Update config.yaml

fix(mcp-stdio): initialize result storage so query tools work

README.md

@@ -211,7 +211,7 @@ go build -o cyberstrike-ai cmd/server/main.go
 
 **CyberStrikeAI one-click upgrade (recommended):**
 1. (First time) enable the script: `chmod +x upgrade.sh`
-2. Upgrade with: `./upgrade.sh` (optional flags: `--tag vX.Y.Z`, `--no-venv`, `--preserve-custom`, `--yes`)
+2. Upgrade with: `./upgrade.sh` (optional flags: `--tag vX.Y.Z`, `--no-venv`, `--yes`). Local `tools/`, `roles/`, and `skills/` are always preserved.
 3. The script will back up your `config.yaml` and `data/`, upgrade the code from GitHub Release, update `config.yaml`'s `version`, then restart the server.
 
 Recommended one-liner:

README_CN.md

@@ -209,7 +209,7 @@ go build -o cyberstrike-ai cmd/server/main.go
 ### CyberStrikeAI 版本更新（无兼容性问题）
 
 1. （首次使用）启用脚本：`chmod +x upgrade.sh`
-2. 一键升级：`./upgrade.sh`（可选参数：`--tag vX.Y.Z`、`--no-venv`、`--preserve-custom`、`--yes`）
+2. 一键升级：`./upgrade.sh`（可选参数：`--tag vX.Y.Z`、`--no-venv`、`--yes`）。本地的 `tools/`、`roles/`、`skills/` 会始终保留不被覆盖。
 3. 脚本会备份你的 `config.yaml` 和 `data/`，从 GitHub Release 升级代码，更新 `config.yaml` 的 `version` 字段后重启服务。
 
 推荐的一键指令：

cmd/mcp-stdio/main.go

@@ -5,6 +5,7 @@ import (
 	"cyberstrike-ai/internal/logger"
 	"cyberstrike-ai/internal/mcp"
 	"cyberstrike-ai/internal/security"
+	"cyberstrike-ai/internal/storage"
 	"flag"
 	"fmt"
 	"os"
@@ -32,6 +33,23 @@ func main() {
 	// 创建安全工具执行器
 	executor := security.NewExecutor(&cfg.Security, mcpServer, log.Logger)
 
+	// 初始化结果存储（与 internal/app/app.go 同样的逻辑）。
+	// stdio 模式下原本不初始化，导致 'exec' 等查询型工具报"结果存储未初始化"。
+	resultStorageDir := "tmp"
+	if cfg.Agent.ResultStorageDir != "" {
+		resultStorageDir = cfg.Agent.ResultStorageDir
+	}
+	if err := os.MkdirAll(resultStorageDir, 0755); err != nil {
+		fmt.Fprintf(os.Stderr, "创建结果存储目录失败: %v\n", err)
+		os.Exit(1)
+	}
+	resultStorage, err := storage.NewFileResultStorage(resultStorageDir, log.Logger)
+	if err != nil {
+		fmt.Fprintf(os.Stderr, "初始化结果存储失败: %v\n", err)
+		os.Exit(1)
+	}
+	executor.SetResultStorage(resultStorage)
+
 	// 注册工具
 	executor.RegisterTools(mcpServer)
 

config.yaml

@@ -10,7 +10,7 @@
 # ============================================
 
 # 前端显示的版本号（可选，不填则显示默认版本）
-version: "v1.6.7"
+version: "v1.6.10"
 # 服务器配置
 server:
   host: 0.0.0.0 # 监听地址，0.0.0.0 表示监听所有网络接口
@@ -41,6 +41,13 @@ openai:
   api_key: sk-xxxxxxx # API 密钥（必填）
   model: qwen3-max # 模型名称（必填）
   max_total_tokens: 120000 # LLM 相关上下文的最大 Token 数限制（内存压缩和攻击链构建会共用此配置）
+  # Eino 路径模型推理：DeepSeek/OpenAI 为 thinking / reasoning_effort 等；provider 为 claude 时合并为 Anthropic 顶层 thinking（extended thinking），mode: off 关闭
+  reasoning:
+    mode: off # auto | on | off；off 时不附加任何推理扩展字段
+    effort: max # low | medium | high | max；空表示不指定（openai_compat 下 auto 且无强度时不发请求扩展）
+    allow_client_reasoning: true # false 时忽略对话请求体 reasoning，仅以下方为准
+    profile: openai_compat # auto | deepseek_compat | openai_compat | output_config_effort
+    # extra_request_fields: {} # 可选：管理员自定义根级 JSON 片段（高级）
 # ============================================
 # 信息收集（FOFA）配置（可选）
 # ============================================
@@ -110,6 +117,21 @@ multi_agent:
     deep_output_key: "" # 非空：将最终助手输出写入 adk session 的键名（Deep 与 Supervisor 主代理）；空表示不写入
     deep_model_retry_max_retries: 0 # >0：ChatModel 调用失败时的框架级最大重试次数（Deep 与 Supervisor 主）；0：不重试
     task_tool_description_prefix: "" # 非空：仅 Deep 的 task 工具使用自定义描述前缀，运行时会拼接子代理名称；空则走 Eino 默认生成逻辑
+  # Eino callbacks + OpenTelemetry：框架级 span（与 Zap 对齐）；默认不向终端用户 UI 推 eino_trace_*（见 sse_trace_to_client）
+  eino_callbacks:
+    enabled: true
+    # log_only=仅 Zap+OTel（推荐默认）| sse/full=才启用流式回调副本关闭等（full 含 stream hooks）
+    mode: log_only
+    sse_trace_to_client: false # true：且 mode 为 sse/full 时，向前端时间线推送 eino_trace_*（排障/内网演示用）
+    max_input_summary_runes: 400
+    max_output_summary_runes: 400
+    zap_verbose: false # true：Debug 附带 input/output 摘要
+    otel:
+      enabled: true
+      service_name: cyberstrike-ai
+      exporter: stdout # none | stdout（开发/本机）| otlphttp（生产接 Collector）
+      otlp_endpoint: localhost:4318 # otlphttp 时使用，host:port，路径固定 /v1/traces
+      sample_ratio: 1.0 # 0~1，ParentBased+TraceIDRatio
 # 数据库配置
 database:
   path: data/conversations.db # SQLite 数据库文件路径，用于存储对话历史和消息

go.mod

@@ -27,6 +27,11 @@ require (
 	github.com/open-dingtalk/dingtalk-stream-sdk-go v0.9.1
 	github.com/pkoukk/tiktoken-go v0.1.8
 	github.com/robfig/cron/v3 v3.0.1
+	go.opentelemetry.io/otel v1.34.0
+	go.opentelemetry.io/otel/exporters/otlp/otlptrace/otlptracehttp v1.34.0
+	go.opentelemetry.io/otel/exporters/stdout/stdouttrace v1.34.0
+	go.opentelemetry.io/otel/sdk v1.34.0
+	go.opentelemetry.io/otel/trace v1.34.0
 	go.uber.org/zap v1.26.0
 	golang.org/x/text v0.26.0
 	golang.org/x/time v0.14.0
@@ -39,6 +44,7 @@ require (
 	github.com/buger/jsonparser v1.1.1 // indirect
 	github.com/bytedance/gopkg v0.1.3 // indirect
 	github.com/bytedance/sonic/loader v0.5.0 // indirect
+	github.com/cenkalti/backoff/v4 v4.3.0 // indirect
 	github.com/cloudwego/base64x v0.1.6 // indirect
 	github.com/cloudwego/eino-ext/libs/acl/openai v0.1.17 // indirect
 	github.com/dlclark/regexp2 v1.10.0 // indirect
@@ -46,6 +52,8 @@ require (
 	github.com/evanphx/json-patch v0.5.2 // indirect
 	github.com/gabriel-vasile/mimetype v1.4.2 // indirect
 	github.com/gin-contrib/sse v0.1.0 // indirect
+	github.com/go-logr/logr v1.4.2 // indirect
+	github.com/go-logr/stdr v1.2.2 // indirect
 	github.com/go-playground/locales v0.14.1 // indirect
 	github.com/go-playground/universal-translator v0.18.1 // indirect
 	github.com/go-playground/validator/v10 v10.14.0 // indirect
@@ -53,6 +61,7 @@ require (
 	github.com/gogo/protobuf v1.3.2 // indirect
 	github.com/google/jsonschema-go v0.3.0 // indirect
 	github.com/goph/emperror v0.17.2 // indirect
+	github.com/grpc-ecosystem/grpc-gateway/v2 v2.25.1 // indirect
 	github.com/json-iterator/go v1.1.12 // indirect
 	github.com/klauspost/cpuid/v2 v2.2.10 // indirect
 	github.com/leodido/go-urn v1.2.4 // indirect
@@ -71,14 +80,21 @@ require (
 	github.com/wk8/go-ordered-map/v2 v2.1.8 // indirect
 	github.com/yargevad/filepathx v1.0.0 // indirect
 	github.com/yosida95/uritemplate/v3 v3.0.2 // indirect
+	go.opentelemetry.io/auto/sdk v1.1.0 // indirect
+	go.opentelemetry.io/otel/exporters/otlp/otlptrace v1.34.0 // indirect
+	go.opentelemetry.io/otel/metric v1.34.0 // indirect
+	go.opentelemetry.io/proto/otlp v1.5.0 // indirect
 	go.uber.org/multierr v1.11.0 // indirect
 	golang.org/x/arch v0.15.0 // indirect
 	golang.org/x/crypto v0.39.0 // indirect
 	golang.org/x/exp v0.0.0-20250305212735-054e65f0b394 // indirect
-	golang.org/x/net v0.24.0 // indirect
+	golang.org/x/net v0.34.0 // indirect
 	golang.org/x/oauth2 v0.30.0 // indirect
 	golang.org/x/sys v0.33.0 // indirect
-	google.golang.org/protobuf v1.30.0 // indirect
+	google.golang.org/genproto/googleapis/api v0.0.0-20250115164207-1a7da9e5054f // indirect
+	google.golang.org/genproto/googleapis/rpc v0.0.0-20250115164207-1a7da9e5054f // indirect
+	google.golang.org/grpc v1.69.4 // indirect
+	google.golang.org/protobuf v1.36.3 // indirect
 )
 
