feat: 添加记忆维护的错误处理逻辑，优化传输错误的上下文信息

Co-authored-by: Copilot <copilot@github.com>

src/gateway/memory_maintenance.rs

@@ -238,8 +238,18 @@ impl MemoryMaintenanceService {
         let mut results = Vec::new();
 
         for scope_key in scope_keys {
-            let Some(output) = self.run_for_scope(&scope_key).await? else {
+            let output = match self.run_for_scope(&scope_key).await {
+                Ok(Some(output)) => output,
+                Ok(None) => continue,
+                Err(error) if is_recoverable_maintenance_scope_error(&error) => {
+                    tracing::warn!(
+                        scope_key = %scope_key,
+                        error = %error,
+                        "Memory maintenance skipped scope after recoverable model failure"
+                    );
                     continue;
+                }
+                Err(error) => return Err(error),
             };
 
             results.push(MemoryMaintenanceScopeResult { scope_key, output });
@@ -319,6 +329,10 @@ pub(crate) fn is_recoverable_maintenance_llm_error(error: &str) -> bool {
         || normalized.contains("timeout")
 }
 
+fn is_recoverable_maintenance_scope_error(error: &AgentError) -> bool {
+    is_recoverable_maintenance_llm_error(&error.to_string())
+}
+
 pub(crate) fn strip_json_code_fence(content: &str) -> &str {
     let trimmed = content.trim();
     if let Some(rest) = trimmed.strip_prefix("```json") {

src/gateway/session.rs

@@ -1059,6 +1059,67 @@ mod tests {
         );
     }
 
+    #[tokio::test]
+    async fn test_summarize_memory_maintenance_transport_error_includes_provider_context() {
+        let provider_config = LLMProviderConfig {
+            provider_type: "openai".to_string(),
+            name: "maintenance-provider".to_string(),
+            base_url: "https://example.invalid/v1".to_string(),
+            api_key: "test-key".to_string(),
+            extra_headers: HashMap::new(),
+            model_id: "maintenance-model".to_string(),
+            temperature: Some(0.0),
+            max_tokens: Some(256),
+            context_window_tokens: None,
+            model_extra: HashMap::new(),
+            max_tool_iterations: 1,
+            llm_timeout_secs: 1,
+            tool_result_max_chars: 20_000,
+            context_tool_result_trim_chars: 20_000,
+        };
+
+        let session_manager = SessionManager::new(
+            4,
+            100,
+            false,
+            "Asia/Shanghai".to_string(),
+            provider_config.clone(),
+            HashMap::from([("default".to_string(), provider_config)]),
+            Arc::new(SkillRuntime::default()),
+        )
+        .unwrap();
+
+        session_manager
+            .store()
+            .put_memory(&crate::storage::MemoryUpsert {
+                scope_kind: "user".to_string(),
+                scope_key: "feishu:user-1".to_string(),
+                namespace: "profile".to_string(),
+                memory_key: "work".to_string(),
+                content: "用户在做AI产品".to_string(),
+                source_type: "message".to_string(),
+                source_session_id: None,
+                source_message_id: None,
+                source_message_seq: None,
+                source_channel_name: None,
+                source_chat_id: None,
+            })
+            .unwrap();
+
+        let error = session_manager
+            .summarize_memory_maintenance_for_scope("feishu:user-1")
+            .await
+            .unwrap_err()
+            .to_string();
+
+        assert!(error.contains("memory maintenance model error: transport error:"));
+        assert!(error.contains("provider=maintenance-provider"));
+        assert!(error.contains("model=maintenance-model"));
+        assert!(error.contains("url=https://example.invalid/v1/chat/completions"));
+        assert!(error.contains("timeout_secs=1"));
+        assert!(error.contains("error sending request for url"));
+    }
+
     #[tokio::test]
     async fn test_summarize_memory_maintenance_retries_recoverable_provider_errors() {
         let mock_response_content = serde_json::to_string(&json!({
@@ -1249,6 +1310,63 @@ mod tests {
         assert!(results.is_empty());
     }
 
