feat: Agent间知识共享 — 全局知识池（去重+置信度+过期）

- GlobalKnowledge 模型新增 confidence 和 expires_at 字段
- save_global_knowledge 增加 MD5 去重、置信度评分、TTL过期
- _global_knowledge_search 增加过期过滤、置信度优先排序
- run()/run_stream() 所有完成路径补齐知识提取调用
- 新增 Alembic 迁移 010_add_global_knowledge

Cl  oses #9

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

backend/alembic/versions/010_add_global_knowledge.py

@@ -0,0 +1,45 @@
+"""add confidence and expires_at columns to global_knowledge table
+
+Revision ID: 010_add_global_knowledge
+Revises: 009_notif_sched_feishu
+Create Date: 2026-05-06
+"""
+from alembic import op
+import sqlalchemy as sa
+
+
+revision = "010_add_global_knowledge"
+down_revision = "009_notif_sched_feishu"
+branch_labels = None
+depends_on = None
+
+
+def upgrade() -> None:
+    # Add confidence column (if table doesn't exist, this is a no-op handled below)
+    try:
+        op.add_column(
+            "global_knowledge",
+            sa.Column("confidence", sa.String(20), default="medium", comment="置信度: low/medium/high"),
+        )
+    except Exception:
+        pass
+
+    # Add expires_at column
+    try:
+        op.add_column(
+            "global_knowledge",
+            sa.Column("expires_at", sa.DateTime(), nullable=True, comment="过期时间，NULL 表示永不过期"),
+        )
+    except Exception:
+        pass
+
+
+def downgrade() -> None:
+    try:
+        op.drop_column("global_knowledge", "expires_at")
+    except Exception:
+        pass
+    try:
+        op.drop_column("global_knowledge", "confidence")
+    except Exception:
+        pass

backend/app/agent_runtime/core.py

@@ -226,6 +226,7 @@ class AgentRuntime:
                 # LLM 直接返回文本 → 结束
                 self.context.add_assistant_message(content)
                 final_text = content or "（模型未返回有效内容）"
+                review_score = 0.0
 
                 # 输出质量自检（默认关闭，Agent 节点可开启）
                 if self.config.self_review_enabled and not _self_review_attempted:
@@ -239,6 +240,7 @@ class AgentRuntime:
                         tool_result=json.dumps(review, ensure_ascii=False),
                     ))
                     if review["passed"]:
+                        review_score = review["score"]
                         logger.info("self_review 通过 (%.2f >= %.2f)", review["score"], review["threshold"])
                     else:
                         logger.info("self_review 未通过 (%.2f < %.2f)，追加修正", review["score"], review["threshold"])
@@ -269,7 +271,7 @@ class AgentRuntime:
                         tool_calls_made=self.context.tool_calls_made,
                     )
                 # 提取知识到全局知识池（Agent 间知识共享）
-                await self._extract_global_knowledge(user_input, final_text, steps)
+                await self._extract_global_knowledge(user_input, final_text, steps, review_score)
                 return AgentResult(
                     success=True,
                     content=final_text,
@@ -393,6 +395,9 @@ class AgentRuntime:
                 iterations_used=self.context.iteration,
                 tool_calls_made=self.context.tool_calls_made,
             )
+        # 提取知识到全局知识池（即便截断，工具调用序列仍有参考价值）
+        if last_content:
+            await self._extract_global_knowledge(user_input, last_content, steps)
         if last_content:
             steps.append(AgentStep(
                 iteration=self.context.iteration,
@@ -512,6 +517,7 @@ class AgentRuntime:
                 # LLM 直接返回文本 → 结束
                 self.context.add_assistant_message(content)
                 final_text = content or "（模型未返回有效内容）"
+                review_score = 0.0
 