 // 修复钉钉 Stream SDK 在长连接断开（熄屏/网络中断）后 "panic: send on closed channel" 问题

go.sum

@@ -17,6 +17,8 @@ github.com/bytedance/sonic v1.15.0 h1:/PXeWFaR5ElNcVE84U0dOHjiMHQOwNIx3K4ymzh/uS
 github.com/bytedance/sonic v1.15.0/go.mod h1:tFkWrPz0/CUCLEF4ri4UkHekCIcdnkqXw9VduqpJh0k=
 github.com/bytedance/sonic/loader v0.5.0 h1:gXH3KVnatgY7loH5/TkeVyXPfESoqSBSBEiDd5VjlgE=
 github.com/bytedance/sonic/loader v0.5.0/go.mod h1:AR4NYCk5DdzZizZ5djGqQ92eEhCCcdf5x77udYiSJRo=
+github.com/cenkalti/backoff/v4 v4.3.0 h1:MyRJ/UdXutAwSAT+s3wNd7MfTIcy71VQueUuFK343L8=
+github.com/cenkalti/backoff/v4 v4.3.0/go.mod h1:Y3VNntkOUPxTVeUxJ/G5vcM//AlwfmyYozVcomhLiZE=
 github.com/certifi/gocertifi v0.0.0-20190105021004-abcd57078448/go.mod h1:GJKEexRPVJrBSOjoqN5VNOIKJ5Q3RViH6eu3puDRwx4=
 github.com/cloudwego/base64x v0.1.6 h1:t11wG9AECkCDk5fMSoxmufanudBtJ+/HemLstXDLI2M=
 github.com/cloudwego/base64x v0.1.6/go.mod h1:OFcloc187FXDaYHvrNIjxSe8ncn0OOM8gEHfghB2IPU=
@@ -59,6 +61,11 @@ github.com/gin-gonic/gin v1.9.1 h1:4idEAncQnU5cB7BeOkPtxjfCSye0AAm1R0RVIqJ+Jmg=
 github.com/gin-gonic/gin v1.9.1/go.mod h1:hPrL7YrpYKXt5YId3A/Tnip5kqbEAP+KLuI3SUcPTeU=
 github.com/go-check/check v0.0.0-20180628173108-788fd7840127 h1:0gkP6mzaMqkmpcJYCFOLkIBwI7xFExG03bbkOkCvUPI=
 github.com/go-check/check v0.0.0-20180628173108-788fd7840127/go.mod h1:9ES+weclKsC9YodN5RgxqK/VD9HM9JsCSh7rNhMZE98=
+github.com/go-logr/logr v1.2.2/go.mod h1:jdQByPbusPIv2/zmleS9BjJVeZ6kBagPoEUsqbVz/1A=
+github.com/go-logr/logr v1.4.2 h1:6pFjapn8bFcIbiKo3XT4j/BhANplGihG6tvd+8rYgrY=
+github.com/go-logr/logr v1.4.2/go.mod h1:9T104GzyrTigFIr8wt5mBrctHMim0Nb2HLGrmQ40KvY=
+github.com/go-logr/stdr v1.2.2 h1:hSWxHoqTgW2S2qGc0LTAI563KZ5YKYRhT3MFKZMbjag=
+github.com/go-logr/stdr v1.2.2/go.mod h1:mMo/vtBO5dYbehREoey6XUKy/eSumjCCveDpRre4VKE=
 github.com/go-playground/assert/v2 v2.2.0 h1:JvknZsQTYeFEAhQwI4qEt9cyV5ONwRHC+lYKSsYSR8s=
 github.com/go-playground/assert/v2 v2.2.0/go.mod h1:VDjEfimB/XKnb+ZQfWdccd7VUvScMdVu0Titje2rxJ4=
 github.com/go-playground/locales v0.14.1 h1:EWaQ/wswjilfKLTECiXz7Rh+3BjFhfDFKv/oXslEjJA=
@@ -75,8 +82,8 @@ github.com/gogo/protobuf v1.3.2/go.mod h1:P1XiOD3dCwIKUDQYPy72D8LYyHL2YPYrpS2s69
 github.com/golang-jwt/jwt/v5 v5.2.2 h1:Rl4B7itRWVtYIHFrSNd7vhTiz9UpLdi6gZhZ3wEeDy8=
 github.com/golang-jwt/jwt/v5 v5.2.2/go.mod h1:pqrtFR0X4osieyHYxtmOUWsAWrfe1Q5UVIyoH402zdk=
 github.com/golang/protobuf v1.2.0/go.mod h1:6lQm79b+lXiMfvg/cZm0SGofjICqVBUtrP5yJMmIC1U=
-github.com/golang/protobuf v1.5.0/go.mod h1:FsONVRAS9T7sI+LIUmWTfcYkHO4aIWwzhcaSAoJOfIk=
-github.com/google/go-cmp v0.5.5/go.mod h1:v8dTdLbMG2kIc/vJvl+f65V22dbkXbowE6jgT/gNBxE=
+github.com/golang/protobuf v1.5.4 h1:i7eJL8qZTpSEXOPTxNKhASYpMn+8e5Q6AdndVa1dWek=
+github.com/golang/protobuf v1.5.4/go.mod h1:lnTiLA8Wa4RWRcIUkrtSVa5nRhsEGBg48fD6rSs7xps=
 github.com/google/go-cmp v0.7.0 h1:wk8382ETsv4JYUZwIsn6YpYiWiBsYLSJiTsyBybVuN8=
 github.com/google/go-cmp v0.7.0/go.mod h1:pXiqmnSA92OHEEa9HXL2W4E7lf9JzCmGVUdgjX3N/iU=
 github.com/google/gofuzz v1.0.0/go.mod h1:dBl0BpW6vV/+mYPU4Po3pmUjxk6FQPldtuIdl/M65Eg=
@@ -90,6 +97,8 @@ github.com/gopherjs/gopherjs v1.17.2 h1:fQnZVsXk8uxXIStYb0N4bGk7jeyTalG/wsZjQ25d
 github.com/gopherjs/gopherjs v1.17.2/go.mod h1:pRRIvn/QzFLrKfvEz3qUuEhtE/zLCWfreZ6J5gM2i+k=
 github.com/gorilla/websocket v1.5.0 h1:PPwGk2jz7EePpoHN/+ClbZu8SPxiqlu12wZP/3sWmnc=
 github.com/gorilla/websocket v1.5.0/go.mod h1:YR8l580nyteQvAITg2hZ9XVh4b55+EU/adAjf1fMHhE=
+github.com/grpc-ecosystem/grpc-gateway/v2 v2.25.1 h1:VNqngBF40hVlDloBruUehVYC3ArSgIyScOAyMRqBxRg=
+github.com/grpc-ecosystem/grpc-gateway/v2 v2.25.1/go.mod h1:RBRO7fro65R6tjKzYgLAFo0t1QEXY1Dp+i/bvpRiqiQ=
 github.com/hpcloud/tail v1.0.0/go.mod h1:ab1qPbhIpdTxEkNHXyeSf5vhxWSCs/tWer42PpOxQnU=
 github.com/jessevdk/go-flags v1.4.0/go.mod h1:4FA24M0QyGHXBuZZK/XkWh8h0e1EYbRYJSGM75WSRxI=
 github.com/json-iterator/go v1.1.12 h1:PV8peI4a0ysnczrg+LtxykD8LfKY9ML6u2jnxaEnrnM=
@@ -191,6 +200,26 @@ github.com/yosida95/uritemplate/v3 v3.0.2 h1:Ed3Oyj9yrmi9087+NczuL5BwkIc4wvTb5zI
 github.com/yosida95/uritemplate/v3 v3.0.2/go.mod h1:ILOh0sOhIJR3+L/8afwt/kE++YT040gmv5BQTMR2HP4=
 github.com/yuin/goldmark v1.1.27/go.mod h1:3hX8gzYuyVAZsxl0MRgGTJEmQBFcNTphYh9decYSb74=
 github.com/yuin/goldmark v1.2.1/go.mod h1:3hX8gzYuyVAZsxl0MRgGTJEmQBFcNTphYh9decYSb74=
+go.opentelemetry.io/auto/sdk v1.1.0 h1:cH53jehLUN6UFLY71z+NDOiNJqDdPRaXzTel0sJySYA=
+go.opentelemetry.io/auto/sdk v1.1.0/go.mod h1:3wSPjt5PWp2RhlCcmmOial7AvC4DQqZb7a7wCow3W8A=
+go.opentelemetry.io/otel v1.34.0 h1:zRLXxLCgL1WyKsPVrgbSdMN4c0FMkDAskSTQP+0hdUY=
+go.opentelemetry.io/otel v1.34.0/go.mod h1:OWFPOQ+h4G8xpyjgqo4SxJYdDQ/qmRH+wivy7zzx9oI=
+go.opentelemetry.io/otel/exporters/otlp/otlptrace v1.34.0 h1:OeNbIYk/2C15ckl7glBlOBp5+WlYsOElzTNmiPW/x60=
+go.opentelemetry.io/otel/exporters/otlp/otlptrace v1.34.0/go.mod h1:7Bept48yIeqxP2OZ9/AqIpYS94h2or0aB4FypJTc8ZM=
+go.opentelemetry.io/otel/exporters/otlp/otlptrace/otlptracehttp v1.34.0 h1:BEj3SPM81McUZHYjRS5pEgNgnmzGJ5tRpU5krWnV8Bs=
+go.opentelemetry.io/otel/exporters/otlp/otlptrace/otlptracehttp v1.34.0/go.mod h1:9cKLGBDzI/F3NoHLQGm4ZrYdIHsvGt6ej6hUowxY0J4=
+go.opentelemetry.io/otel/exporters/stdout/stdouttrace v1.34.0 h1:jBpDk4HAUsrnVO1FsfCfCOTEc/MkInJmvfCHYLFiT80=
+go.opentelemetry.io/otel/exporters/stdout/stdouttrace v1.34.0/go.mod h1:H9LUIM1daaeZaz91vZcfeM0fejXPmgCYE8ZhzqfJuiU=
+go.opentelemetry.io/otel/metric v1.34.0 h1:+eTR3U0MyfWjRDhmFMxe2SsW64QrZ84AOhvqS7Y+PoQ=
+go.opentelemetry.io/otel/metric v1.34.0/go.mod h1:CEDrp0fy2D0MvkXE+dPV7cMi8tWZwX3dmaIhwPOaqHE=
+go.opentelemetry.io/otel/sdk v1.34.0 h1:95zS4k/2GOy069d321O8jWgYsW3MzVV+KuSPKp7Wr1A=
+go.opentelemetry.io/otel/sdk v1.34.0/go.mod h1:0e/pNiaMAqaykJGKbi+tSjWfNNHMTxoC9qANsCzbyxU=
+go.opentelemetry.io/otel/sdk/metric v1.31.0 h1:i9hxxLJF/9kkvfHppyLL55aW7iIJz4JjxTeYusH7zMc=
+go.opentelemetry.io/otel/sdk/metric v1.31.0/go.mod h1:CRInTMVvNhUKgSAMbKyTMxqOBC0zgyxzW55lZzX43Y8=
+go.opentelemetry.io/otel/trace v1.34.0 h1:+ouXS2V8Rd4hp4580a8q23bg0azF2nI8cqLYnC8mh/k=
+go.opentelemetry.io/otel/trace v1.34.0/go.mod h1:Svm7lSjQD7kG7KJ/MUHPVXSDGz2OX4h0M2jHBhmSfRE=
+go.opentelemetry.io/proto/otlp v1.5.0 h1:xJvq7gMzB31/d406fB8U5CBdyQGw4P399D1aQWU/3i4=
+go.opentelemetry.io/proto/otlp v1.5.0/go.mod h1:keN8WnHxOy8PG0rQZjJJ5A2ebUoafqWp0eVQ4yIXvJ4=
 go.uber.org/goleak v1.2.0 h1:xqgm/S+aQvhWFTtR0XK3Jvg7z8kGV8P4X14IzwN3Eqk=
 go.uber.org/goleak v1.2.0/go.mod h1:XJYK+MuIchqpmGmUSAzotztawfKvYLUIgg7guXrwVUo=
 go.uber.org/mock v0.4.0 h1:VcM4ZOtdbR4f6VXfiOpwpVJDL6lCReaZ6mw31wqh7KU=
@@ -216,8 +245,8 @@ golang.org/x/net v0.0.0-20190404232315-eb5bcb51f2a3/go.mod h1:t9HGtf8HONx5eT2rtn
 golang.org/x/net v0.0.0-20190620200207-3b0461eec859/go.mod h1:z5CRVTTTmAJ677TzLLGU+0bjPO0LkuOLi4/5GtJWs/s=
 golang.org/x/net v0.0.0-20200226121028-0de0cce0169b/go.mod h1:z5CRVTTTmAJ677TzLLGU+0bjPO0LkuOLi4/5GtJWs/s=
 golang.org/x/net v0.0.0-20201021035429-f5854403a974/go.mod h1:sp8m0HH+o8qH0wwXwYZr8TS3Oi6o0r6Gce1SSxlDquU=
-golang.org/x/net v0.24.0 h1:1PcaxkF854Fu3+lvBIx5SYn9wRlBzzcnHZSiaFFAb0w=
-golang.org/x/net v0.24.0/go.mod h1:2Q7sJY5mzlzWjKtYUEXSlBWCdyaioyXzRB2RtU8KVE8=
+golang.org/x/net v0.34.0 h1:Mb7Mrk043xzHgnRM88suvJFwzVrRfHEHJEl5/71CKw0=
+golang.org/x/net v0.34.0/go.mod h1:di0qlW3YNM5oh6GqDGQr92MyTozJPmybPK4Ev/Gm31k=
 golang.org/x/oauth2 v0.30.0 h1:dnDm7JmhM45NNpd8FDDeLhK6FwqbOf4MLCM9zb1BOHI=
 golang.org/x/oauth2 v0.30.0/go.mod h1:B++QgG3ZKulg6sRPGD/mqlHQs5rB3Ml9erfeDY7xKlU=
 golang.org/x/sync v0.0.0-20180314180146-1d60e4601c6f/go.mod h1:RxMgew5VJxzue5/jJTE5uejpjVlOe/izrB70Jof72aM=
@@ -251,9 +280,14 @@ golang.org/x/xerrors v0.0.0-20190717185122-a985d3407aa7/go.mod h1:I/5z698sn9Ka8T
 golang.org/x/xerrors v0.0.0-20191011141410-1b5146add898/go.mod h1:I/5z698sn9Ka8TeJc9MKroUUfqBBauWjQqLJ2OPfmY0=
 golang.org/x/xerrors v0.0.0-20191204190536-9bdfabe68543/go.mod h1:I/5z698sn9Ka8TeJc9MKroUUfqBBauWjQqLJ2OPfmY0=
 golang.org/x/xerrors v0.0.0-20200804184101-5ec99f83aff1/go.mod h1:I/5z698sn9Ka8TeJc9MKroUUfqBBauWjQqLJ2OPfmY0=
-google.golang.org/protobuf v1.26.0-rc.1/go.mod h1:jlhhOSvTdKEhbULTjvd4ARK9grFBp09yW+WbY/TyQbw=
-google.golang.org/protobuf v1.30.0 h1:kPPoIgf3TsEvrm0PFe15JQ+570QVxYzEvvHqChK+cng=
-google.golang.org/protobuf v1.30.0/go.mod h1:HV8QOd/L58Z+nl8r43ehVNZIU/HEI6OcFqwMG9pJV4I=
+google.golang.org/genproto/googleapis/api v0.0.0-20250115164207-1a7da9e5054f h1:gap6+3Gk41EItBuyi4XX/bp4oqJ3UwuIMl25yGinuAA=
+google.golang.org/genproto/googleapis/api v0.0.0-20250115164207-1a7da9e5054f/go.mod h1:Ic02D47M+zbarjYYUlK57y316f2MoN0gjAwI3f2S95o=
+google.golang.org/genproto/googleapis/rpc v0.0.0-20250115164207-1a7da9e5054f h1:OxYkA3wjPsZyBylwymxSHa7ViiW1Sml4ToBrncvFehI=
+google.golang.org/genproto/googleapis/rpc v0.0.0-20250115164207-1a7da9e5054f/go.mod h1:+2Yz8+CLJbIfL9z73EW45avw8Lmge3xVElCP9zEKi50=
+google.golang.org/grpc v1.69.4 h1:MF5TftSMkd8GLw/m0KM6V8CMOCY6NZ1NQDPGFgbTt4A=
+google.golang.org/grpc v1.69.4/go.mod h1:vyjdE6jLBI76dgpDojsFGNaHlxdjXN9ghpnd2o7JGZ4=
+google.golang.org/protobuf v1.36.3 h1:82DV7MYdb8anAVi3qge1wSnMDrnKK7ebr+I0hHRN1BU=
+google.golang.org/protobuf v1.36.3/go.mod h1:9fA7Ob0pmnwhb644+1+CVWFRbNajQ6iRojtC/QF5bRE=
 gopkg.in/check.v1 v0.0.0-20161208181325-20d25e280405/go.mod h1:Co6ibVJAznAaIkqp8huTwlJQCZ016jof/cbN4VW5Yz0=
 gopkg.in/check.v1 v1.0.0-20201130134442-10cb98267c6c h1:Hei/4ADfdWqJk1ZMxUNpqntNwaWcugrBjAiHlqqRiVk=
 gopkg.in/check.v1 v1.0.0-20201130134442-10cb98267c6c/go.mod h1:JHkPIbrfpd72SG/EVd6muEfDQjcINNoR0C8j2r3qZ4Q=

internal/agent/agent.go

@@ -195,6 +195,8 @@ type ChatMessage struct {
 	ToolCallID string     `json:"tool_call_id,omitempty"`
 	// ToolName 仅 tool 角色：从 Eino/轨迹 JSON 的 name 或 tool_name 恢复，供续跑构造 ToolMessage。
 	ToolName string `json:"tool_name,omitempty"`
+	// ReasoningContent 对应 OpenAI/DeepSeek 的 reasoning_content；思考模式 + 工具调用后续跑须回传（见 DeepSeek 文档）。
+	ReasoningContent string `json:"reasoning_content,omitempty"`
 }
 
 // MarshalJSON 自定义JSON序列化，将tool_calls中的arguments转换为JSON字符串
@@ -208,6 +210,9 @@ func (cm ChatMessage) MarshalJSON() ([]byte, error) {
 	if cm.Content != "" {
 		aux["content"] = cm.Content
 	}
+	if cm.ReasoningContent != "" {
+		aux["reasoning_content"] = cm.ReasoningContent
+	}
 
 	// 添加tool_call_id（如果存在）
 	if cm.ToolCallID != "" {
@@ -663,8 +668,8 @@ func (a *Agent) AgentLoopWithProgress(ctx context.Context, userInput string, his
 
 		// 检查是否有工具调用
 		if len(choice.Message.ToolCalls) > 0 {
-			// 思考内容：如果本轮启用了思考流式增量（thinking_stream_*），前端会去重；
-			// 同时也需要在该“思考阶段结束”时补一条可落库的 thinking（用于刷新后持久化展示）。
+			// ReAct 助手正文流式增量（thinking_stream_*）在 UI 上归为「思考」；若与 streamId 重复则前端会去重。
+			// 该条 thinking 用于刷新后持久化展示（与流式聚合一致）。
 			if choice.Message.Content != "" {
 				sendProgress("thinking", choice.Message.Content, map[string]interface{}{
 					"iteration": i + 1,

internal/app/app.go

@@ -16,6 +16,7 @@ import (
 	"cyberstrike-ai/internal/c2"
 	"cyberstrike-ai/internal/config"
 	"cyberstrike-ai/internal/database"
+	"cyberstrike-ai/internal/einoobserve"
 	"cyberstrike-ai/internal/handler"
 	"cyberstrike-ai/internal/knowledge"
 	"cyberstrike-ai/internal/logger"
@@ -90,6 +91,7 @@ func New(cfg *config.Config, log *logger.Logger) (*App, error) {
 
 	// 创建MCP服务器（带数据库持久化）
 	mcpServer := mcp.NewServerWithStorage(log.Logger, db)
+	mcpServer.ConfigureHTTPToolCallTimeoutFromAgentMinutes(cfg.Agent.ToolTimeoutMinutes)
 
 	// 创建安全工具执行器
 	executor := security.NewExecutor(&cfg.Security, mcpServer, log.Logger)
@@ -557,6 +559,10 @@ func (a *App) RunWithContext(ctx context.Context) error {
 
 // Shutdown 关闭应用
 func (a *App) Shutdown() {
+	shutdownCtx, shutdownCancel := context.WithTimeout(context.Background(), 5*time.Second)
+	_ = einoobserve.ShutdownOtel(shutdownCtx)
+	shutdownCancel()
+
 	// 停止钉钉/飞书长连接
 	a.robotMu.Lock()
 	if a.dingCancel != nil {

internal/attackchain/builder.go

@@ -301,7 +301,7 @@ func (b *Builder) formatProcessDetailsForAttackChain(details []database.ProcessD
 		// 目标：以主 agent（编排器）视角输出整轮迭代
 		// - 保留：编排器工具调用/结果、对子代理的 task 调度、子代理最终回复（不含推理）
 		// - 丢弃：thinking/planning/progress 等噪声、子代理的工具细节与推理过程
-		if d.EventType == "progress" || d.EventType == "thinking" || d.EventType == "planning" {
+		if d.EventType == "progress" || d.EventType == "thinking" || d.EventType == "reasoning_chain" || d.EventType == "planning" {
 			continue
 		}
 

internal/config/config.go

@@ -63,6 +63,126 @@ type MultiAgentConfig struct {
 	EinoSkills MultiAgentEinoSkillsConfig `yaml:"eino_skills,omitempty" json:"eino_skills,omitempty"`
 	// EinoMiddleware wires optional ADK middleware (patchtoolcalls, toolsearch, plantask, reduction) and Deep extras.
 	EinoMiddleware MultiAgentEinoMiddlewareConfig `yaml:"eino_middleware,omitempty" json:"eino_middleware,omitempty"`
+	// EinoCallbacks attaches CloudWeGo eino callbacks.InitCallbacks on ADK Runner context (structured logs + optional SSE trace).
+	EinoCallbacks MultiAgentEinoCallbacksConfig `yaml:"eino_callbacks,omitempty" json:"eino_callbacks,omitempty"`
+}
+
+// MultiAgentEinoCallbacksConfig enables Eino unified callbacks on each ADK agent run (deep / plan_execute / supervisor / eino_single).
+// Modes: log_only (zap + optional OTel; no SSE to browser), sse (adds client SSE eino_trace_* when sse_trace_to_client), full (sse rules + stream callback copies closed).
+type MultiAgentEinoCallbacksConfig struct {
+	Enabled bool   `yaml:"enabled" json:"enabled"`
+	Mode    string `yaml:"mode,omitempty" json:"mode,omitempty"` // log_only | sse | full; empty with enabled=true defaults to log_only
+	// SseTraceToClient when true emits eino_trace_* SSE for UI (use only for admin/debug; nil/false recommended in production).
+	SseTraceToClient *bool `yaml:"sse_trace_to_client,omitempty" json:"sse_trace_to_client,omitempty"`
+	// Otel configures OpenTelemetry trace export (independent of mode; exporter none disables export even if enabled).
+	Otel MultiAgentEinoCallbacksOtelConfig `yaml:"otel,omitempty" json:"otel,omitempty"`
+	// MaxInputSummaryRunes / MaxOutputSummaryRunes cap text placed in SSE payloads and debug logs (not full payloads).
+	MaxInputSummaryRunes  int `yaml:"max_input_summary_runes,omitempty" json:"max_input_summary_runes,omitempty"`
+	MaxOutputSummaryRunes int `yaml:"max_output_summary_runes,omitempty" json:"max_output_summary_runes,omitempty"`
+	// ZapVerbose when true logs input/output summaries at zap.Debug on start/end; false uses Info with short fields only.
+	ZapVerbose bool `yaml:"zap_verbose,omitempty" json:"zap_verbose,omitempty"`
+}
+
+// MultiAgentEinoCallbacksOtelConfig OpenTelemetry for Eino callback spans (W3C trace in collector / stdout).
+type MultiAgentEinoCallbacksOtelConfig struct {
+	Enabled     bool    `yaml:"enabled" json:"enabled"`
+	ServiceName string  `yaml:"service_name,omitempty" json:"service_name,omitempty"`
+	Exporter    string  `yaml:"exporter,omitempty" json:"exporter,omitempty"`         // none | stdout | otlphttp
+	OTLPEndpoint string `yaml:"otlp_endpoint,omitempty" json:"otlp_endpoint,omitempty"` // host:port, e.g. localhost:4318 (path /v1/traces)
+	SampleRatio float64 `yaml:"sample_ratio,omitempty" json:"sample_ratio,omitempty"`   // 0–1, default 1.0
+}
+
+// EinoCallbacksModeEffective returns off | log_only | sse | full.
+func (c MultiAgentEinoCallbacksConfig) EinoCallbacksModeEffective() string {
+	if !c.Enabled {
+		return "off"
+	}
+	m := strings.TrimSpace(strings.ToLower(c.Mode))
+	switch m {
+	case "log_only":
+		return "log_only"
+	case "sse":
+		return "sse"
+	case "full":
+		return "full"
+	case "":
+		return "log_only"
+	default:
+		return "log_only"
+	}
+}
+
+// SseTraceToClientEffective is false unless explicitly set true (best practice: do not expose framework traces to end users by default).
+func (c MultiAgentEinoCallbacksConfig) SseTraceToClientEffective() bool {
+	if c.SseTraceToClient == nil {
+		return false
+	}
+	return *c.SseTraceToClient
+}
+
+// ShouldEmitEinoTraceSSE is true when client-visible trace events should be sent over progress/SSE.
+func (c MultiAgentEinoCallbacksConfig) ShouldEmitEinoTraceSSE(mode string) bool {
+	if !c.SseTraceToClientEffective() {
+		return false
+	}
+	return mode == "sse" || mode == "full"
+}
+
+// OtelExporterEffective returns none | stdout | otlphttp.
+func (c MultiAgentEinoCallbacksOtelConfig) OtelExporterEffective() string {
+	e := strings.TrimSpace(strings.ToLower(c.Exporter))
+	switch e {
+	case "none", "stdout", "otlphttp":
+		return e
+	case "":
+		if c.Enabled {
+			return "stdout"
+		}
+		return "none"
+	default:
+		return "none"
+	}
+}
+
+// OtelTracingActive is true when spans should be started (enabled + non-none exporter).
+func (c MultiAgentEinoCallbacksConfig) OtelTracingActive() bool {
+	if !c.Otel.Enabled {
+		return false
+	}
+	return c.Otel.OtelExporterEffective() != "none"
+}
+
+func (c MultiAgentEinoCallbacksOtelConfig) ServiceNameEffective() string {
+	s := strings.TrimSpace(c.ServiceName)
+	if s != "" {
+		return s
+	}
+	return "cyberstrike-ai"
+}
+
+func (c MultiAgentEinoCallbacksOtelConfig) SampleRatioEffective() float64 {
+	r := c.SampleRatio
+	if r <= 0 {
+		return 1.0
+	}
+	if r > 1 {
+		return 1.0
+	}
+	return r
+}
+
+func (c MultiAgentEinoCallbacksConfig) EinoCallbacksMaxInputSummaryRunes() int {
+	if c.MaxInputSummaryRunes > 0 {
+		return c.MaxInputSummaryRunes
+	}
+	return 400
+}
+
+func (c MultiAgentEinoCallbacksConfig) EinoCallbacksMaxOutputSummaryRunes() int {
+	if c.MaxOutputSummaryRunes > 0 {
+		return c.MaxOutputSummaryRunes
+	}
+	return 400
 }
 
