitly_requested") or []),
- resolver=_plan_muscle_resolver,
- check_subject=False,
- )
- corrected_alignment = await run_in_threadpool(
- deps.validate_plan_goal_alignment,
- corrected_answer,
- user_request=message,
- requested_subjects=(getattr(chat_intent, "requested_subjects", []) if chat_intent is not None else []),
- fallback_subject=expected_plan_subject,
- weekly_subject=str(resolved_plan_context.get("scheduled_subject") or ""),
- schedule_decision=str(resolved_plan_context.get("decision") or "follow_schedule"),
- allow_external_models=external_models_enabled,
- )
- if not corrected_alignment.get("passed"):
- missing = "、".join(corrected_alignment.get("missing_subjects") or [expected_plan_subject])
- plan_validation.append(f"计划未覆盖用户要求的训练目标:{missing}")
- plan_auto_correction["final_violations"] = list(plan_validation)
- plan_auto_correction["passed"] = not plan_validation
- if plan_validation:
- artifact = None
- answer = (corrected_answer or answer) + "\n\n## 自动修正后仍未通过计划校验\n" + "\n".join(
- f"- {item}" for item in plan_validation
- ) + "\n本次不提供可保存的训练计划,请调整限制或训练目标后重试。"
- else:
- artifact = corrected_artifact
- answer = corrected_answer
- if pending_reply_prefix:
- answer = pending_reply_prefix + "\n\n" + answer
- if artifact is not None and not plan_validation:
- # BUG-05:校验通过后立刻在服务端留一份完整正文;自动修正通过
- # 的计划也必须走同一条服务端草稿链。
- draft = deps.plan_draft_cache.remember(
- content=str(artifact.get("content") or ""),
- subject=str(artifact.get("subject") or ""),
- title=str(artifact.get("title") or ""),
- suggested_date=str(artifact.get("suggested_date") or ""),
- )
- artifact["draft_id"] = draft.id
- if artifact is not None or first_plan_validation:
- try:
- usage_violations = first_plan_validation or plan_validation
- for fact_id, detail in plan_validation_fact_usage(
- enforceable_memories, usage_violations
- ):
- deps.info_store.log_fact_usage(
- [fact_id],
- action="plan_auto_correction" if first_plan_validation else "plan_validation",
- detail=(detail + (";自动修正通过" if first_plan_validation and not plan_validation else ""))[:500],
- )
- except (MemoryStoreDegradedError, OSError):
- logger.exception("记忆使用日志保存失败")
- response_artifact = artifact or profile_update_artifact or pending_view_artifact
- if response_artifact is None and source == "chat" and any(
- word in message for word in ("饮食", "营养", "摄入", "碳水", "蛋白质", "脂肪", "热量")
- ):
- response_artifact = {
- "type": "nutrition_records",
- "default_date": datetime.now(ZoneInfo("Asia/Shanghai")).date().isoformat(),
- }
- response_extra: dict[str, object] = {
- "plan_context": resolved_plan_context,
- "workflow_state": str(
- state_after_generation(artifact=artifact, validation_failed=bool(plan_validation))
- ),
- }
- if profile_update_artifact is not None:
- response_extra["profile_update"] = profile_update_artifact
- if health_risk is not None:
- response_extra["health_risk"] = {
- "level": health_risk.level,
- "signals": list(health_risk.labels),
- }
- if immediate_memory_candidate is not None:
- response_extra["memory_candidate"] = immediate_memory_candidate
- if memory_candidates:
- response_extra["memory_candidates"] = memory_candidates
- if plan_validation:
- response_extra["plan_validation"] = {"passed": False, "violations": plan_validation}
- elif artifact and artifact.get("type") == "training_plan":
- response_extra["plan_validation"] = {"passed": True, "violations": []}
- if plan_auto_correction["attempted"]:
- response_extra["plan_auto_correction"] = plan_auto_correction
- return _chat_response(
- answer,
- artifact=response_artifact,
- source=pending_view_source or "agent",
- **response_extra,
- )
- except HelloAgentsException:
- logger.exception("Agent 模型配置或调用失败")
- return _chat_response(
- "模型服务当前不可用;健康数据查询仍可通过“查看本周数据”使用本地汇总。",
- status_code=503,
- )
- except Exception: # noqa: BLE001
- logger.exception("Agent 对话处理失败")
- return _chat_response("服务暂时不可用,请稍后重试。", status_code=500)
|