+    #[tokio::test]
+    async fn test_run_memory_maintenance_for_all_scopes_skips_recoverable_transport_failures() {
+        let provider_config = LLMProviderConfig {
+            provider_type: "openai".to_string(),
+            name: "maintenance-provider".to_string(),
+            base_url: "https://example.invalid/v1".to_string(),
+            api_key: "test-key".to_string(),
+            extra_headers: HashMap::new(),
+            model_id: "maintenance-model".to_string(),
+            temperature: Some(0.0),
+            max_tokens: Some(256),
+            context_window_tokens: None,
+            model_extra: HashMap::new(),
+            max_tool_iterations: 1,
+            llm_timeout_secs: 1,
+            tool_result_max_chars: 20_000,
+            context_tool_result_trim_chars: 20_000,
+        };
+
+        let session_manager = SessionManager::new(
+            4,
+            100,
+            false,
+            "Asia/Shanghai".to_string(),
+            provider_config.clone(),
+            HashMap::from([("default".to_string(), provider_config)]),
+            Arc::new(SkillRuntime::default()),
+        )
+        .unwrap();
+
+        for scope_key in ["feishu:user-1", "feishu:user-2"] {
+            session_manager
+                .store()
+                .put_memory(&crate::storage::MemoryUpsert {
+                    scope_kind: "user".to_string(),
+                    scope_key: scope_key.to_string(),
+                    namespace: "profile".to_string(),
+                    memory_key: "work".to_string(),
+                    content: format!("{} 在做AI产品", scope_key),
+                    source_type: "message".to_string(),
+                    source_session_id: None,
+                    source_message_id: None,
+                    source_message_seq: None,
+                    source_channel_name: None,
+                    source_chat_id: None,
+                })
+                .unwrap();
+        }
+
+        let results = session_manager
+            .run_memory_maintenance_for_all_scopes(None)
+            .await
+            .unwrap();
+
+        assert!(results.is_empty());
+    }
+
     #[test]
     fn test_apply_memory_maintenance_output_merges_and_deletes_low_value_records() {
         let store = SessionStore::in_memory().unwrap();

src/providers/openai.rs

@@ -23,6 +23,23 @@ fn format_error_chain(error: &(dyn std::error::Error + 'static)) -> String {
     details.join("\ncaused by: ")
 }
 
+fn format_transport_error_context(
+    provider_name: &str,
+    model_id: &str,
+    url: &str,
+    timeout_secs: u64,
+    error: &(dyn std::error::Error + 'static),
+) -> String {
+    format!(
+        "transport error: provider={} model={} url={} timeout_secs={} details={}",
+        provider_name,
+        model_id,
+        url,
+        timeout_secs,
+        format_error_chain(error)
+    )
+}
+
 fn convert_content_blocks(blocks: &[ContentBlock]) -> Value {
     if blocks.len() == 1 {
         if let ContentBlock::Text { text } = &blocks[0] {
@@ -294,7 +311,25 @@ impl LLMProvider for OpenAIProvider {
             req_builder = req_builder.header(key.as_str(), value.as_str());
         }
 
-        let resp = req_builder.json(&body).send().await?;
+        let resp = req_builder.json(&body).send().await.map_err(|err| {
+            let error_context = format_transport_error_context(
+                &self.name,
+                &self.model_id,
+                &url,
+                self.llm_timeout_secs,
+                &err,
+            );
+            tracing::error!(
+                provider = %self.name,
+                model = %self.model_id,
+                url = %url,
+                base_url = %self.base_url,
+                timeout_secs = self.llm_timeout_secs,
+                error = %error_context,
+                "OpenAI-compatible API transport request failed"
+            );
+            error_context
+        })?;
 
         let status = resp.status();
         let text = resp.text().await?;