 // MultiAgentEinoMiddlewareConfig optional Eino ADK middleware and Deep / supervisor tuning.
@@ -346,6 +466,48 @@ type OpenAIConfig struct {
 	BaseURL        string `yaml:"base_url" json:"base_url"`
 	Model          string `yaml:"model" json:"model"`
 	MaxTotalTokens int    `yaml:"max_total_tokens,omitempty" json:"max_total_tokens,omitempty"`
+	// Reasoning 控制 Eino ChatModel 的 thinking / reasoning_effort / output_config 等（仅 Eino 路径生效；原生 ReAct 忽略）。
+	Reasoning OpenAIReasoningConfig `yaml:"reasoning,omitempty" json:"reasoning,omitempty"`
+}
+
+// OpenAIReasoningConfig 全局默认与网关 profile（对话页可通过 ChatRequest.reasoning 覆盖，受 AllowClientReasoning 约束）。
+type OpenAIReasoningConfig struct {
+	// Mode: auto（默认）| on | off | default（与 auto 相同）。off 时不向模型附加推理扩展字段。
+	Mode string `yaml:"mode,omitempty" json:"mode,omitempty"`
+	// Effort: low | medium | high | max；空表示不单独指定强度（各 profile 行为见 internal/reasoning）。
+	Effort string `yaml:"effort,omitempty" json:"effort,omitempty"`
+	// AllowClientReasoning 为 false 时忽略请求体 reasoning；nil 或未设置等同于 true。
+	AllowClientReasoning *bool `yaml:"allow_client_reasoning,omitempty" json:"allow_client_reasoning,omitempty"`
+	// Profile: auto | deepseek_compat | openai_compat | output_config_effort
+	Profile string `yaml:"profile,omitempty" json:"profile,omitempty"`
+	// ExtraRequestFields 合并进 Chat Completions 根 JSON（管理员用；与自动字段同名时后者覆盖）。
+	ExtraRequestFields map[string]interface{} `yaml:"extra_request_fields,omitempty" json:"extra_request_fields,omitempty"`
+}
+
+// ModeEffective returns auto when empty or default.
+func (c OpenAIReasoningConfig) ModeEffective() string {
+	m := strings.ToLower(strings.TrimSpace(c.Mode))
+	if m == "" || m == "default" {
+		return "auto"
+	}
+	return m
+}
+
+// ProfileEffective returns auto when empty.
+func (c OpenAIReasoningConfig) ProfileEffective() string {
+	p := strings.ToLower(strings.TrimSpace(c.Profile))
+	if p == "" {
+		return "auto"
+	}
+	return p
+}
+
+// AllowClientReasoningEffective true when client may send ChatRequest.reasoning.
+func (c OpenAIReasoningConfig) AllowClientReasoningEffective() bool {
+	if c.AllowClientReasoning == nil {
+		return true
+	}
+	return *c.AllowClientReasoning
 }
 
 type FofaConfig struct {

internal/database/conversation.go

@@ -25,14 +25,15 @@ type Conversation struct {
 
 // Message 消息
 type Message struct {
-	ID              string                   `json:"id"`
-	ConversationID  string                   `json:"conversationId"`
-	Role            string                   `json:"role"`
-	Content         string                   `json:"content"`
-	MCPExecutionIDs []string                 `json:"mcpExecutionIds,omitempty"`
-	ProcessDetails  []map[string]interface{} `json:"processDetails,omitempty"`
-	CreatedAt       time.Time                `json:"createdAt"`
-	UpdatedAt       time.Time                `json:"updatedAt"`
+	ID               string                   `json:"id"`
+	ConversationID string                   `json:"conversationId"`
+	Role             string                   `json:"role"`
+	Content          string                   `json:"content"`
+	ReasoningContent string                   `json:"reasoningContent,omitempty"`
+	MCPExecutionIDs  []string                 `json:"mcpExecutionIds,omitempty"`
+	ProcessDetails   []map[string]interface{} `json:"processDetails,omitempty"`
+	CreatedAt        time.Time                `json:"createdAt"`
+	UpdatedAt        time.Time                `json:"updatedAt"`
 }
 
 // CreateConversation 创建新对话
@@ -498,8 +499,8 @@ func (db *DB) AddMessage(conversationID, role, content string, mcpExecutionIDs [
 	}
 
 	_, err := db.Exec(
-		"INSERT INTO messages (id, conversation_id, role, content, mcp_execution_ids, created_at, updated_at) VALUES (?, ?, ?, ?, ?, ?, ?)",
-		id, conversationID, role, content, mcpIDsJSON, now, now,
+		"INSERT INTO messages (id, conversation_id, role, content, reasoning_content, mcp_execution_ids, created_at, updated_at) VALUES (?, ?, ?, ?, ?, ?, ?, ?)",
+		id, conversationID, role, content, "", mcpIDsJSON, now, now,
 	)
 	if err != nil {
 		return nil, fmt.Errorf("添加消息失败: %w", err)
@@ -523,10 +524,30 @@ func (db *DB) AddMessage(conversationID, role, content string, mcpExecutionIDs [
 	return message, nil
 }
 
+// UpdateAssistantMessageFinalize 更新助手消息终态（正文、MCP id、思考链聚合文本，供无轨迹回退时回放）。
+func (db *DB) UpdateAssistantMessageFinalize(messageID, content string, mcpExecutionIDs []string, reasoningContent string) error {
+	var mcpIDsJSON string
+	if len(mcpExecutionIDs) > 0 {
+		jsonData, err := json.Marshal(mcpExecutionIDs)
+		if err != nil {
+			return fmt.Errorf("序列化MCP执行ID失败: %w", err)
+		}
+		mcpIDsJSON = string(jsonData)
+	}
+	_, err := db.Exec(
+		"UPDATE messages SET content = ?, mcp_execution_ids = ?, reasoning_content = ?, updated_at = ? WHERE id = ?",
+		content, mcpIDsJSON, strings.TrimSpace(reasoningContent), time.Now(), messageID,
+	)
+	if err != nil {
+		return fmt.Errorf("更新助手消息失败: %w", err)
+	}
+	return nil
+}
+
 // GetMessages 获取对话的所有消息
 func (db *DB) GetMessages(conversationID string) ([]Message, error) {
 	rows, err := db.Query(
-		"SELECT id, conversation_id, role, content, mcp_execution_ids, created_at, updated_at FROM messages WHERE conversation_id = ? ORDER BY created_at ASC",
+		"SELECT id, conversation_id, role, content, reasoning_content, mcp_execution_ids, created_at, updated_at FROM messages WHERE conversation_id = ? ORDER BY created_at ASC",
 		conversationID,
 	)
 	if err != nil {
@@ -537,13 +558,17 @@ func (db *DB) GetMessages(conversationID string) ([]Message, error) {
 	var messages []Message
 	for rows.Next() {
 		var msg Message
+		var reasoning sql.NullString
 		var mcpIDsJSON sql.NullString
 		var createdAt string
 		var updatedAt sql.NullString
 
-		if err := rows.Scan(&msg.ID, &msg.ConversationID, &msg.Role, &msg.Content, &mcpIDsJSON, &createdAt, &updatedAt); err != nil {
+		if err := rows.Scan(&msg.ID, &msg.ConversationID, &msg.Role, &msg.Content, &reasoning, &mcpIDsJSON, &createdAt, &updatedAt); err != nil {
 			return nil, fmt.Errorf("扫描消息失败: %w", err)
 		}
+		if reasoning.Valid {
+			msg.ReasoningContent = reasoning.String
+		}
 
 		// 尝试多种时间格式解析
 		var err error
@@ -683,7 +708,7 @@ type ProcessDetail struct {
 	ID             string    `json:"id"`
 	MessageID      string    `json:"messageId"`
 	ConversationID string    `json:"conversationId"`
-	EventType      string    `json:"eventType"` // iteration, thinking, tool_calls_detected, tool_call, tool_result, progress, error
+	EventType      string    `json:"eventType"` // iteration, thinking, reasoning_chain, tool_calls_detected, tool_call, tool_result, progress, error
 	Message        string    `json:"message"`
 	Data           string    `json:"data"` // JSON格式的数据
 	CreatedAt      time.Time `json:"createdAt"`

internal/database/database.go

@@ -594,6 +594,25 @@ func (db *DB) migrateMessagesTable() error {
 
 	// 回填已有数据：让 updated_at 至少等于 created_at，避免前端出现空/当前时间回退。
 	_, _ = db.Exec("UPDATE messages SET updated_at = created_at WHERE updated_at IS NULL OR updated_at = ''")
+
+	// reasoning_content：DeepSeek 思考模式 + 工具调用续跑；与 last_react_input 互补，供消息表回退路径回放
+	var rcColCount int
+	errRC := db.QueryRow("SELECT COUNT(*) FROM pragma_table_info('messages') WHERE name='reasoning_content'").Scan(&rcColCount)
+	if errRC != nil {
+		if _, addErr := db.Exec("ALTER TABLE messages ADD COLUMN reasoning_content TEXT"); addErr != nil {
+			errMsg := strings.ToLower(addErr.Error())
+			if !strings.Contains(errMsg, "duplicate column") && !strings.Contains(errMsg, "already exists") {
+				return fmt.Errorf("添加 messages.reasoning_content 字段失败: %w", addErr)
+			}
+		}
+	} else if rcColCount == 0 {
+		if _, err := db.Exec("ALTER TABLE messages ADD COLUMN reasoning_content TEXT"); err != nil {
+			errMsg := strings.ToLower(err.Error())
+			if !strings.Contains(errMsg, "duplicate column") && !strings.Contains(errMsg, "already exists") {
+				return fmt.Errorf("添加 messages.reasoning_content 字段失败: %w", err)
+			}
+		}
+	}
 	return nil
 }
 

internal/einoobserve/attach.go

@@ -0,0 +1,435 @@
+// Package einoobserve attaches CloudWeGo Eino [callbacks.Handler] to ADK Runner contexts for
+// structured logging and optional SSE trace events (eino_trace_*).
+package einoobserve
+
+import (
+	"context"
+	"encoding/json"
+	"fmt"
+	"strings"
+	"sync"
+	"sync/atomic"
+	"time"
+
+	"cyberstrike-ai/internal/config"
+
+	"github.com/cloudwego/eino/adk"
+	"github.com/cloudwego/eino/callbacks"
+	"github.com/cloudwego/eino/components"
+	"github.com/cloudwego/eino/components/model"
+	"github.com/cloudwego/eino/components/tool"
+	"github.com/cloudwego/eino/schema"
+	"github.com/google/uuid"
+	"go.opentelemetry.io/otel"
+	"go.opentelemetry.io/otel/attribute"
+	"go.opentelemetry.io/otel/codes"
+	"go.opentelemetry.io/otel/trace"
+	"go.uber.org/zap"
+)
+
+type ctxSpanKey struct{}
+
+type ctxOtelSpanKey struct{}
+
+// Params for attaching per-run callback instrumentation.
+type Params struct {
+	Logger           *zap.Logger
+	Progress         func(eventType, message string, data interface{})
+	ConversationID   string
+	OrchMode         string
+	OrchestratorName string
+}
+
+// AttachAgentRunCallbacks returns ctx wrapped with callbacks.InitCallbacks when enabled.
+// Safe to call with nil cfg or disabled cfg (returns ctx unchanged).
+func AttachAgentRunCallbacks(ctx context.Context, cfg *config.MultiAgentEinoCallbacksConfig, p Params) context.Context {
+	if ctx == nil {
+		return ctx
+	}
+	if cfg == nil || !cfg.Enabled {
+		return ctx
+	}
+	mode := cfg.EinoCallbacksModeEffective()
+	if mode == "off" {
+		return ctx
+	}
+	runID := uuid.New().String()
+	if p.Progress != nil && cfg.ShouldEmitEinoTraceSSE(mode) {
+		p.Progress("eino_trace_run", "Eino callbacks session", map[string]interface{}{
+			"runId":            runID,
+			"conversationId":   strings.TrimSpace(p.ConversationID),
+			"orchestration":    strings.TrimSpace(p.OrchMode),
+			"orchestratorName": strings.TrimSpace(p.OrchestratorName),
+			"observeMode":      mode,
+			"source":           "eino_callbacks",
+		})
+	}
+	h := &runHandler{
+		cfg:    *cfg,
+		mode:   mode,
+		params: p,
+		runID:  runID,
+	}
+	b := callbacks.NewHandlerBuilder().
+		OnStartFn(h.onStart).
+		OnEndFn(h.onEnd).
+		OnErrorFn(h.onError)
+	if mode == "full" {
+		b = b.OnStartWithStreamInputFn(h.onStartStreamIn).OnEndWithStreamOutputFn(h.onEndStreamOut)
+	}
+	ri := &callbacks.RunInfo{
+		Name:      "CyberStrikeADKRun",
+		Type:      strings.TrimSpace(p.OrchMode),
+		Component: components.Component("AgentSession"),
+	}
+	return callbacks.InitCallbacks(ctx, ri, b.Build())
+}
+
+type runHandler struct {
+	cfg    config.MultiAgentEinoCallbacksConfig
+	mode   string
+	params Params
+	runID  string
+
+	mu        sync.Mutex
+	spanStack []string
+	seq       atomic.Uint64
+}
+
+func (h *runHandler) genSpanID() string {
+	return fmt.Sprintf("%s-%d", h.runID, h.seq.Add(1))
+}
+
+func (h *runHandler) popSpan() (id string) {
+	h.mu.Lock()
+	defer h.mu.Unlock()
+	if len(h.spanStack) == 0 {
+		return ""
+	}
+	id = h.spanStack[len(h.spanStack)-1]
+	h.spanStack = h.spanStack[:len(h.spanStack)-1]
+	return id
+}
+
+// popMatching removes the given id from the stack top if it matches; otherwise pops until empty or match (rare ordering mismatch).
+func (h *runHandler) popMatching(want string) string {
+	h.mu.Lock()
+	defer h.mu.Unlock()
+	if want == "" {
+		if len(h.spanStack) == 0 {
+			return ""
+		}
+		id := h.spanStack[len(h.spanStack)-1]
+		h.spanStack = h.spanStack[:len(h.spanStack)-1]
+		return id
+	}
+	for len(h.spanStack) > 0 {
+		top := h.spanStack[len(h.spanStack)-1]
+		h.spanStack = h.spanStack[:len(h.spanStack)-1]
+		if top == want {
+			return top
+		}
+	}
+	return want
+}
+
+func (h *runHandler) onStart(ctx context.Context, info *callbacks.RunInfo, input callbacks.CallbackInput) context.Context {
+	var parentID string
+	h.mu.Lock()
+	if len(h.spanStack) > 0 {
+		parentID = h.spanStack[len(h.spanStack)-1]
+	}
+	spanID := h.genSpanID()
+	h.spanStack = append(h.spanStack, spanID)
+	h.mu.Unlock()
+
+	inSum := summarizeCallbackInput(input, h.cfg.EinoCallbacksMaxInputSummaryRunes())
+	if h.cfg.OtelTracingActive() {
+		tracer := otel.Tracer("cyberstrike/eino")
+		spanName := callbackSpanName(info)
+		var sp trace.Span
+		ctx, sp = tracer.Start(ctx, spanName,
+			trace.WithSpanKind(trace.SpanKindInternal),
+			trace.WithAttributes(
+				attribute.String("eino.component", string(info.Component)),
+				attribute.String("eino.name", info.Name),
+				attribute.String("eino.type", info.Type),
+				attribute.String("cyberstrike.run_id", h.runID),
+				attribute.String("cyberstrike.conversation_id", strings.TrimSpace(h.params.ConversationID)),
+				attribute.String("cyberstrike.orchestration", strings.TrimSpace(h.params.OrchMode)),
+			),
+		)
+		if inSum != "" {
+			sp.SetAttributes(attribute.String("eino.input.summary", truncateForAttr(inSum, 256)))
+		}
+		ctx = context.WithValue(ctx, ctxOtelSpanKey{}, sp)
+	}
+	if h.params.Logger != nil {
+		fields := []zap.Field{
+			zap.String("runId", h.runID),
+			zap.String("spanId", spanID),
+			zap.String("parentSpanId", parentID),
+			zap.String("component", string(info.Component)),
+			zap.String("name", info.Name),
+			zap.String("type", info.Type),
+			zap.String("phase", "start"),
+		}
+		if sp, ok := ctx.Value(ctxOtelSpanKey{}).(trace.Span); ok && sp != nil {
+			if sc := sp.SpanContext(); sc.IsValid() {
+				fields = append(fields,
+					zap.String("trace_id", sc.TraceID().String()),
+					zap.String("otel_span_id", sc.SpanID().String()),
+				)
+			}
+		}
+		if h.cfg.ZapVerbose {
+			h.params.Logger.Debug("eino_callback", append(fields, zap.String("inputSummary", inSum))...)
+		} else {
+			h.params.Logger.Info("eino_callback", fields...)
+		}
+	}
+	if h.params.Progress != nil && h.cfg.ShouldEmitEinoTraceSSE(h.mode) {
+		h.params.Progress("eino_trace_start", "", map[string]interface{}{
+			"runId":          h.runID,
+			"spanId":         spanID,
+			"parentSpanId":   parentID,
+			"conversationId": strings.TrimSpace(h.params.ConversationID),
+			"orchestration":    strings.TrimSpace(h.params.OrchMode),
+			"component":      string(info.Component),
+			"name":           info.Name,
+			"type":           info.Type,
+			"ts":             time.Now().UTC().Format(time.RFC3339Nano),
+			"inputSummary":   inSum,
+			"source":         "eino_callbacks",
+		})
+	}
+	ctx = context.WithValue(ctx, ctxSpanKey{}, spanID)
+	return ctx
+}
+
+func (h *runHandler) onEnd(ctx context.Context, info *callbacks.RunInfo, output callbacks.CallbackOutput) context.Context {
+	spanID, _ := ctx.Value(ctxSpanKey{}).(string)
+	if spanID == "" {
+		spanID = h.popSpan()
+	} else {
+		spanID = h.popMatching(spanID)
+	}
+	outSum := summarizeCallbackOutput(output, h.cfg.EinoCallbacksMaxOutputSummaryRunes())
+	if sp, ok := ctx.Value(ctxOtelSpanKey{}).(trace.Span); ok && sp != nil {
+		if outSum != "" {
+			sp.SetAttributes(attribute.String("eino.output.summary", truncateForAttr(outSum, 256)))
+		}
+		sp.SetStatus(codes.Ok, "")
+		sp.End()
+	}
+	if h.params.Logger != nil {
+		fields := []zap.Field{
+			zap.String("runId", h.runID),
+			zap.String("spanId", spanID),
+			zap.String("component", string(info.Component)),
+			zap.String("name", info.Name),
+			zap.String("type", info.Type),
+			zap.String("phase", "end"),
+		}
+		if h.cfg.ZapVerbose {
+			h.params.Logger.Debug("eino_callback", append(fields, zap.String("outputSummary", outSum))...)
+		} else {
+			h.params.Logger.Info("eino_callback", fields...)
+		}
+	}
+	if h.params.Progress != nil && h.cfg.ShouldEmitEinoTraceSSE(h.mode) {
+		h.params.Progress("eino_trace_end", "", map[string]interface{}{
+			"runId":          h.runID,
+			"spanId":         spanID,
+			"conversationId": strings.TrimSpace(h.params.ConversationID),
+			"orchestration":    strings.TrimSpace(h.params.OrchMode),
+			"component":      string(info.Component),
+			"name":           info.Name,
+			"type":           info.Type,
+			"ts":             time.Now().UTC().Format(time.RFC3339Nano),
+			"outputSummary":  outSum,
+			"source":         "eino_callbacks",
+		})
+	}
+	return ctx
+}
+
+func (h *runHandler) onError(ctx context.Context, info *callbacks.RunInfo, err error) context.Context {
+	spanID, _ := ctx.Value(ctxSpanKey{}).(string)
+	if spanID == "" {
+		spanID = h.popSpan()
+	} else {
+		spanID = h.popMatching(spanID)
+	}
+	msg := ""
+	if err != nil {
+		msg = truncateRunes(err.Error(), h.cfg.EinoCallbacksMaxOutputSummaryRunes())
+	}
+	if sp, ok := ctx.Value(ctxOtelSpanKey{}).(trace.Span); ok && sp != nil {
+		if err != nil {
+			sp.RecordError(err)
+		}
+		sp.SetStatus(codes.Error, msg)
+		sp.End()
+	}
+	if h.params.Logger != nil {
+		h.params.Logger.Warn("eino_callback_error",
+			zap.String("runId", h.runID),
+			zap.String("spanId", spanID),
+			zap.String("component", string(info.Component)),
+			zap.String("name", info.Name),
+			zap.String("type", info.Type),
+			zap.Error(err),
+		)
+	}
+	if h.params.Progress != nil && h.cfg.ShouldEmitEinoTraceSSE(h.mode) {
+		h.params.Progress("eino_trace_error", msg, map[string]interface{}{
+			"runId":          h.runID,
+			"spanId":         spanID,
+			"conversationId": strings.TrimSpace(h.params.ConversationID),
+			"orchestration":    strings.TrimSpace(h.params.OrchMode),
+			"component":      string(info.Component),
+			"name":           info.Name,
+			"type":           info.Type,
+			"ts":             time.Now().UTC().Format(time.RFC3339Nano),
+			"error":          msg,
+			"source":         "eino_callbacks",
+		})
+	}
+	return ctx
+}
+
+func (h *runHandler) onStartStreamIn(ctx context.Context, info *callbacks.RunInfo, input *schema.StreamReader[callbacks.CallbackInput]) context.Context {
+	if input != nil {
+		input.Close()
+	}
+	if h.params.Logger != nil {
+		h.params.Logger.Debug("eino_callback_stream_in",
+			zap.String("runId", h.runID),
+			zap.String("component", string(info.Component)),
+			zap.String("name", info.Name),
+		)
+	}
+	return ctx
+}
+
+func (h *runHandler) onEndStreamOut(ctx context.Context, info *callbacks.RunInfo, output *schema.StreamReader[callbacks.CallbackOutput]) context.Context {
+	if output != nil {
+		output.Close()
+	}
+	if h.params.Logger != nil {
+		h.params.Logger.Debug("eino_callback_stream_out",
+			zap.String("runId", h.runID),
+			zap.String("component", string(info.Component)),
+			zap.String("name", info.Name),
+		)
+	}
+	return ctx
+}
+
+func callbackSpanName(info *callbacks.RunInfo) string {
+	if info == nil {
+		return "eino.callback"
+	}
+	comp := strings.TrimSpace(string(info.Component))
+	name := strings.TrimSpace(info.Name)
+	typ := strings.TrimSpace(info.Type)
+	if name != "" && comp != "" {
+		return comp + "/" + name
+	}
+	if typ != "" && comp != "" {
+		return comp + "[" + typ + "]"
+	}
+	if comp != "" {
+		return comp
+	}
+	return "eino.callback"
+}
+
+func truncateForAttr(s string, maxRunes int) string {
+	return truncateRunes(s, maxRunes)
+}
+
+func summarizeCallbackInput(in callbacks.CallbackInput, maxRunes int) string {
+	if in == nil {
+		return ""
+	}
+	if ai := adk.ConvAgentCallbackInput(in); ai != nil {
+		parts := []string{"agent"}
+		if ai.Input != nil {
+			parts = append(parts, fmt.Sprintf("messages=%d", len(ai.Input.Messages)))
+		}
+		if ai.ResumeInfo != nil {
+			parts = append(parts, "resume=true")
+		}
+		return strings.Join(parts, " ")
+	}
+	if mi := model.ConvCallbackInput(in); mi != nil {
+		return fmt.Sprintf("chatModel messages=%d tools=%d", len(mi.Messages), len(mi.Tools))
+	}
+	if ti := tool.ConvCallbackInput(in); ti != nil {
+		raw := ti.ArgumentsInJSON
+		return "tool args=" + truncateRunes(raw, maxRunes)
+	}
+	b, err := json.Marshal(in)
+	if err != nil {
+		return fmt.Sprintf("%T", in)
+	}
+	return truncateRunes(string(b), maxRunes)
+}
+
+func summarizeCallbackOutput(out callbacks.CallbackOutput, maxRunes int) string {
+	if out == nil {
+		return ""
+	}
+	if ao := adk.ConvAgentCallbackOutput(out); ao != nil {
+		return "agent_events=stream"
+	}
+	if mo := model.ConvCallbackOutput(out); mo != nil && mo.Message != nil {
+		s := ""
+		if mo.Message.Content != "" {
+			s = mo.Message.Content
+		}
+		if mo.TokenUsage != nil {
+			return fmt.Sprintf("tokens total=%d completion=%d prompt=%d text=%s",
+				mo.TokenUsage.TotalTokens, mo.TokenUsage.CompletionTokens, mo.TokenUsage.PromptTokens,
+				truncateRunes(s, minInt(120, maxRunes)))
+		}
+		return "assistant len=" + itoa(len(s))
+	}
+	if to := tool.ConvCallbackOutput(out); to != nil {
+		if to.Response != "" {
+			return truncateRunes(to.Response, maxRunes)
+		}
+		if to.ToolOutput != nil {
+			return "tool_result multimodal"
+		}
+	}
+	b, err := json.Marshal(out)
+	if err != nil {
+		return fmt.Sprintf("%T", out)
+	}
+	return truncateRunes(string(b), maxRunes)
+}
+
+func minInt(a, b int) int {
+	if a < b {
+		return a
+	}
+	return b
+}
+
+func itoa(n int) string {
+	return fmt.Sprintf("%d", n)
+}
+
+func truncateRunes(s string, maxRunes int) string {
+	if maxRunes <= 0 {
+		return ""
+	}
+	r := []rune(s)
+	if len(r) <= maxRunes {
+		return s
+	}
+	return string(r[:maxRunes]) + "…"
+}