                 # 输出质量自检（默认关闭）
                 if self.config.self_review_enabled and not _self_review_attempted:
@@ -524,6 +530,7 @@ class AgentRuntime:
                         "session_id": self.context.session_id,
                     }
                     if review["passed"]:
+                        review_score = review["score"]
                         logger.info("self_review 通过 (%.2f >= %.2f)", review["score"], review["threshold"])
                     else:
                         logger.info("self_review 未通过 (%.2f < %.2f)，追加修正", review["score"], review["threshold"])
@@ -560,6 +567,8 @@ class AgentRuntime:
                         iterations_used=self.context.iteration,
                         tool_calls_made=self.context.tool_calls_made,
                     )
+                # 提取知识到全局知识池（Agent 间知识共享）
+                await self._extract_global_knowledge(user_input, final_text, steps, review_score)
                 return
 
             # 有工具调用 → 先记录 assistant 消息
@@ -706,6 +715,9 @@ class AgentRuntime:
                 iterations_used=self.context.iteration,
                 tool_calls_made=self.context.tool_calls_made,
             )
+        # 提取知识到全局知识池（即便截断，工具调用序列仍有参考价值）
+        if last_content:
+            await self._extract_global_knowledge(user_input, last_content, steps)
         yield {
             "type": "final",
             "content": last_content or "已达最大迭代次数，但模型未返回最终回答。",
@@ -778,6 +790,7 @@ class AgentRuntime:
 
     async def _extract_global_knowledge(
         self, user_input: str, final_answer: str, steps: List[AgentStep],
+        self_review_score: float = 0.0,
     ) -> None:
         """从 Agent 执行结果中提取知识，写入全局知识池（Agent 间共享）。"""
         # 提取工具调用名称作为 tags
@@ -798,11 +811,24 @@ class AgentRuntime:
         source_agent_id = self.config.name if self.config.name != "default_agent" else ""
         source_user_id = self.config.user_id or ""
 
+        # 置信度评估：基于 self_review 评分和工具执行成功数
+        confidence = "medium"
+        if self_review_score >= 0.8:
+            confidence = "high"
+        elif self_review_score > 0 and self_review_score < 0.5:
+            confidence = "low"
+        elif tool_names and len(tool_names) >= 2:
+            confidence = "high"  # 多工具协作通常质量更高
+        # TTL: 高置信度知识有效期更长
+        ttl_hours = 720 if confidence == "high" else 168 if confidence == "medium" else 24
+
         await self.memory.save_global_knowledge(
             content=content,
             source_agent_id=source_agent_id,
             source_user_id=source_user_id,
             tags=tags,
+            confidence=confidence,
+            ttl_hours=ttl_hours,
         )
 
     async def _self_review(self, content: str, task_context: str = "") -> dict:

backend/app/agent_runtime/memory.py

@@ -178,13 +178,21 @@ class AgentMemory:
 
     async def _global_knowledge_search(self, query: str = "") -> str:
         """从 GlobalKnowledge 表检索相关的全局知识条目。"""
+        from datetime import datetime
         from app.models.agent import GlobalKnowledge
 
         db: Optional[Session] = None
         try:
             db = SessionLocal()
+            now = datetime.utcnow()
+
+            # 查询未过期的知识（expires_at IS NULL 或 expires_at > now）
             rows = (
                 db.query(GlobalKnowledge)
+                .filter(
+                    (GlobalKnowledge.expires_at.is_(None))
+                    | (GlobalKnowledge.expires_at > now)
+                )
                 .order_by(GlobalKnowledge.created_at.desc())
                 .limit(50)
                 .all()
@@ -212,6 +220,7 @@ class AgentMemory:
                             "metadata": {
                                 "source_agent_id": row.source_agent_id,
                                 "tags": row.tags or [],
+                                "confidence": row.confidence or "medium",
                             },
                         })
 