internal/einoobserve/attach_test.go

@@ -0,0 +1,26 @@
+package einoobserve
+
+import (
+	"context"
+	"testing"
+
+	"cyberstrike-ai/internal/config"
+)
+
+func TestAttachAgentRunCallbacks_Disabled(t *testing.T) {
+	ctx := context.Background()
+	cfg := &config.MultiAgentEinoCallbacksConfig{Enabled: false}
+	out := AttachAgentRunCallbacks(ctx, cfg, Params{})
+	if out != ctx {
+		t.Fatalf("expected same ctx when disabled")
+	}
+}
+
+func TestTruncateRunes(t *testing.T) {
+	if got := truncateRunes("abc", 10); got != "abc" {
+		t.Fatalf("got %q", got)
+	}
+	if got := truncateRunes("abcdefghij", 4); got != "abcd…" {
+		t.Fatalf("got %q", got)
+	}
+}

internal/einoobserve/otel.go

@@ -0,0 +1,111 @@
+package einoobserve
+
+import (
+	"context"
+	"fmt"
+	"strings"
+	"sync"
+
+	"cyberstrike-ai/internal/config"
+
+	"go.opentelemetry.io/otel"
+	"go.opentelemetry.io/otel/exporters/otlp/otlptrace/otlptracehttp"
+	"go.opentelemetry.io/otel/exporters/stdout/stdouttrace"
+	"go.opentelemetry.io/otel/sdk/resource"
+	sdktrace "go.opentelemetry.io/otel/sdk/trace"
+	semconv "go.opentelemetry.io/otel/semconv/v1.26.0"
+	"go.uber.org/zap"
+)
+
+var (
+	otelMu           sync.Mutex
+	otelShutdown     func(context.Context) error
+	otelInitialized  bool
+)
+
+// InitOtelFromConfig installs the global OpenTelemetry TracerProvider when
+// eino_callbacks.otel is enabled and exporter is not none. Safe to call multiple times.
+func InitOtelFromConfig(cfg *config.MultiAgentEinoCallbacksConfig, log *zap.Logger) (shutdown func(context.Context) error, err error) {
+	shutdown = func(context.Context) error { return nil }
+	if cfg == nil || !cfg.OtelTracingActive() {
+		return shutdown, nil
+	}
+
+	otelMu.Lock()
+	defer otelMu.Unlock()
+	if otelInitialized {
+		if otelShutdown != nil {
+			return otelShutdown, nil
+		}
+		return shutdown, nil
+	}
+
+	oc := cfg.Otel
+	expKind := oc.OtelExporterEffective()
+	ctx := context.Background()
+
+	var exporter sdktrace.SpanExporter
+	switch expKind {
+	case "stdout":
+		exporter, err = stdouttrace.New()
+		if err != nil {
+			return shutdown, fmt.Errorf("eino otel stdout exporter: %w", err)
+		}
+	case "otlphttp":
+		ep := strings.TrimSpace(oc.OTLPEndpoint)
+		if ep == "" {
+			ep = "localhost:4318"
+		}
+		exporter, err = otlptracehttp.New(ctx,
+			otlptracehttp.WithEndpoint(ep),
+			otlptracehttp.WithURLPath("/v1/traces"),
+		)
+		if err != nil {
+			return shutdown, fmt.Errorf("eino otel otlphttp exporter: %w", err)
+		}
+	default:
+		return shutdown, nil
+	}
+
+	res, err := resource.New(ctx,
+		resource.WithAttributes(
+			semconv.ServiceName(oc.ServiceNameEffective()),
+		),
+	)
+	if err != nil {
+		return shutdown, fmt.Errorf("eino otel resource: %w", err)
+	}
+
+	sampler := sdktrace.ParentBased(sdktrace.TraceIDRatioBased(oc.SampleRatioEffective()))
+	tp := sdktrace.NewTracerProvider(
+		sdktrace.WithBatcher(exporter),
+		sdktrace.WithResource(res),
+		sdktrace.WithSampler(sampler),
+	)
+	otel.SetTracerProvider(tp)
+
+	otelShutdown = tp.Shutdown
+	otelInitialized = true
+	if log != nil {
+		log.Info("eino otel: tracer provider initialized",
+			zap.String("exporter", expKind),
+			zap.String("service", oc.ServiceNameEffective()),
+			zap.Float64("sample_ratio", oc.SampleRatioEffective()),
+		)
+	}
+	return otelShutdown, nil
+}
+
+// ShutdownOtel flushes and shuts down the global TracerProvider if it was installed.
+func ShutdownOtel(ctx context.Context) error {
+	otelMu.Lock()
+	fn := otelShutdown
+	otelShutdown = nil
+	inited := otelInitialized
+	otelInitialized = false
+	otelMu.Unlock()
+	if !inited || fn == nil {
+		return nil
+	}
+	return fn(ctx)
+}

internal/handler/agent.go

@@ -19,6 +19,7 @@ import (
 	"cyberstrike-ai/internal/agent"
 	"cyberstrike-ai/internal/config"
 	"cyberstrike-ai/internal/database"
+	"cyberstrike-ai/internal/reasoning"
 	"cyberstrike-ai/internal/mcp"
 	"cyberstrike-ai/internal/mcp/builtin"
 	"cyberstrike-ai/internal/multiagent"
@@ -201,6 +202,14 @@ type ChatAttachment struct {
 	ServerPath string `json:"serverPath,omitempty"` // 已保存在 chat_uploads 下的绝对路径（由 POST /api/chat-uploads 返回）
 }
 
+// ChatReasoningRequest 对话页「模型推理」意图（仅 Eino 路径消费；原生 agent-loop 忽略）。
+type ChatReasoningRequest struct {
+	// Mode: default（跟随系统）| off | on | auto
+	Mode string `json:"mode,omitempty"`
+	// Effort: low | medium | high | max；空表示不指定（由系统默认与各 profile 决定）。
+	Effort string `json:"effort,omitempty"`
+}
+
 // ChatRequest 聊天请求
 type ChatRequest struct {
 	Message              string           `json:"message" binding:"required"`
@@ -209,10 +218,18 @@ type ChatRequest struct {
 	Attachments          []ChatAttachment `json:"attachments,omitempty"`
 	WebShellConnectionID string           `json:"webshellConnectionId,omitempty"` // WebShell 管理 - AI 助手：当前选中的连接 ID，仅使用 webshell_* 工具
 	Hitl                 *HITLRequest     `json:"hitl,omitempty"`
+	Reasoning            *ChatReasoningRequest `json:"reasoning,omitempty"`
 	// Orchestration 仅对 /api/multi-agent、/api/multi-agent/stream：deep | plan_execute | supervisor；空则等同 deep。机器人/批量等无请求体时由服务端默认 deep。/api/eino-agent* 不使用此字段。
 	Orchestration string `json:"orchestration,omitempty"`
 }
 
+func chatReasoningToClientIntent(r *ChatReasoningRequest) *reasoning.ClientIntent {
+	if r == nil {
+		return nil
+	}
+	return &reasoning.ClientIntent{Mode: r.Mode, Effort: r.Effort}
+}
+
 type HITLRequest struct {
 	Enabled        bool     `json:"enabled"`
 	Mode           string   `json:"mode,omitempty"`
@@ -567,14 +584,7 @@ func (h *AgentHandler) AgentLoop(c *gin.Context) {
 			h.logger.Warn("获取历史消息失败", zap.Error(err))
 			agentHistoryMessages = []agent.ChatMessage{}
 		} else {
-			// 将数据库消息转换为Agent消息格式
-			agentHistoryMessages = make([]agent.ChatMessage, 0, len(historyMessages))
-			for _, msg := range historyMessages {
-				agentHistoryMessages = append(agentHistoryMessages, agent.ChatMessage{
-					Role:    msg.Role,
-					Content: msg.Content,
-				})
-			}
+			agentHistoryMessages = dbMessagesToAgentChatMessages(historyMessages)
 			h.logger.Info("从消息表加载历史消息", zap.Int("count", len(agentHistoryMessages)))
 		}
 	} else {
@@ -775,6 +785,7 @@ func (h *AgentHandler) ProcessMessageForRobot(ctx context.Context, conversationI
 			progressCallback,
 			h.agentsMarkdownDir,
 			"deep",
+			nil,
 		)
 		if errMA != nil {
 			if shouldPersistEinoAgentTraceAfterRunError(ctx) {
@@ -788,17 +799,8 @@ func (h *AgentHandler) ProcessMessageForRobot(ctx context.Context, conversationI
 			return "", conversationID, errMA
 		}
 		if assistantMessageID != "" {
-			mcpIDsJSON := ""
-			if len(resultMA.MCPExecutionIDs) > 0 {
-				jsonData, _ := json.Marshal(resultMA.MCPExecutionIDs)
-				mcpIDsJSON = string(jsonData)
-			}
-			_, err = h.db.Exec(
-				"UPDATE messages SET content = ?, mcp_execution_ids = ?, updated_at = ? WHERE id = ?",
-				resultMA.Response, mcpIDsJSON, time.Now(), assistantMessageID,
-			)
-			if err != nil {
-				h.logger.Warn("机器人：更新助手消息失败", zap.Error(err))
+			if errU := h.db.UpdateAssistantMessageFinalize(assistantMessageID, resultMA.Response, resultMA.MCPExecutionIDs, multiagent.AggregatedReasoningFromTraceJSON(resultMA.LastAgentTraceInput)); errU != nil {
+				h.logger.Warn("机器人：更新助手消息失败", zap.Error(errU))
 			}
 		} else {
 			if _, err = h.db.AddMessage(conversationID, "assistant", resultMA.Response, resultMA.MCPExecutionIDs); err != nil {
@@ -823,17 +825,8 @@ func (h *AgentHandler) ProcessMessageForRobot(ctx context.Context, conversationI
 
 	// 更新助手消息内容与 MCP 执行 ID（与 stream 一致）
 	if assistantMessageID != "" {
-		mcpIDsJSON := ""
-		if len(result.MCPExecutionIDs) > 0 {
-			jsonData, _ := json.Marshal(result.MCPExecutionIDs)
-			mcpIDsJSON = string(jsonData)
-		}
-		_, err = h.db.Exec(
-		"UPDATE messages SET content = ?, mcp_execution_ids = ?, updated_at = ? WHERE id = ?",
-		result.Response, mcpIDsJSON, time.Now(), assistantMessageID,
-		)
-		if err != nil {
-			h.logger.Warn("机器人：更新助手消息失败", zap.Error(err))
+		if errU := h.db.UpdateAssistantMessageFinalize(assistantMessageID, result.Response, result.MCPExecutionIDs, multiagent.AggregatedReasoningFromTraceJSON(result.LastAgentTraceInput)); errU != nil {
+			h.logger.Warn("机器人：更新助手消息失败", zap.Error(errU))
 		}
 	} else {
 		if _, err = h.db.AddMessage(conversationID, "assistant", result.Response, result.MCPExecutionIDs); err != nil {
@@ -891,10 +884,12 @@ func (h *AgentHandler) createProgressCallback(runCtx context.Context, cancelRun
 		return ""
 	}
 
-	// thinking_stream_*：不逐条落库，按 streamId 聚合，在后续关键事件前补一条可持久化的 thinking
+	// thinking_stream_*（ReAct 等助手正文流）与 reasoning_chain_stream_*（Eino ReasoningContent）：
+	// 不逐条落库，按 streamId 聚合，flush 时分别落 thinking / reasoning_chain。
 	type thinkingBuf struct {
-		b    strings.Builder
-		meta map[string]interface{}
+		b         strings.Builder
+		meta      map[string]interface{}
+		persistAs string // "thinking" | "reasoning_chain"
 	}
 	thinkingStreams := make(map[string]*thinkingBuf) // streamId -> buf
 	flushedThinking := make(map[string]bool)         // streamId -> flushed
@@ -948,8 +943,12 @@ func (h *AgentHandler) createProgressCallback(runCtx context.Context, cancelRun
 				}
 				data[k] = v
 			}
-			if err := h.db.AddProcessDetail(assistantMessageID, conversationID, "thinking", content, data); err != nil {
-				h.logger.Warn("保存过程详情失败", zap.Error(err), zap.String("eventType", "thinking"))
+			persist := tb.persistAs
+			if persist != "reasoning_chain" {
+				persist = "thinking"
+			}
+			if err := h.db.AddProcessDetail(assistantMessageID, conversationID, persist, content, data); err != nil {
+				h.logger.Warn("保存过程详情失败", zap.Error(err), zap.String("eventType", persist))
 			}
 			flushedThinking[sid] = true
 		}
@@ -1177,14 +1176,20 @@ func (h *AgentHandler) createProgressCallback(runCtx context.Context, cancelRun
 			return
 		}
 
-		// 聚合 thinking_stream_*（ReasoningContent），不逐条落库
-		if eventType == "thinking_stream_start" {
+		// 聚合 thinking_stream_* / reasoning_chain_stream_*，不逐条落库
+		if eventType == "thinking_stream_start" || eventType == "reasoning_chain_stream_start" {
+			persistAs := "thinking"
+			if eventType == "reasoning_chain_stream_start" {
+				persistAs = "reasoning_chain"
+			}
 			if dataMap, ok := data.(map[string]interface{}); ok {
 				if sid, ok2 := dataMap["streamId"].(string); ok2 && sid != "" {
 					tb := thinkingStreams[sid]
 					if tb == nil {
-						tb = &thinkingBuf{meta: map[string]interface{}{}}
+						tb = &thinkingBuf{meta: map[string]interface{}{}, persistAs: persistAs}
 						thinkingStreams[sid] = tb
+					} else {
+						tb.persistAs = persistAs
 					}
 					// 记录元信息（source/einoAgent/einoRole/iteration 等）
 					for k, v := range dataMap {
@@ -1194,15 +1199,21 @@ func (h *AgentHandler) createProgressCallback(runCtx context.Context, cancelRun
 			}
 			return
 		}
-		if eventType == "thinking_stream_delta" {
+		if eventType == "thinking_stream_delta" || eventType == "reasoning_chain_stream_delta" {
+			persistAs := "thinking"
+			if eventType == "reasoning_chain_stream_delta" {
+				persistAs = "reasoning_chain"
+			}
 			if dataMap, ok := data.(map[string]interface{}); ok {
 				if sid, ok2 := dataMap["streamId"].(string); ok2 && sid != "" {
 					tb := thinkingStreams[sid]
 					if tb == nil {
-						tb = &thinkingBuf{meta: map[string]interface{}{}}
+						tb = &thinkingBuf{meta: map[string]interface{}{}, persistAs: persistAs}
 						thinkingStreams[sid] = tb
+					} else if tb.persistAs == "" {
+						tb.persistAs = persistAs
 					}
-					// delta 片段直接拼接；message 本身就是 reasoning content
+					// delta 片段直接拼接
 					tb.b.WriteString(message)
 					// 有时 delta 先到 start 未到，补充元信息
 					for k, v := range dataMap {
@@ -1213,10 +1224,9 @@ func (h *AgentHandler) createProgressCallback(runCtx context.Context, cancelRun
 			return
 		}
 
-		// 当 Agent 同时发送 thinking_stream_* 和 thinking（带同一 streamId）时，
-		// thinking_stream_* 已经会在 flushThinkingStreams() 聚合落库；
-		// 这里跳过同 streamId 的 thinking，避免 processDetails 双份展示。
-		if eventType == "thinking" {
+		// 当 Agent 同时发送 *_stream_* 与同名 streamId 的 thinking/reasoning_chain 时，
+		// 流式聚合已会在 flushThinkingStreams() 落库；此处跳过逐条重复。
+		if eventType == "thinking" || eventType == "reasoning_chain" {
 			if dataMap, ok := data.(map[string]interface{}); ok {
 				if sid, ok2 := dataMap["streamId"].(string); ok2 && sid != "" {
 					if tb, exists := thinkingStreams[sid]; exists && tb != nil {
@@ -1239,13 +1249,17 @@ func (h *AgentHandler) createProgressCallback(runCtx context.Context, cancelRun
 			eventType != "response_start" &&
 			eventType != "response_delta" &&
 			eventType != "tool_result_delta" &&
+			eventType != "eino_trace_run" &&
+			eventType != "eino_trace_start" &&
+			eventType != "eino_trace_end" &&
+			eventType != "eino_trace_error" &&
 			eventType != "eino_agent_reply_stream_start" &&
 			eventType != "eino_agent_reply_stream_delta" &&
 			eventType != "eino_agent_reply_stream_end" {
 			if eventType == "tool_result" {
 				discardPlanningIfEchoesToolResult(&respPlan, data)
 			}
-			// 在关键过程事件落库前，先把「规划中」与 thinking_stream 落库
+			// 在关键过程事件落库前，先把「规划中」与聚合中的 thinking / reasoning_chain 流落库
 			flushResponsePlan()
 			flushThinkingStreams()
 			if err := h.db.AddProcessDetail(assistantMessageID, conversationID, eventType, message, data); err != nil {
@@ -1427,14 +1441,7 @@ func (h *AgentHandler) AgentLoopStream(c *gin.Context) {
 			h.logger.Warn("获取历史消息失败", zap.Error(err))
 			agentHistoryMessages = []agent.ChatMessage{}
 		} else {
-			// 将数据库消息转换为Agent消息格式
-			agentHistoryMessages = make([]agent.ChatMessage, 0, len(historyMessages))
-			for _, msg := range historyMessages {
-				agentHistoryMessages = append(agentHistoryMessages, agent.ChatMessage{
-					Role:    msg.Role,
-					Content: msg.Content,
-				})
-			}
+			agentHistoryMessages = dbMessagesToAgentChatMessages(historyMessages)
 			h.logger.Info("从消息表加载历史消息", zap.Int("count", len(agentHistoryMessages)))
 		}
 	} else {
@@ -1727,20 +1734,8 @@ func (h *AgentHandler) AgentLoopStream(c *gin.Context) {
 
 	// 更新助手消息内容
 	if assistantMsg != nil {
-		_, err = h.db.Exec(
-			"UPDATE messages SET content = ?, mcp_execution_ids = ?, updated_at = ? WHERE id = ?",
-			result.Response,
-			func() string {
-				if len(result.MCPExecutionIDs) > 0 {
-					jsonData, _ := json.Marshal(result.MCPExecutionIDs)
-					return string(jsonData)
-				}
-				return ""
-			}(),
-			time.Now(), assistantMessageID,
-		)
-		if err != nil {
-			h.logger.Error("更新助手消息失败", zap.Error(err))
+		if errU := h.db.UpdateAssistantMessageFinalize(assistantMessageID, result.Response, result.MCPExecutionIDs, multiagent.AggregatedReasoningFromTraceJSON(result.LastAgentTraceInput)); errU != nil {
+			h.logger.Error("更新助手消息失败", zap.Error(errU))
 		}
 	} else {
 		// 如果之前创建失败，现在创建
@@ -2664,12 +2659,12 @@ func (h *AgentHandler) executeBatchQueue(queueID string) {
 			var runErr error
 			switch {
 			case useBatchMulti:
-				resultMA, runErr = multiagent.RunDeepAgent(taskCtx, h.config, &h.config.MultiAgent, h.agent, h.logger, conversationID, finalMessage, []agent.ChatMessage{}, roleTools, progressCallback, h.agentsMarkdownDir, batchOrch)
+				resultMA, runErr = multiagent.RunDeepAgent(taskCtx, h.config, &h.config.MultiAgent, h.agent, h.logger, conversationID, finalMessage, []agent.ChatMessage{}, roleTools, progressCallback, h.agentsMarkdownDir, batchOrch, nil)
 			case useEinoSingle:
 				if h.config == nil {
 					runErr = fmt.Errorf("服务器配置未加载")
 				} else {
-					resultMA, runErr = multiagent.RunEinoSingleChatModelAgent(taskCtx, h.config, &h.config.MultiAgent, h.agent, h.logger, conversationID, finalMessage, []agent.ChatMessage{}, roleTools, progressCallback)
+					resultMA, runErr = multiagent.RunEinoSingleChatModelAgent(taskCtx, h.config, &h.config.MultiAgent, h.agent, h.logger, conversationID, finalMessage, []agent.ChatMessage{}, roleTools, progressCallback, nil)
 				}
 			default:
 				result, runErr = h.agent.AgentLoopWithProgress(taskCtx, finalMessage, []agent.ChatMessage{}, conversationID, progressCallback, roleTools)
@@ -2768,17 +2763,7 @@ func (h *AgentHandler) executeBatchQueue(queueID string) {
 
 			// 更新助手消息内容
 			if assistantMessageID != "" {
-				mcpIDsJSON := ""
-				if len(mcpIDs) > 0 {
-					jsonData, _ := json.Marshal(mcpIDs)
-					mcpIDsJSON = string(jsonData)
-				}
-				if _, updateErr := h.db.Exec(
-					"UPDATE messages SET content = ?, mcp_execution_ids = ?, updated_at = ? WHERE id = ?",
-					resText,
-					mcpIDsJSON,
-					time.Now(), assistantMessageID,
-				); updateErr != nil {
+				if updateErr := h.db.UpdateAssistantMessageFinalize(assistantMessageID, resText, mcpIDs, multiagent.AggregatedReasoningFromTraceJSON(lastIn)); updateErr != nil {
 					h.logger.Warn("更新助手消息失败", zap.String("queueId", queueID), zap.String("taskId", task.ID), zap.Error(updateErr))
 					// 如果更新失败，尝试创建新消息
 					_, err = h.db.AddMessage(conversationID, "assistant", resText, mcpIDs)
@@ -2870,6 +2855,10 @@ func (h *AgentHandler) loadHistoryFromAgentTrace(conversationID string) ([]agent
 		if content, ok := msgMap["content"].(string); ok {
 			msg.Content = content
 		}
+		// DeepSeek 思考模式：含工具调用的 assistant 须在后续请求中回传 reasoning_content
+		if rc, ok := msgMap["reasoning_content"].(string); ok && strings.TrimSpace(rc) != "" {
+			msg.ReasoningContent = rc
+		}
 
 		// 解析tool_calls（如果存在）
 		if toolCallsRaw, ok := msgMap["tool_calls"]; ok && toolCallsRaw != nil {
@@ -2975,3 +2964,18 @@ func (h *AgentHandler) loadHistoryFromAgentTrace(conversationID string) ([]agent
 	)
 	return agentMessages, nil
 }
+
+// dbMessagesToAgentChatMessages maps DB rows to agent ChatMessage for history fallback
+// (includes reasoning_content for DeepSeek thinking + tool replay).
+func dbMessagesToAgentChatMessages(msgs []database.Message) []agent.ChatMessage {
+	out := make([]agent.ChatMessage, 0, len(msgs))
+	for i := range msgs {
+		m := msgs[i]
+		out = append(out, agent.ChatMessage{
+			Role:             m.Role,
+			Content:          m.Content,
+			ReasoningContent: m.ReasoningContent,
+		})
+	}
+	return out
+}

internal/handler/config.go

@@ -609,15 +609,46 @@ func (h *ConfigHandler) GetTools(c *gin.Context) {
 
 // UpdateConfigRequest 更新配置请求
 type UpdateConfigRequest struct {
-	OpenAI     *config.OpenAIConfig        `json:"openai,omitempty"`
-	FOFA       *config.FofaConfig          `json:"fofa,omitempty"`
-	MCP        *config.MCPConfig           `json:"mcp,omitempty"`
-	Tools      []ToolEnableStatus          `json:"tools,omitempty"`
-	Agent      *config.AgentConfig         `json:"agent,omitempty"`
-	Knowledge  *config.KnowledgeConfig     `json:"knowledge,omitempty"`
-	Robots     *config.RobotsConfig        `json:"robots,omitempty"`
-	MultiAgent *config.MultiAgentAPIUpdate `json:"multi_agent,omitempty"`
-	C2         *config.C2APIUpdate          `json:"c2,omitempty"`
+	OpenAI     *config.OpenAIConfig         `json:"openai,omitempty"`
+	FOFA       *config.FofaConfig           `json:"fofa,omitempty"`
+	MCP        *config.MCPConfig            `json:"mcp,omitempty"`
+	Tools      []ToolEnableStatus           `json:"tools,omitempty"`
+	Agent      *AgentConfigUpdate           `json:"agent,omitempty"`
+	Knowledge  *config.KnowledgeConfig      `json:"knowledge,omitempty"`
+	Robots     *config.RobotsConfig         `json:"robots,omitempty"`
+	MultiAgent *config.MultiAgentAPIUpdate  `json:"multi_agent,omitempty"`
+	C2         *config.C2APIUpdate           `json:"c2,omitempty"`
+}
+
+// AgentConfigUpdate 用于 PATCH /api/config 的 agent 段：仅 JSON 中出现的字段（指针非 nil）覆盖内存配置。
+// 避免旧版「整包替换 *AgentConfig」时，未传的整型字段被反序列化为 0 误覆盖（例如 tool_timeout_minutes 变成 0）。
+type AgentConfigUpdate struct {
+	MaxIterations        *int    `json:"max_iterations,omitempty"`
+	LargeResultThreshold *int    `json:"large_result_threshold,omitempty"`
+	ResultStorageDir     *string `json:"result_storage_dir,omitempty"`
+	ToolTimeoutMinutes   *int    `json:"tool_timeout_minutes,omitempty"`
+	SystemPromptPath     *string `json:"system_prompt_path,omitempty"`
+}
+
+func applyAgentConfigUpdate(dst *config.AgentConfig, src *AgentConfigUpdate) {
+	if dst == nil || src == nil {
+		return
+	}
+	if src.MaxIterations != nil {
+		dst.MaxIterations = *src.MaxIterations
+	}
+	if src.LargeResultThreshold != nil {
+		dst.LargeResultThreshold = *src.LargeResultThreshold
+	}
+	if src.ResultStorageDir != nil {
+		dst.ResultStorageDir = *src.ResultStorageDir
+	}
+	if src.ToolTimeoutMinutes != nil {
+		dst.ToolTimeoutMinutes = *src.ToolTimeoutMinutes
+	}
+	if src.SystemPromptPath != nil {
+		dst.SystemPromptPath = *src.SystemPromptPath
+	}
 }
 
 // ToolEnableStatus 工具启用状态
@@ -664,12 +695,19 @@ func (h *ConfigHandler) UpdateConfig(c *gin.Context) {
 		)
 	}
 
-	// 更新Agent配置
+	// 更新Agent配置（按字段合并，避免部分 JSON 把未出现的字段写成 0）
 	if req.Agent != nil {
-		h.config.Agent = *req.Agent
+		applyAgentConfigUpdate(&h.config.Agent, req.Agent)
 		h.logger.Info("更新Agent配置",
 			zap.Int("max_iterations", h.config.Agent.MaxIterations),
+			zap.Int("tool_timeout_minutes", h.config.Agent.ToolTimeoutMinutes),
 		)
+		if h.agent != nil && req.Agent.MaxIterations != nil {
+			h.agent.UpdateMaxIterations(h.config.Agent.MaxIterations)
+		}
+		if h.mcpServer != nil {
+			h.mcpServer.ConfigureHTTPToolCallTimeoutFromAgentMinutes(h.config.Agent.ToolTimeoutMinutes)
+		}
 	}
 
 	// 更新Knowledge配置
@@ -1116,6 +1154,9 @@ func (h *ConfigHandler) ApplyConfig(c *gin.Context) {
 		h.agent.UpdateToolDescriptionMode(h.config.Security.ToolDescriptionMode)
 		h.logger.Info("Agent配置已更新")
 	}
+	if h.mcpServer != nil {
+		h.mcpServer.ConfigureHTTPToolCallTimeoutFromAgentMinutes(h.config.Agent.ToolTimeoutMinutes)
+	}
 
 	// 更新AttackChainHandler的OpenAI配置
 	if h.attackChainHandler != nil {
@@ -1181,7 +1222,7 @@ func (h *ConfigHandler) saveConfig() error {
 		return fmt.Errorf("解析配置文件失败: %w", err)
 	}
 
-	updateAgentConfig(root, h.config.Agent.MaxIterations)
+	updateAgentConfig(root, h.config.Agent)
 	updateMCPConfig(root, h.config.MCP)
 	updateOpenAIConfig(root, h.config.OpenAI)
 	updateFOFAConfig(root, h.config.FOFA)
@@ -1286,10 +1327,14 @@ func writeYAMLDocument(path string, doc *yaml.Node) error {
 	return os.WriteFile(path, buf.Bytes(), 0644)
 }
 
-func updateAgentConfig(doc *yaml.Node, maxIterations int) {
+func updateAgentConfig(doc *yaml.Node, agent config.AgentConfig) {
 	root := doc.Content[0]
 	agentNode := ensureMap(root, "agent")
-	setIntInMap(agentNode, "max_iterations", maxIterations)
+	setIntInMap(agentNode, "max_iterations", agent.MaxIterations)
+	setIntInMap(agentNode, "tool_timeout_minutes", agent.ToolTimeoutMinutes)
+	setIntInMap(agentNode, "large_result_threshold", agent.LargeResultThreshold)
+	setStringInMap(agentNode, "result_storage_dir", agent.ResultStorageDir)
+	setStringInMap(agentNode, "system_prompt_path", agent.SystemPromptPath)
 }
 
 func updateMCPConfig(doc *yaml.Node, cfg config.MCPConfig) {
@@ -1312,6 +1357,19 @@ func updateOpenAIConfig(doc *yaml.Node, cfg config.OpenAIConfig) {
 	if cfg.MaxTotalTokens > 0 {
 		setIntInMap(openaiNode, "max_total_tokens", cfg.MaxTotalTokens)
 	}
+	rn := ensureMap(openaiNode, "reasoning")
+	if strings.TrimSpace(cfg.Reasoning.Mode) != "" {
+		setStringInMap(rn, "mode", cfg.Reasoning.Mode)
+	}
+	if strings.TrimSpace(cfg.Reasoning.Effort) != "" {
+		setStringInMap(rn, "effort", cfg.Reasoning.Effort)
+	}
+	if cfg.Reasoning.AllowClientReasoning != nil {
+		setBoolInMap(rn, "allow_client_reasoning", *cfg.Reasoning.AllowClientReasoning)
+	}
+	if strings.TrimSpace(cfg.Reasoning.Profile) != "" {
+		setStringInMap(rn, "profile", cfg.Reasoning.Profile)
+	}
 }
 
 func updateFOFAConfig(doc *yaml.Node, cfg config.FofaConfig) {

internal/handler/eino_single_agent.go

@@ -196,6 +196,7 @@ func (h *AgentHandler) EinoSingleAgentLoopStream(c *gin.Context) {
 			curHistory,
 			roleTools,
 			progressCallback,
+			chatReasoningToClientIntent(req.Reasoning),
 		)
 		timeoutCancel()
 