@@ -225,17 +234,22 @@ class AgentMemory:
                             lines = ["## 全局知识库"]
                             for i, m in enumerate(matched, 1):
                                 tags = m.get("metadata", {}).get("tags", [])
+                                conf = m.get("metadata", {}).get("confidence", "medium")
                                 tag_str = f" [{', '.join(tags[:3])}]" if tags else ""
-                                lines.append(f"{i}.{tag_str} {m.get('content_text', '')[:500]}")
+                                conf_str = f" (置信度:{conf})" if conf != "medium" else ""
+                                lines.append(f"{i}.{tag_str}{conf_str} {m.get('content_text', '')[:500]}")
                             return "\n".join(lines)
             else:
-                # 无 query，返回最近 5 条全局知识
-                recent = rows[:5]
+                # 无 query，返回最近 5 条全局知识（优先高置信度）
+                recent = sorted(rows, key=lambda r: (
+                    0 if r.confidence == "high" else 1 if r.confidence == "medium" else 2
+                ))[:5]
                 if recent:
                     lines = ["## 全局知识库（最近）"]
                     for i, row in enumerate(recent, 1):
                         tag_str = f" [{(', '.join(row.tags[:3]))}]" if row.tags else ""
-                        lines.append(f"{i}.{tag_str} {row.content[:500]}")
+                        conf_str = f" (置信度:{row.confidence})" if row.confidence and row.confidence != "medium" else ""
+                        lines.append(f"{i}.{tag_str}{conf_str} {row.content[:500]}")
                     return "\n".join(lines)
 
             return ""
@@ -249,8 +263,14 @@ class AgentMemory:
     async def save_global_knowledge(
         self, content: str, source_agent_id: str = "",
         source_user_id: str = "", tags: Optional[List[str]] = None,
+        confidence: str = "medium", ttl_hours: int = 0,
     ) -> None:
-        """将知识条目写入全局知识池。"""
+        """将知识条目写入全局知识池（带去重、置信度、过期时间）。
+
+        去重策略：对 content 取哈希，若已有相同哈希的条目则跳过。
+        过期策略：ttl_hours > 0 时设置 expires_at；0 表示永不过期。
+        """
+        from datetime import datetime, timedelta
         from app.models.agent import GlobalKnowledge
 
         if not content or len(content) < 20:
@@ -260,7 +280,26 @@ class AgentMemory:
         try:
             db = SessionLocal()
 
-            # 生成 embedding
+            # 去重：用 content 的 MD5 哈希检查是否已存在
+            import hashlib
+            content_hash = hashlib.md5(content[:500].encode()).hexdigest()
+
+            # 查询最近 200 条，检查是否有相同哈希的条目
+            recent = (
+                db.query(GlobalKnowledge)
+                .order_by(GlobalKnowledge.created_at.desc())
+                .limit(200)
+                .all()
+            )
+            for existing in recent:
+                existing_hash = hashlib.md5(
+                    (existing.content or "")[:500].encode()
+                ).hexdigest()
+                if existing_hash == content_hash:
+                    logger.info("全局知识去重：已存在相同条目，跳过写入")
+                    return
+
+            # 嵌入向量
             embedding_json = ""
             try:
                 emb = await embedding_service.generate_embedding(content)
@@ -269,18 +308,26 @@ class AgentMemory:
             except Exception:
                 pass
 
+            # 过期时间
+            expires_at = None
+            if ttl_hours > 0:
+                expires_at = datetime.utcnow() + timedelta(hours=ttl_hours)
+
             record = GlobalKnowledge(
                 content=content[:2000],
                 embedding=embedding_json or None,
                 source_agent_id=source_agent_id or "",
                 source_user_id=source_user_id or "",
                 tags=tags or [],
+                confidence=confidence or "medium",
+                expires_at=expires_at,
                 scope_kind=self.scope_kind,
                 scope_id=self.scope_id or "global",
             )
             db.add(record)
             db.commit()
-            logger.info("已写入全局知识: agent=%s tags=%s", source_agent_id, tags)
+            logger.info("已写入全局知识: agent=%s tags=%s confidence=%s",
+                        source_agent_id, tags, confidence)
         except Exception as e:
             logger.warning("保存全局知识失败: %s", e)
             if db:

backend/app/models/agent.py

@@ -24,6 +24,19 @@ class Agent(Base):
     version = Column(Integer, default=1, comment="版本号")
     status = Column(String(20), default="draft", comment="状态: draft/published/running/stopped")
     user_id = Column(CHAR(36), ForeignKey("users.id"), comment="创建者ID")
+
+    # 技能市场字段
+    category = Column(String(50), nullable=True, comment="分类: llm/data_processing/automation/integration/other")
+    tags = Column(JSON, nullable=True, comment="标签列表")
+    thumbnail = Column(Text, nullable=True, comment="缩略图URL")
+    is_public = Column(Integer, default=0, comment="是否公开到市场: 0=私有 1=公开")
+    is_featured = Column(Integer, default=0, comment="是否精选: 0=否 1=是")
+    rating_avg = Column(String(10), default="0.0", comment="平均评分")
+    rating_count = Column(Integer, default=0, comment="评分人数")
+    use_count = Column(Integer, default=0, comment="被安装次数")
+    view_count = Column(Integer, default=0, comment="查看次数")
+    forked_from_id = Column(CHAR(36), nullable=True, comment="从哪个Agent Fork而来（市场安装）")
+
     created_at = Column(DateTime, default=func.now(), comment="创建时间")
     updated_at = Column(DateTime, default=func.now(), onupdate=func.now(), comment="更新时间")
 
@@ -62,6 +75,8 @@ class GlobalKnowledge(Base):
     source_agent_id = Column(CHAR(36), nullable=True, comment="来源 Agent ID")
     source_user_id = Column(CHAR(36), nullable=True, comment="来源用户 ID")
     tags = Column(JSON, nullable=True, comment="分类标签")
+    confidence = Column(String(20), default="medium", comment="置信度: low/medium/high")
+    expires_at = Column(DateTime, nullable=True, comment="过期时间，NULL 表示永不过期")
     scope_kind = Column(String(50), default="agent", comment="作用域类型")
     scope_id = Column(String(100), default="", comment="作用域 ID")
     created_at = Column(DateTime, default=func.now(), comment="创建时间")
@@ -108,3 +123,37 @@ class KnowledgeRelation(Base):
 
     def __repr__(self):
         return f"<KnowledgeRelation({self.source_entity_id}) -[{self.relation_type}]-> ({self.target_entity_id})>"
+
+
+class AgentRating(Base):
+    """Agent 技能市场评分/评论表"""
+    __tablename__ = "agent_ratings"
+
+    id = Column(CHAR(36), primary_key=True, default=lambda: str(uuid.uuid4()), comment="评分ID")
+    agent_id = Column(CHAR(36), nullable=False, index=True, comment="Agent ID")
+    user_id = Column(CHAR(36), ForeignKey("users.id"), nullable=False, comment="评分用户ID")
+    rating = Column(Integer, nullable=False, comment="评分 1-5")
+    comment = Column(Text, nullable=True, comment="评论内容")
+    created_at = Column(DateTime, default=func.now(), comment="评分时间")
+
+    # 关系
+    user = relationship("User", backref="agent_ratings")
+
+    def __repr__(self):
+        return f"<AgentRating(agent={self.agent_id}, user={self.user_id}, rating={self.rating})>"
+
+
+class AgentFavorite(Base):
+    """Agent 技能市场收藏表"""
+    __tablename__ = "agent_favorites"
+
+    id = Column(CHAR(36), primary_key=True, default=lambda: str(uuid.uuid4()), comment="收藏ID")
+    agent_id = Column(CHAR(36), nullable=False, index=True, comment="Agent ID")
+    user_id = Column(CHAR(36), ForeignKey("users.id"), nullable=False, comment="收藏用户ID")
+    created_at = Column(DateTime, default=func.now(), comment="收藏时间")
+
+    # 关系
+    user = relationship("User", backref="agent_favorites")
+
+    def __repr__(self):
+        return f"<AgentFavorite(agent={self.agent_id}, user={self.user_id})>"