@@ -297,18 +298,7 @@ func (h *AgentHandler) EinoSingleAgentLoopStream(c *gin.Context) {
 	}
 
 	if assistantMessageID != "" {
-		mcpIDsJSON := ""
-		if len(cumulativeMCPExecutionIDs) > 0 {
-			jsonData, _ := json.Marshal(cumulativeMCPExecutionIDs)
-			mcpIDsJSON = string(jsonData)
-		}
-		_, _ = h.db.Exec(
-			"UPDATE messages SET content = ?, mcp_execution_ids = ?, updated_at = ? WHERE id = ?",
-			result.Response,
-			mcpIDsJSON,
-			time.Now(),
-			assistantMessageID,
-		)
+		_ = h.db.UpdateAssistantMessageFinalize(assistantMessageID, result.Response, cumulativeMCPExecutionIDs, multiagent.AggregatedReasoningFromTraceJSON(result.LastAgentTraceInput))
 	}
 
 	if result.LastAgentTraceInput != "" || result.LastAgentTraceOutput != "" {
@@ -376,6 +366,7 @@ func (h *AgentHandler) EinoSingleAgentLoop(c *gin.Context) {
 		prep.History,
 		prep.RoleTools,
 		progressCallback,
+		chatReasoningToClientIntent(req.Reasoning),
 	)
 	if runErr != nil {
 		if shouldPersistEinoAgentTraceAfterRunError(baseCtx) {
@@ -386,18 +377,7 @@ func (h *AgentHandler) EinoSingleAgentLoop(c *gin.Context) {
 	}
 
 	if prep.AssistantMessageID != "" {
-		mcpIDsJSON := ""
-		if len(result.MCPExecutionIDs) > 0 {
-			jsonData, _ := json.Marshal(result.MCPExecutionIDs)
-			mcpIDsJSON = string(jsonData)
-		}
-		_, _ = h.db.Exec(
-			"UPDATE messages SET content = ?, mcp_execution_ids = ?, updated_at = ? WHERE id = ?",
-			result.Response,
-			mcpIDsJSON,
-			time.Now(),
-			prep.AssistantMessageID,
-		)
+		_ = h.db.UpdateAssistantMessageFinalize(prep.AssistantMessageID, result.Response, result.MCPExecutionIDs, multiagent.AggregatedReasoningFromTraceJSON(result.LastAgentTraceInput))
 	}
 	if result.LastAgentTraceInput != "" || result.LastAgentTraceOutput != "" {
 		_ = h.db.SaveAgentTrace(prep.ConversationID, result.LastAgentTraceInput, result.LastAgentTraceOutput)

internal/handler/multi_agent.go

@@ -208,6 +208,7 @@ func (h *AgentHandler) MultiAgentLoopStream(c *gin.Context) {
 			progressCallback,
 			h.agentsMarkdownDir,
 			orch,
+			chatReasoningToClientIntent(req.Reasoning),
 		)
 		timeoutCancel()
 
@@ -309,18 +310,7 @@ func (h *AgentHandler) MultiAgentLoopStream(c *gin.Context) {
 	}
 
 	if assistantMessageID != "" {
-		mcpIDsJSON := ""
-		if len(cumulativeMCPExecutionIDs) > 0 {
-			jsonData, _ := json.Marshal(cumulativeMCPExecutionIDs)
-			mcpIDsJSON = string(jsonData)
-		}
-		_, _ = h.db.Exec(
-			"UPDATE messages SET content = ?, mcp_execution_ids = ?, updated_at = ? WHERE id = ?",
-			result.Response,
-			mcpIDsJSON,
-			time.Now(),
-			assistantMessageID,
-		)
+		_ = h.db.UpdateAssistantMessageFinalize(assistantMessageID, result.Response, cumulativeMCPExecutionIDs, multiagent.AggregatedReasoningFromTraceJSON(result.LastAgentTraceInput))
 	}
 
 	if result.LastAgentTraceInput != "" || result.LastAgentTraceOutput != "" {
@@ -390,6 +380,7 @@ func (h *AgentHandler) MultiAgentLoop(c *gin.Context) {
 		progressCallback,
 		h.agentsMarkdownDir,
 		strings.TrimSpace(req.Orchestration),
+		chatReasoningToClientIntent(req.Reasoning),
 	)
 	if runErr != nil {
 		if shouldPersistEinoAgentTraceAfterRunError(baseCtx) {
@@ -405,18 +396,7 @@ func (h *AgentHandler) MultiAgentLoop(c *gin.Context) {
 	}
 
 	if prep.AssistantMessageID != "" {
-		mcpIDsJSON := ""
-		if len(result.MCPExecutionIDs) > 0 {
-			jsonData, _ := json.Marshal(result.MCPExecutionIDs)
-			mcpIDsJSON = string(jsonData)
-		}
-		_, _ = h.db.Exec(
-			"UPDATE messages SET content = ?, mcp_execution_ids = ?, updated_at = ? WHERE id = ?",
-			result.Response,
-			mcpIDsJSON,
-			time.Now(),
-			prep.AssistantMessageID,
-		)
+		_ = h.db.UpdateAssistantMessageFinalize(prep.AssistantMessageID, result.Response, result.MCPExecutionIDs, multiagent.AggregatedReasoningFromTraceJSON(result.LastAgentTraceInput))
 	}
 
 	if result.LastAgentTraceInput != "" || result.LastAgentTraceOutput != "" {

internal/handler/multi_agent_prepare.go

@@ -55,13 +55,7 @@ func (h *AgentHandler) prepareMultiAgentSession(req *ChatRequest) (*multiAgentPr
 		if getErr != nil {
 			agentHistoryMessages = []agent.ChatMessage{}
 		} else {
-			agentHistoryMessages = make([]agent.ChatMessage, 0, len(historyMessages))
-			for _, msg := range historyMessages {
-				agentHistoryMessages = append(agentHistoryMessages, agent.ChatMessage{
-					Role:    msg.Role,
-					Content: msg.Content,
-				})
-			}
+			agentHistoryMessages = dbMessagesToAgentChatMessages(historyMessages)
 		}
 	}
 

internal/mcp/server.go

@@ -44,6 +44,10 @@ type Server struct {
 	runningCancels        map[string]context.CancelFunc
 	runningCancelsMu      sync.Mutex
 	abortUserNotes        map[string]string // 监控页终止时附带的用户说明，与 executionID 对应
+	// httpToolTimeoutMinutes 同步 agent.tool_timeout_minutes，用于 POST /api/mcp 的 tools/call（不经 Agent 包装的路径）。
+	// nil 表示未配置，沿用默认 30 分钟；指向 0 表示不限制；>0 为分钟数。
+	httpToolTimeoutMinutes *int
+	httpToolTimeoutMu      sync.RWMutex
 }
 
 type sseClient struct {
@@ -90,6 +94,39 @@ func NewServerWithStorage(logger *zap.Logger, storage MonitorStorage) *Server {
 	return s
 }
 
+// ConfigureHTTPToolCallTimeoutFromAgentMinutes 将 agent.tool_timeout_minutes 同步到经 HTTP POST /api/mcp 触发的 tools/call。
+// minutes<=0 表示不设置硬性截止时间（与配置「0 不限制」一致）；minutes>0 为该次调用的最长等待时间。
+// 未调用前对 tools/call 使用默认 30 分钟（与历史硬编码一致）。
+func (s *Server) ConfigureHTTPToolCallTimeoutFromAgentMinutes(minutes int) {
+	if s == nil {
+		return
+	}
+	v := minutes
+	if v < 0 {
+		v = 0
+	}
+	s.httpToolTimeoutMu.Lock()
+	defer s.httpToolTimeoutMu.Unlock()
+	s.httpToolTimeoutMinutes = &v
+}
+
+func (s *Server) effectiveHTTPToolCallDeadline() (context.Context, context.CancelFunc) {
+	const defaultDur = 30 * time.Minute
+	if s == nil {
+		return context.WithTimeout(context.Background(), defaultDur)
+	}
+	s.httpToolTimeoutMu.RLock()
+	mPtr := s.httpToolTimeoutMinutes
+	s.httpToolTimeoutMu.RUnlock()
+	if mPtr == nil {
+		return context.WithTimeout(context.Background(), defaultDur)
+	}
+	if *mPtr <= 0 {
+		return context.WithCancel(context.Background())
+	}
+	return context.WithTimeout(context.Background(), time.Duration(*mPtr)*time.Minute)
+}
+
 // RegisterTool 注册工具
 func (s *Server) RegisterTool(tool Tool, handler ToolHandler) {
 	s.mu.Lock()
@@ -457,7 +494,7 @@ func (s *Server) handleCallTool(msg *Message) *Message {
 		}
 	}
 
-	baseCtx, timeoutCancel := context.WithTimeout(context.Background(), 30*time.Minute)
+	baseCtx, timeoutCancel := s.effectiveHTTPToolCallDeadline()
 	defer timeoutCancel()
 	execCtx, runCancel := context.WithCancel(baseCtx)
 	s.registerRunningCancel(executionID, runCancel)

internal/multiagent/eino_adk_run_loop.go

@@ -14,7 +14,10 @@ import (
 	"unicode/utf8"
 
 	"cyberstrike-ai/internal/agent"
+	"cyberstrike-ai/internal/config"
+	"cyberstrike-ai/internal/einoobserve"
 	"cyberstrike-ai/internal/einomcp"
+	"cyberstrike-ai/internal/openai"
 
 	"github.com/cloudwego/eino/adk"
 	"github.com/cloudwego/eino/schema"
@@ -94,6 +97,9 @@ type einoADKRunLoopArgs struct {
 	// ModelFacingTrace 可选：由各 ChatModelAgent Handlers 链末尾中间件写入「即将送入模型」的消息快照；
 	// 非空时优先用于 LastAgentTraceInput 序列化，使续跑与 summarization/reduction 后的上下文一致。
 	ModelFacingTrace *modelFacingTraceHolder
+
+	// EinoCallbacks 可选：为 ADK Runner 注入 eino [callbacks] 全链路观测（见 internal/einoobserve）。
+	EinoCallbacks *config.MultiAgentEinoCallbacksConfig
 }
 
 func runEinoADKAgentLoop(ctx context.Context, args *einoADKRunLoopArgs, baseMsgs []adk.Message) (*RunResult, error) {
@@ -288,6 +294,16 @@ func runEinoADKAgentLoop(ctx context.Context, args *einoADKRunLoopArgs, baseMsgs
 		})
 	}
 
+	if args.EinoCallbacks != nil {
+		ctx = einoobserve.AttachAgentRunCallbacks(ctx, args.EinoCallbacks, einoobserve.Params{
+			Logger:           logger,
+			Progress:         progress,
+			ConversationID:   conversationID,
+			OrchMode:         orchMode,
+			OrchestratorName: orchestratorName,
+		})
+	}
+
 	runnerCfg := adk.RunnerConfig{
 		Agent:           da,
 		EnableStreaming: true,
@@ -550,6 +566,7 @@ func runEinoADKAgentLoop(ctx context.Context, args *einoADKRunLoopArgs, baseMsgs
 			var mainAssistantBuf string
 			var mainAssistDupTarget string // 非空表示本段主助手流需缓冲至 EOF，与 execute 输出比对去重
 			var reasoningBuf string
+			var prevReasoningDisplay string // UI 用：剥离 Claude 内部 signature 尾缀后的累计展示
 			var streamRecvErr error
 			type streamMsg struct {
 				chunk *schema.Message
@@ -597,19 +614,29 @@ func runEinoADKAgentLoop(ctx context.Context, args *einoADKRunLoopArgs, baseMsgs
 						var reasoningDelta string
 						reasoningBuf, reasoningDelta = normalizeStreamingDelta(reasoningBuf, chunk.ReasoningContent)
 						if reasoningDelta != "" {
-							if reasoningStreamID == "" {
-								reasoningStreamID = fmt.Sprintf("eino-reasoning-%s-%d", conversationID, atomic.AddInt64(&reasoningStreamSeq, 1))
-								progress("thinking_stream_start", " ", map[string]interface{}{
-									"streamId":      reasoningStreamID,
-									"source":        "eino",
-									"einoAgent":     ev.AgentName,
-									"einoRole":      einoRoleTag(ev.AgentName),
-									"orchestration": orchMode,
+							fullDisplay := openai.DisplayReasoningContent(reasoningBuf)
+							var displayDelta string
+							if strings.HasPrefix(fullDisplay, prevReasoningDisplay) {
+								displayDelta = fullDisplay[len(prevReasoningDisplay):]
+							} else {
+								displayDelta = fullDisplay
+							}
+							prevReasoningDisplay = fullDisplay
+							if displayDelta != "" {
+								if reasoningStreamID == "" {
+									reasoningStreamID = fmt.Sprintf("eino-reasoning-%s-%d", conversationID, atomic.AddInt64(&reasoningStreamSeq, 1))
+									progress("reasoning_chain_stream_start", " ", map[string]interface{}{
+										"streamId":      reasoningStreamID,
+										"source":        "eino",
+										"einoAgent":     ev.AgentName,
+										"einoRole":      einoRoleTag(ev.AgentName),
+										"orchestration": orchMode,
+									})
+								}
+								progress("reasoning_chain_stream_delta", displayDelta, map[string]interface{}{
+									"streamId": reasoningStreamID,
 								})
 							}
-							progress("thinking_stream_delta", reasoningDelta, map[string]interface{}{
-								"streamId": reasoningStreamID,
-							})
 						}
 					}
 					if chunk.Content != "" {
@@ -777,7 +804,7 @@ func runEinoADKAgentLoop(ctx context.Context, args *einoADKRunLoopArgs, baseMsgs
 
 		if mv.Role == schema.Assistant {
 			if progress != nil && strings.TrimSpace(msg.ReasoningContent) != "" {
-				progress("thinking", strings.TrimSpace(msg.ReasoningContent), map[string]interface{}{
+				progress("reasoning_chain", openai.DisplayReasoningContent(strings.TrimSpace(msg.ReasoningContent)), map[string]interface{}{
 					"conversationId": conversationID,
 					"source":         "eino",
 					"einoAgent":      ev.AgentName,

internal/multiagent/eino_execute_streaming_wrap.go

@@ -22,6 +22,9 @@ import (
 //
 // 使用 Pipe 将内层流转发给调用方：在 inner EOF 后、关闭 Pipe 前同步调用 ToolInvokeNotify.Fire，
 // 保证 run loop 在模型开始下一轮输出前已记录 execute 结果（用于 UI 与「重复助手复述」去重）。
+//
+// 若 inner 在校验阶段直接返回 error（未建立 reader），不会进入下方 goroutine，也必须 Fire；
+// 否则 pending tool_call 要等整轮 run 结束才被 force-close，与已展示的助手/工具软错误文案不同步。
 type einoStreamingShellWrap struct {
 	inner         filesystem.StreamingShell
 	invokeNotify  *einomcp.ToolInvokeNotifyHolder
@@ -42,17 +45,24 @@ func (w *einoStreamingShellWrap) ExecuteStreaming(ctx context.Context, input *fi
 	if security.IsBackgroundShellCommand(req.Command) && !req.RunInBackendGround {
 		req.RunInBackendGround = true
 	}
+	tid := strings.TrimSpace(compose.GetToolCallID(ctx))
+	agentTag := strings.TrimSpace(w.einoAgentName)
+
 	sr, err := w.inner.ExecuteStreaming(ctx, &req)
 	if err != nil {
+		if w.recordMonitor != nil {
+			w.recordMonitor(cmd, "", false, err)
+		}
+		if w.invokeNotify != nil && tid != "" {
+			w.invokeNotify.Fire(tid, "execute", agentTag, false, "", err)
+		}
 		return nil, err
 	}
-	tid := strings.TrimSpace(compose.GetToolCallID(ctx))
 	if sr == nil || w.invokeNotify == nil || tid == "" {
 		return sr, nil
 	}
 
 	outR, outW := schema.Pipe[*filesystem.ExecuteResponse](32)
-	agentTag := strings.TrimSpace(w.einoAgentName)
 
 	go func(inner *schema.StreamReader[*filesystem.ExecuteResponse], command string) {
 		defer inner.Close()

internal/multiagent/eino_middleware.go

@@ -161,6 +161,8 @@ func buildReductionMiddleware(ctx context.Context, mw config.MultiAgentEinoMiddl
 }
 
 // prependEinoMiddlewares returns handlers to prepend (outermost first) and optionally replaces tools when tool_search is used.
+// toolSearchActive is true when the toolsearch middleware was mounted (dynamic tools split off); callers should pass this to
+// injectToolNamesOnlyInstruction — tool_search is not part of the pre-middleware tools list, so name-scanning alone cannot detect it.
 func prependEinoMiddlewares(
 	ctx context.Context,
 	mw *config.MultiAgentEinoMiddlewareConfig,
@@ -170,16 +172,16 @@ func prependEinoMiddlewares(
 	skillsRoot string,
 	conversationID string,
 	logger *zap.Logger,
-) (outTools []tool.BaseTool, extraHandlers []adk.ChatModelAgentMiddleware, err error) {
+) (outTools []tool.BaseTool, extraHandlers []adk.ChatModelAgentMiddleware, toolSearchActive bool, err error) {
 	if mw == nil {
-		return tools, nil, nil
+		return tools, nil, false, nil
 	}
 	outTools = tools
 
 	if mw.PatchToolCallsEffective() {
 		patchMW, perr := patchtoolcalls.New(ctx, &patchtoolcalls.Config{})
 		if perr != nil {
-			return nil, nil, fmt.Errorf("patchtoolcalls: %w", perr)
+			return nil, nil, false, fmt.Errorf("patchtoolcalls: %w", perr)
 		}
 		extraHandlers = append(extraHandlers, patchMW)
 	}
@@ -190,7 +192,7 @@ func prependEinoMiddlewares(
 		} else {
 			redMW, rerr := buildReductionMiddleware(ctx, *mw, conversationID, einoLoc, logger)
 			if rerr != nil {
-				return nil, nil, rerr
+				return nil, nil, false, rerr
 			}
 			extraHandlers = append(extraHandlers, redMW)
 		}
@@ -209,10 +211,11 @@ func prependEinoMiddlewares(
 		if split && len(dynamic) > 0 {
 			ts, terr := toolsearch.New(ctx, &toolsearch.Config{DynamicTools: dynamic})
 			if terr != nil {
-				return nil, nil, fmt.Errorf("toolsearch: %w", terr)
+				return nil, nil, false, fmt.Errorf("toolsearch: %w", terr)
 			}
 			extraHandlers = append(extraHandlers, ts)
 			outTools = static
+			toolSearchActive = true
 			if logger != nil {
 				logger.Info("eino middleware: tool_search enabled",
 					zap.Int("static_tools", len(static)),
@@ -233,12 +236,12 @@ func prependEinoMiddlewares(
 			}
 			baseDir := filepath.Join(skillsRoot, rel, sanitizeEinoPathSegment(conversationID))
 			if mk := os.MkdirAll(baseDir, 0o755); mk != nil {
-				return nil, nil, fmt.Errorf("plantask mkdir: %w", mk)
+				return nil, nil, toolSearchActive, fmt.Errorf("plantask mkdir: %w", mk)
 			}
 			ptBE := &localPlantaskBackend{Local: einoLoc}
 			pt, perr := plantask.New(ctx, &plantask.Config{Backend: ptBE, BaseDir: baseDir})
 			if perr != nil {
-				return nil, nil, fmt.Errorf("plantask: %w", perr)
+				return nil, nil, toolSearchActive, fmt.Errorf("plantask: %w", perr)
 			}
 			extraHandlers = append(extraHandlers, pt)
 			if logger != nil {
@@ -247,7 +250,7 @@ func prependEinoMiddlewares(
 		}
 	}
 
-	return outTools, extraHandlers, nil
+	return outTools, extraHandlers, toolSearchActive, nil
 }
 
 func deepExtrasFromConfig(ma *config.MultiAgentConfig) (outputKey string, retry *adk.ModelRetryConfig, taskDesc func(context.Context, []adk.Agent) (string, error)) {

internal/multiagent/eino_single_runner.go

@@ -13,6 +13,7 @@ import (
 	"cyberstrike-ai/internal/config"
 	"cyberstrike-ai/internal/einomcp"
 	"cyberstrike-ai/internal/openai"
+	"cyberstrike-ai/internal/reasoning"
 
 	einoopenai "github.com/cloudwego/eino-ext/components/model/openai"
 	"github.com/cloudwego/eino/adk"
@@ -37,6 +38,7 @@ func RunEinoSingleChatModelAgent(
 	history []agent.ChatMessage,
 	roleTools []string,
 	progress func(eventType, message string, data interface{}),
+	reasoningClient *reasoning.ClientIntent,
 ) (*RunResult, error) {
 	if appCfg == nil || ag == nil {
 		return nil, fmt.Errorf("eino single: 配置或 Agent 为空")
@@ -94,7 +96,7 @@ func RunEinoSingleChatModelAgent(
 		return nil, err
 	}
 
-	mainToolsForCfg, mainOrchestratorPre, err := prependEinoMiddlewares(ctx, &ma.EinoMiddleware, einoMWMain, mainTools, einoLoc, skillsRoot, conversationID, logger)
+	mainToolsForCfg, mainOrchestratorPre, singleToolSearchActive, err := prependEinoMiddlewares(ctx, &ma.EinoMiddleware, einoMWMain, mainTools, einoLoc, skillsRoot, conversationID, logger)
 	if err != nil {
 		return nil, fmt.Errorf("eino single eino 中间件: %w", err)
 	}
@@ -121,6 +123,7 @@ func RunEinoSingleChatModelAgent(
 		Model:      appCfg.OpenAI.Model,
 		HTTPClient: httpClient,
 	}
+	reasoning.ApplyToEinoChatModelConfig(baseModelCfg, &appCfg.OpenAI, reasoningClient)
 
 	mainModel, err := einoopenai.NewChatModel(ctx, baseModelCfg)
 	if err != nil {
@@ -170,27 +173,20 @@ func RunEinoSingleChatModelAgent(
 			UnknownToolsHandler: einomcp.UnknownToolReminderHandler(),
 			ToolCallMiddlewares: []compose.ToolMiddleware{
 				hitlToolCallMiddleware(),
-				{Invokable: softRecoveryToolCallMiddleware()},
+				softRecoveryToolMiddleware(),
 			},
 		},
 		EmitInternalEvents: true,
 	}
-	ins := injectToolNamesOnlyInstruction(ctx, ag.EinoSingleAgentSystemInstruction(), mainTools)
+	ins := injectToolNamesOnlyInstruction(ctx, ag.EinoSingleAgentSystemInstruction(), mainTools, singleToolSearchActive)
 	if logger != nil {
 		names := collectToolNames(ctx, mainTools)
 		mountedNames := collectToolNames(ctx, mainToolsForCfg)
-		hasToolSearch := false
-		for _, n := range names {
-			if strings.EqualFold(strings.TrimSpace(n), "tool_search") {
-				hasToolSearch = true
-				break
-			}
-		}
 		logger.Info("eino tool-name injection",
 			zap.String("scope", "eino_single"),
 			zap.Int("tool_names", len(names)),
 			zap.Int("mounted_tool_names", len(mountedNames)),
-			zap.Bool("has_tool_search", hasToolSearch),
+			zap.Bool("tool_search_middleware", singleToolSearchActive),
 		)
 	}
 
@@ -244,6 +240,7 @@ func RunEinoSingleChatModelAgent(
 		ToolInvokeNotify:        toolInvokeNotify,
 		DA:                      chatAgent,
 		ModelFacingTrace:        modelFacingTrace,
+		EinoCallbacks:           &ma.EinoCallbacks,
 		EmptyResponseMessage: "(Eino ADK single-agent session completed but no assistant text was captured. Check process details or logs.) " +
 			"（Eino ADK 单代理会话已完成，但未捕获到助手文本输出。请查看过程详情或日志。）",
 	}, baseMsgs)

internal/multiagent/eino_summarize.go

@@ -214,7 +214,7 @@ func summarizeFinalizeWithRecentAssistantToolTrail(
 		selectedCount++
 	}
 
-	// 还原时间顺序
+	// 还原时间顺序。round 内为原始 *schema.Message 指针，保留 ReasoningContent（DeepSeek 工具续跑所必需）。
 	selectedMsgs := make([]adk.Message, 0, 8)
 	for i := len(selectedRoundsReverse) - 1; i >= 0; i-- {
 		selectedMsgs = append(selectedMsgs, selectedRoundsReverse[i].messages...)

internal/multiagent/eino_tool_name_injection.go

@@ -9,34 +9,43 @@ import (
 
 // injectToolNamesOnlyInstruction prepends a compact tool-name-only section into
 // the system instruction so the model can reference current callable names.
-func injectToolNamesOnlyInstruction(ctx context.Context, instruction string, tools []tool.BaseTool) string {
+// toolSearchMiddlewareActive must be true when prependEinoMiddlewares mounted toolsearch (dynamic tools); do not infer this
+// by scanning tool names — tool_search is injected by middleware and is usually absent from the pre-split tools list.
+func injectToolNamesOnlyInstruction(ctx context.Context, instruction string, tools []tool.BaseTool, toolSearchMiddlewareActive bool) string {
 	names := collectToolNames(ctx, tools)
 	if len(names) == 0 {
 		return strings.TrimSpace(instruction)
 	}
-	hasToolSearch := false
-	for _, n := range names {
-		if strings.EqualFold(strings.TrimSpace(n), "tool_search") {
-			hasToolSearch = true
-			break
+	hasToolSearch := toolSearchMiddlewareActive
+	if !hasToolSearch {
+		for _, n := range names {
+			if strings.EqualFold(strings.TrimSpace(n), "tool_search") {
+				hasToolSearch = true
+				break
+			}
 		}
 	}
 
 	var sb strings.Builder
-	sb.WriteString("以下是当前会话中可调用的工具名称列表（仅名称，无参数定义）：\n")
+	sb.WriteString("以下是当前会话绑定的工具名称索引（仅名称，无参数 JSON Schema）。\n")
+	sb.WriteString("说明：若启用了 tool_search，则列表里可能含「非常驻」工具——它们不一定出现在当前轮次下发给模型的工具定义中；在未看到该工具的完整 schema 前，禁止凭名称臆测参数。\n")
 	for _, name := range names {
 		sb.WriteString("- ")
 		sb.WriteString(name)
 		sb.WriteByte('\n')
 	}
 	sb.WriteString("\n使用规则：\n")
-	sb.WriteString("1) 上述仅为名称列表，不包含参数定义。\n")
+	sb.WriteString("1) 上表仅为名称索引，不含参数定义。禁止猜测参数名、类型、枚举取值或是否必填。\n")
 	if hasToolSearch {
-		sb.WriteString("2) 在调用具体工具前，应先使用 tool_search 查看工具详情与参数要求，再发起调用。\n")
+		sb.WriteString("【强制 / 最高优先级】本会话已启用 tool_search（动态工具池）。凡名称索引里出现、但你在「当前请求所附 tools 定义」中看不到其完整参数 schema 的工具，一律必须先调用 tool_search；为省 token 或赶进度而跳过 tool_search、直接调用业务工具，属于明确禁止的错误流程。\n")
+		sb.WriteString("2) 默认策略：只要对目标工具的参数定义有任何不确定，就先 tool_search；宁可多一次 tool_search，也不要在未见 schema 时盲调业务工具。\n")
+		sb.WriteString("3) 调用顺序：先 tool_search（唯一必填参数 regex_pattern：按工具名匹配的正则，如子串 nuclei 或 ^exact_tool_name$）→ 在后续轮次确认目标工具已出现在 tools 列表且已阅读其 schema → 再发起对该工具的真实调用。\n")
+		sb.WriteString("4) tool_search 的返回仅为匹配到的工具名列表；schema 在解锁后的下一轮才会下发。禁止在 schema 未出现时编造 JSON 参数。\n")
+		sb.WriteString("5) 不要臆造不存在的工具名。\n\n")
 	} else {
-		sb.WriteString("2) 调用具体工具前，请先确认该工具的参数要求；不确定时先澄清再调用。\n")
+		sb.WriteString("2) 调用具体工具前，请先确认该工具的参数要求（以当前请求中的工具定义为准）；不确定时先澄清再调用。\n")
+		sb.WriteString("3) 不要臆造不存在的工具名。\n\n")
 	}
-	sb.WriteString("3) 不要臆造不存在的工具名。\n\n")
 	if s := strings.TrimSpace(instruction); s != "" {
 		sb.WriteString(s)
 	}

internal/multiagent/reasoning_trace.go

@@ -0,0 +1,52 @@
+package multiagent
+
+import (
+	"encoding/json"
+	"fmt"
+	"strings"
+)
+
+// AggregatedReasoningFromTraceJSON concatenates non-empty assistant `reasoning_content`
+// fields from last_react-style JSON (slice of message objects) in document order.
+// Used to persist on the single assistant bubble row for audit and for GetMessages fallback
+// when the full trace JSON is unavailable. For strict per-message replay, prefer last_react_input.
+func AggregatedReasoningFromTraceJSON(traceJSON string) string {
+	traceJSON = strings.TrimSpace(traceJSON)
+	if traceJSON == "" {
+		return ""
+	}
+	var arr []map[string]interface{}
+	if err := json.Unmarshal([]byte(traceJSON), &arr); err != nil {
+		return ""
+	}
+	var b strings.Builder
+	for _, m := range arr {
+		role, _ := m["role"].(string)
+		if !strings.EqualFold(strings.TrimSpace(role), "assistant") {
+			continue
+		}
+		rc := reasoningContentFromMessageMap(m)
+		if rc == "" {
+			continue
+		}
+		if b.Len() > 0 {
+			b.WriteByte('\n')
+		}
+		b.WriteString(rc)
+	}
+	return b.String()
+}
+
+func reasoningContentFromMessageMap(m map[string]interface{}) string {
+	if m == nil {
+		return ""
+	}
+	switch v := m["reasoning_content"].(type) {
+	case string:
+		return strings.TrimSpace(v)
+	case nil:
+		return ""
+	default:
+		return strings.TrimSpace(fmt.Sprint(v))
+	}
+}

internal/multiagent/reasoning_trace_test.go

@@ -0,0 +1,20 @@
+package multiagent
+
+import "testing"
+
+func TestAggregatedReasoningFromTraceJSON(t *testing.T) {
+	const j = `[
+{"role":"user","content":"hi"},
+{"role":"assistant","content":"c1","reasoning_content":"r1","tool_calls":[{"id":"1","type":"function","function":{"name":"f","arguments":"{}"}}]},
+{"role":"tool","tool_call_id":"1","content":"out"},
+{"role":"assistant","content":"c2","reasoning_content":"r2"}
+]`
+	got := AggregatedReasoningFromTraceJSON(j)
+	want := "r1\nr2"
+	if got != want {
+		t.Fatalf("got %q want %q", got, want)
+	}
+	if AggregatedReasoningFromTraceJSON("") != "" || AggregatedReasoningFromTraceJSON("[]") != "" {
+		t.Fatal("empty expected")
+	}
+}

internal/multiagent/runner.go

@@ -17,6 +17,7 @@ import (
 	"cyberstrike-ai/internal/config"
 	"cyberstrike-ai/internal/einomcp"
 	"cyberstrike-ai/internal/openai"
+	"cyberstrike-ai/internal/reasoning"
 
 	einoopenai "github.com/cloudwego/eino-ext/components/model/openai"
 	"github.com/cloudwego/eino/adk"
@@ -48,6 +49,7 @@ type toolCallPendingInfo struct {
 
 // RunDeepAgent 使用 Eino 多代理预置编排执行一轮对话（deep / plan_execute / supervisor；流式事件通过 progress 回调输出）。
 // orchestrationOverride 非空时优先（如聊天/WebShell 请求体）；否则用 multi_agent.orchestration（遗留 yaml）；皆空则按 deep。
+// reasoningClient 来自 ChatRequest.reasoning；可为 nil（机器人/批量等走全局 openai.reasoning）。
 func RunDeepAgent(
 	ctx context.Context,
 	appCfg *config.Config,
@@ -61,6 +63,7 @@ func RunDeepAgent(
 	progress func(eventType, message string, data interface{}),
 	agentsMarkdownDir string,
 	orchestrationOverride string,
+	reasoningClient *reasoning.ClientIntent,
 ) (*RunResult, error) {
 	if appCfg == nil || ma == nil || ag == nil {
 		return nil, fmt.Errorf("multiagent: 配置或 Agent 为空")
@@ -163,6 +166,7 @@ func RunDeepAgent(
 		Model:      appCfg.OpenAI.Model,
 		HTTPClient: httpClient,
 	}
+	reasoning.ApplyToEinoChatModelConfig(baseModelCfg, &appCfg.OpenAI, reasoningClient)
 
 	deepMaxIter := ma.MaxIteration
 	if deepMaxIter <= 0 {
@@ -219,7 +223,7 @@ func RunDeepAgent(
 				return nil, fmt.Errorf("子代理 %q 工具: %w", id, err)
 			}
 
-			subToolsForCfg, subPre, err := prependEinoMiddlewares(ctx, &ma.EinoMiddleware, einoMWSub, subTools, einoLoc, skillsRoot, conversationID, logger)
+			subToolsForCfg, subPre, subToolSearchActive, err := prependEinoMiddlewares(ctx, &ma.EinoMiddleware, einoMWSub, subTools, einoLoc, skillsRoot, conversationID, logger)
 			if err != nil {
 				return nil, fmt.Errorf("子代理 %q eino 中间件: %w", id, err)
 			}
@@ -256,23 +260,16 @@ func RunDeepAgent(
 				subHandlers = append(subHandlers, teleMw)
 			}
 
-			subInstrFinal := injectToolNamesOnlyInstruction(ctx, instr, subTools)
+			subInstrFinal := injectToolNamesOnlyInstruction(ctx, instr, subTools, subToolSearchActive)
 			if logger != nil {
 				subNames := collectToolNames(ctx, subTools)
 				mountedNames := collectToolNames(ctx, subToolsForCfg)
-				hasToolSearch := false
-				for _, n := range subNames {
-					if strings.EqualFold(strings.TrimSpace(n), "tool_search") {
-						hasToolSearch = true
-						break
-					}
-				}
 				logger.Info("eino tool-name injection",
 					zap.String("scope", "sub_agent"),
 					zap.String("agent", id),
 					zap.Int("tool_names", len(subNames)),
 					zap.Int("mounted_tool_names", len(mountedNames)),
-					zap.Bool("has_tool_search", hasToolSearch),
+					zap.Bool("tool_search_middleware", subToolSearchActive),
 				)
 			}
 			sa, err := adk.NewChatModelAgent(ctx, &adk.ChatModelAgentConfig{
@@ -286,7 +283,7 @@ func RunDeepAgent(
 						UnknownToolsHandler: einomcp.UnknownToolReminderHandler(),
 						ToolCallMiddlewares: []compose.ToolMiddleware{
 							hitlToolCallMiddleware(),
-							{Invokable: softRecoveryToolCallMiddleware()},
+							softRecoveryToolMiddleware(),
 						},
 					},
 					EmitInternalEvents: true,
@@ -337,28 +334,21 @@ func RunDeepAgent(
 	if err != nil {
 		return nil, err
 	}
-	mainToolsForCfg, mainOrchestratorPre, err := prependEinoMiddlewares(ctx, &ma.EinoMiddleware, einoMWMain, mainTools, einoLoc, skillsRoot, conversationID, logger)
+	mainToolsForCfg, mainOrchestratorPre, mainToolSearchActive, err := prependEinoMiddlewares(ctx, &ma.EinoMiddleware, einoMWMain, mainTools, einoLoc, skillsRoot, conversationID, logger)
 	if err != nil {
 		return nil, err
 	}
 
-	orchInstruction = injectToolNamesOnlyInstruction(ctx, orchInstruction, mainTools)
+	orchInstruction = injectToolNamesOnlyInstruction(ctx, orchInstruction, mainTools, mainToolSearchActive)
 	if logger != nil {
 		mainNames := collectToolNames(ctx, mainTools)
 		mountedNames := collectToolNames(ctx, mainToolsForCfg)
-		hasToolSearch := false
-		for _, n := range mainNames {
-			if strings.EqualFold(strings.TrimSpace(n), "tool_search") {
-				hasToolSearch = true
-				break
-			}
-		}
 		logger.Info("eino tool-name injection",
 			zap.String("scope", "orchestrator"),
 			zap.String("orchestration", orchMode),
 			zap.Int("tool_names", len(mainNames)),
 			zap.Int("mounted_tool_names", len(mountedNames)),
-			zap.Bool("has_tool_search", hasToolSearch),
+			zap.Bool("tool_search_middleware", mainToolSearchActive),
 		)
 	}
 
@@ -435,7 +425,7 @@ func RunDeepAgent(
 			UnknownToolsHandler: einomcp.UnknownToolReminderHandler(),
 			ToolCallMiddlewares: []compose.ToolMiddleware{
 				hitlToolCallMiddleware(),
-				{Invokable: softRecoveryToolCallMiddleware()},
+				softRecoveryToolMiddleware(),
 			},
 		},
 		EmitInternalEvents: true,
@@ -581,6 +571,7 @@ func RunDeepAgent(
 		ToolInvokeNotify:        toolInvokeNotify,
 		DA:                      da,
 		ModelFacingTrace:        modelFacingTrace,
+		EinoCallbacks:           &ma.EinoCallbacks,
 		EmptyResponseMessage: "(Eino multi-agent orchestration completed but no assistant text was captured. Check process details or logs.) " +
 			"（Eino 多代理编排已完成，但未捕获到助手文本输出。请查看过程详情或日志。）",
 	}, baseMsgs)
@@ -636,8 +627,13 @@ func historyToMessages(history []agent.ChatMessage, appCfg *config.Config, mwCfg
 			}
 		case "assistant":
 			toolSchema := chatToolCallsToSchema(h.ToolCalls)
-			if len(toolSchema) > 0 || strings.TrimSpace(h.Content) != "" {
-				raw = append(raw, schema.AssistantMessage(h.Content, toolSchema))
+			hasRC := strings.TrimSpace(h.ReasoningContent) != ""
+			if len(toolSchema) > 0 || strings.TrimSpace(h.Content) != "" || hasRC {
+				am := schema.AssistantMessage(h.Content, toolSchema)
+				if hasRC {
+					am.ReasoningContent = strings.TrimSpace(h.ReasoningContent)
+				}
+				raw = append(raw, am)
 			}
 		case "tool":
 			if strings.TrimSpace(h.ToolCallID) == "" && strings.TrimSpace(h.Content) == "" {

internal/multiagent/runner_reasoning_history_test.go

@@ -0,0 +1,22 @@
+package multiagent
+
+import (
+	"testing"
+
+	"cyberstrike-ai/internal/agent"
+)
+
+func TestHistoryToMessagesPreservesReasoningContent(t *testing.T) {
+	h := []agent.ChatMessage{
+		{Role: "user", Content: "u"},
+		{Role: "assistant", Content: "c", ReasoningContent: "r1", ToolCalls: []agent.ToolCall{{ID: "t1", Type: "function", Function: agent.FunctionCall{Name: "f", Arguments: map[string]interface{}{}}}}},
+	}
+	msgs := historyToMessages(h, nil, nil)
+	if len(msgs) != 2 {
+		t.Fatalf("len=%d", len(msgs))
+	}
+	am := msgs[1]
+	if am.ReasoningContent != "r1" || am.Content != "c" {
+		t.Fatalf("got reasoning=%q content=%q", am.ReasoningContent, am.Content)
+	}
+}

internal/multiagent/tool_error_middleware.go

@@ -8,6 +8,7 @@ import (
 	"strings"
 
 	"github.com/cloudwego/eino/compose"
+	"github.com/cloudwego/eino/schema"
 )
 
 // softRecoveryToolCallMiddleware returns an InvokableToolMiddleware that catches
@@ -16,8 +17,9 @@ import (
 // returned to the LLM. This allows the model to self-correct within the same
 // iteration rather than crashing the entire graph and requiring a full replay.
 //
-// Without this middleware, a JSON parse failure in any tool's InvokableRun propagates
-// as a hard error through the Eino ToolsNode → [NodeRunError] → ev.Err, which
+// Without Invokable (+ Streamable where applicable) registration, a JSON parse failure
+// in InvokableRun / StreamableRun propagates as a hard error through the Eino ToolsNode
+// → [NodeRunError] → ev.Err, which
 // either triggers the full-replay retry loop (expensive) or terminates the run
 // entirely once retries are exhausted. With it, the LLM simply sees an error message
 // in the tool result and can adjust its next tool call accordingly.
@@ -39,6 +41,44 @@ func softRecoveryToolCallMiddleware() compose.InvokableToolMiddleware {
 	}
 }
 
+// softRecoveryStreamableToolCallMiddleware mirrors softRecoveryToolCallMiddleware for
+// tools that implement StreamableTool only (e.g. Eino ADK filesystem execute).
+// Eino applies Invokable vs Streamable middleware to disjoint code paths in ToolsNode;
+// registering only Invokable leaves streaming tools uncovered — empty/malformed JSON
+// then fails inside [LocalStreamFunc] before the inner endpoint runs.
+func softRecoveryStreamableToolCallMiddleware() compose.StreamableToolMiddleware {
+	return func(next compose.StreamableToolEndpoint) compose.StreamableToolEndpoint {
+		return func(ctx context.Context, input *compose.ToolInput) (*compose.StreamToolOutput, error) {
+			out, err := next(ctx, input)
+			if err == nil {
+				return out, nil
+			}
+			if !isSoftRecoverableToolError(err) {
+				return out, err
+			}
+			toolName := ""
+			args := ""
+			if input != nil {
+				toolName = input.Name
+				args = input.Arguments
+			}
+			msg := buildSoftRecoveryMessage(toolName, args, err)
+			return &compose.StreamToolOutput{
+				Result: schema.StreamReaderFromArray([]string{msg}),
+			}, nil
+		}
+	}
+}
+
+// softRecoveryToolMiddleware returns a ToolMiddleware with both Invokable and Streamable
+// soft recovery (same semantics as hitlToolCallMiddleware bundling).
+func softRecoveryToolMiddleware() compose.ToolMiddleware {
+	return compose.ToolMiddleware{
+		Invokable:  softRecoveryToolCallMiddleware(),
+		Streamable: softRecoveryStreamableToolCallMiddleware(),
+	}
+}
+
 // isSoftRecoverableToolError determines whether a tool execution error should be
 // silently converted to a tool-result message rather than crashing the graph.
 //

internal/multiagent/tool_error_middleware_test.go

@@ -4,6 +4,8 @@ import (
 	"context"
 	"encoding/json"
 	"errors"
+	"io"
+	"strings"
 	"testing"
 
 	"github.com/cloudwego/eino/compose"
@@ -108,6 +110,39 @@ func TestSoftRecoveryToolCallMiddleware_PassesThrough(t *testing.T) {
 	}
 }
 
+func TestSoftRecoveryStreamableToolCallMiddleware_LocalStreamFuncJSONError(t *testing.T) {
+	mw := softRecoveryStreamableToolCallMiddleware()
+	next := func(ctx context.Context, input *compose.ToolInput) (*compose.StreamToolOutput, error) {
+		return nil, errors.New(`[LocalStreamFunc] failed to unmarshal arguments in json, toolName=execute, err="Syntax error no sources available, the input json is empty`)
+	}
+	wrapped := mw(next)
+	out, err := wrapped(context.Background(), &compose.ToolInput{
+		Name:      "execute",
+		Arguments: "",
+	})
+	if err != nil {
+		t.Fatalf("expected nil error (soft recovery), got: %v", err)
+	}
+	if out == nil || out.Result == nil {
+		t.Fatal("expected stream result")
+	}
+	var sb strings.Builder
+	for {
+		chunk, rerr := out.Result.Recv()
+		if errors.Is(rerr, io.EOF) {
+			break
+		}
+		if rerr != nil {
+			t.Fatalf("recv: %v", rerr)
+		}
+		sb.WriteString(chunk)
+	}
+	text := sb.String()
+	if !containsAll(text, "[Tool Error]", "execute", "JSON") {
+		t.Fatalf("recovery message missing expected content: %s", text)
+	}
+}
+
 func TestSoftRecoveryToolCallMiddleware_ConvertsJSONError(t *testing.T) {
 	mw := softRecoveryToolCallMiddleware()
 	next := func(ctx context.Context, input *compose.ToolInput) (*compose.ToolOutput, error) {

internal/openai/claude_bridge.go

@@ -9,6 +9,9 @@ package openai
 //   Stream:   Claude SSE (event: content_block_delta / message_delta) → OpenAI SSE 格式
 //   Auth:     Bearer → x-api-key
 //   Tools:    OpenAI tools[] → Claude tools[] (input_schema)
+//
+// Extended thinking: 顶层 `thinking` 从 OpenAI 请求体透传；响应中 `thinking` block 映射为
+// `reasoning_content`（可读前缀 + 内部 JSON 尾缀以保留 signature，供多轮工具续跑；UI 用 openai.DisplayReasoningContent 剥离）。
 
 import (
 	"bufio"
@@ -38,6 +41,7 @@ type claudeRequest struct {
 	Messages  []claudeMessage `json:"messages"`
 	Tools     []claudeTool    `json:"tools,omitempty"`
 	Stream    bool            `json:"stream,omitempty"`
+	Thinking  json.RawMessage `json:"thinking,omitempty"`
 }
 
 type claudeMessage struct {
@@ -76,6 +80,10 @@ type claudeContentBlock struct {
 	// text block
 	Text string `json:"text,omitempty"`
 
+	// thinking block (extended thinking)
+	Thinking  string `json:"thinking,omitempty"`
+	Signature string `json:"signature,omitempty"`
+
 	// tool_use block (assistant 返回)
 	ID    string          `json:"id,omitempty"`
 	Name  string          `json:"name,omitempty"`
@@ -176,7 +184,13 @@ func convertOpenAIToClaude(payload interface{}) (*claudeRequest, error) {
 
 		// tool_calls (assistant 消息中包含工具调用)
 		if role == "assistant" {
+			rc, _ := mm["reasoning_content"].(string)
+			_, thinkingReplay := parseClaudeReasoningAssistantBlocks(rc)
+
 			var blocks []claudeContentBlock
+			for _, tb := range thinkingReplay {
+				blocks = append(blocks, tb)
+			}
 			if content != "" {
 				blocks = append(blocks, claudeContentBlock{Type: "text", Text: content})
 			}
@@ -290,6 +304,13 @@ func convertOpenAIToClaude(payload interface{}) (*claudeRequest, error) {
 		}
 	}
 
+	// Extended thinking (Anthropic top-level); merged from Eino ExtraFields / admin extras.
+	if th, ok := oai["thinking"]; ok && th != nil {
+		if raw, err := json.Marshal(th); err == nil && len(raw) > 0 && string(raw) != "null" {
+			req.Thinking = json.RawMessage(raw)
+		}
+	}
+
 	return req, nil
 }
 
@@ -318,9 +339,12 @@ func claudeToOpenAIResponseJSON(claudeBody []byte) ([]byte, error) {
 
 	var textContent string
 	var toolCalls []interface{}
+	var thinkingBlocks []claudeContentBlock
 
 	for _, block := range cr.Content {
 		switch block.Type {
+		case "thinking":
+			thinkingBlocks = append(thinkingBlocks, block)
 		case "text":
 			textContent += block.Text
 		case "tool_use":
@@ -344,6 +368,18 @@ func claudeToOpenAIResponseJSON(claudeBody []byte) ([]byte, error) {
 	if len(toolCalls) > 0 {
 		message["tool_calls"] = toolCalls
 	}
+	if len(thinkingBlocks) > 0 {
+		var parts []string
+		for _, tb := range thinkingBlocks {
+			if strings.TrimSpace(tb.Thinking) != "" {
+				parts = append(parts, tb.Thinking)
+			}
+		}
+		rc := appendClaudeReasoningRoundTrip(strings.Join(parts, "\n\n"), thinkingBlocks)
+		if rc != "" {
+			message["reasoning_content"] = rc
+		}
+	}
 
 	choice := map[string]interface{}{
 		"index":         0,
@@ -901,8 +937,16 @@ func (rt *claudeRoundTripper) RoundTrip(req *http.Request) (*http.Response, erro
 
 		reader := bufio.NewReader(resp.Body)
 		blockToToolIndex := make(map[int]int)
+		blockIndexToType := make(map[int]string)
 		nextToolIndex := 0
 
+		type thinkingAcc struct {
+			text strings.Builder
+			sig  strings.Builder
+		}
+		thinkingByIndex := make(map[int]*thinkingAcc)
+		var finishedThinking []claudeContentBlock
+
 		for {
 			line, readErr := reader.ReadString('\n')
 			if readErr != nil {
@@ -947,6 +991,11 @@ func (rt *claudeRoundTripper) RoundTrip(req *http.Request) (*http.Response, erro
 				blockIdx := int(blockIdxFlt)
 				cb, _ := event["content_block"].(map[string]interface{})
 				bt, _ := cb["type"].(string)
+				blockIndexToType[blockIdx] = bt
+
+				if bt == "thinking" {
+					thinkingByIndex[blockIdx] = &thinkingAcc{}
+				}
 
 				if bt == "tool_use" {
 					id, _ := cb["id"].(string)
@@ -986,7 +1035,35 @@ func (rt *claudeRoundTripper) RoundTrip(req *http.Request) (*http.Response, erro
 				delta, _ := event["delta"].(map[string]interface{})
 				dt, _ := delta["type"].(string)
 
-				if dt == "text_delta" {
+				if dt == "thinking_delta" {
+					tPart, _ := delta["thinking"].(string)
+					if tPart != "" {
+						if acc := thinkingByIndex[blockIdx]; acc != nil {
+							acc.text.WriteString(tPart)
+						}
+						oaiChunk := map[string]interface{}{
+							"choices": []map[string]interface{}{
+								{
+									"delta": map[string]interface{}{
+										"reasoning_content": tPart,
+									},
+								},
+							},
+						}
+						b, _ := json.Marshal(oaiChunk)
+						if !writeLine("data: " + string(b) + "\n\n") {
+							pw.Close()
+							return
+						}
+					}
+				} else if dt == "signature_delta" {
+					sigPart, _ := delta["signature"].(string)
+					if sigPart != "" {
+						if acc := thinkingByIndex[blockIdx]; acc != nil {
+							acc.sig.WriteString(sigPart)
+						}
+					}
+				} else if dt == "text_delta" {
 					text, _ := delta["text"].(string)
 					oaiChunk := map[string]interface{}{
 						"choices": []map[string]interface{}{
@@ -1031,6 +1108,21 @@ func (rt *claudeRoundTripper) RoundTrip(req *http.Request) (*http.Response, erro
 					}
 				}
 
+			case "content_block_stop":
+				blockIdxFlt, _ := event["index"].(float64)
+				blockIdx := int(blockIdxFlt)
+				bt := blockIndexToType[blockIdx]
+				if bt == "thinking" {
+					if acc := thinkingByIndex[blockIdx]; acc != nil {
+						finishedThinking = append(finishedThinking, claudeContentBlock{
+							Type:      "thinking",
+							Thinking:  acc.text.String(),
+							Signature: acc.sig.String(),
+						})
+						delete(thinkingByIndex, blockIdx)
+					}
+				}
+
 			case "message_delta":
 				d, _ := event["delta"].(map[string]interface{})
 				if sr, ok := d["stop_reason"].(string); ok {
@@ -1051,6 +1143,25 @@ func (rt *claudeRoundTripper) RoundTrip(req *http.Request) (*http.Response, erro
 				}
 
 			case "message_stop":
+				if len(finishedThinking) > 0 {
+					suffix := appendClaudeReasoningRoundTrip("", finishedThinking)
+					if strings.TrimSpace(suffix) != "" {
+						oaiChunk := map[string]interface{}{
+							"choices": []map[string]interface{}{
+								{
+									"delta": map[string]interface{}{
+										"reasoning_content": suffix,
+									},
+								},
+							},
+						}
+						b, _ := json.Marshal(oaiChunk)
+						if !writeLine("data: " + string(b) + "\n\n") {
+							pw.Close()
+							return
+						}
+					}
+				}
 				writeLine("data: [DONE]\n\n")
 				pw.Close()
 				return

internal/openai/claude_reasoning_roundtrip.go

@@ -0,0 +1,81 @@
+package openai
+
+import (
+	"encoding/json"
+	"strings"
+)
+
+// claudeReasoningRoundTripSep separates human-readable reasoning from a JSON payload of
+// Anthropic thinking blocks (with signatures) for multi-turn extended thinking + tools.
+// Not shown in UI (see DisplayReasoningContent).
+const claudeReasoningRoundTripSep = "\n---CSAI_CLAUDE_THINKING_BLOCKS---\n"
+
+// DisplayReasoningContent returns reasoning text suitable for the UI (strips internal
+// Claude round-trip JSON suffix). Safe for DeepSeek/plain reasoning strings (no-op).
+func DisplayReasoningContent(s string) string {
+	s = strings.TrimSpace(s)
+	if s == "" {
+		return ""
+	}
+	i := strings.LastIndex(s, claudeReasoningRoundTripSep)
+	if i < 0 {
+		return s
+	}
+	return strings.TrimSpace(s[:i])
+}
+
+func appendClaudeReasoningRoundTrip(display string, blocks []claudeContentBlock) string {
+	var payload []map[string]string
+	for _, b := range blocks {
+		if b.Type != "thinking" {
+			continue
+		}
+		payload = append(payload, map[string]string{
+			"type":      b.Type,
+			"thinking":  b.Thinking,
+			"signature": b.Signature,
+		})
+	}
+	if len(payload) == 0 {
+		return strings.TrimSpace(display)
+	}
+	js, err := json.Marshal(payload)
+	if err != nil {
+		return strings.TrimSpace(display)
+	}
+	d := strings.TrimSpace(display)
+	if d == "" {
+		return claudeReasoningRoundTripSep + string(js)
+	}
+	return d + claudeReasoningRoundTripSep + string(js)
+}
+
+// parseClaudeReasoningAssistantBlocks extracts Anthropic thinking blocks from an OpenAI-style
+// reasoning_content string. When no suffix is present, blocks is nil (caller must not invent signatures).
+func parseClaudeReasoningAssistantBlocks(reasoningContent string) (display string, blocks []claudeContentBlock) {
+	reasoningContent = strings.TrimSpace(reasoningContent)
+	if reasoningContent == "" {
+		return "", nil
+	}
+	idx := strings.LastIndex(reasoningContent, claudeReasoningRoundTripSep)
+	if idx < 0 {
+		return reasoningContent, nil
+	}
+	display = strings.TrimSpace(reasoningContent[:idx])
+	jsonPart := strings.TrimSpace(reasoningContent[idx+len(claudeReasoningRoundTripSep):])
+	var arr []struct {
+		Type      string `json:"type"`
+		Thinking  string `json:"thinking"`
+		Signature string `json:"signature"`
+	}
+	if err := json.Unmarshal([]byte(jsonPart), &arr); err != nil {
+		return reasoningContent, nil
+	}
+	for _, x := range arr {
+		if x.Type != "thinking" {
+			continue
+		}
+		blocks = append(blocks, claudeContentBlock{Type: "thinking", Thinking: x.Thinking, Signature: x.Signature})
+	}
+	return display, blocks
+}

internal/openai/claude_reasoning_roundtrip_test.go

@@ -0,0 +1,102 @@
+package openai
+
+import (
+	"encoding/json"
+	"strings"
+	"testing"
+)
+
+func TestDisplayReasoningContent(t *testing.T) {
+	raw := "hello" + claudeReasoningRoundTripSep + `[{"type":"thinking","thinking":"x","signature":"sig"}]`
+	if d := DisplayReasoningContent(raw); d != "hello" {
+		t.Fatalf("got %q", d)
+	}
+	if DisplayReasoningContent("plain") != "plain" {
+		t.Fatal()
+	}
+}
+
+func TestAppendParseClaudeReasoningRoundTrip(t *testing.T) {
+	blocks := []claudeContentBlock{
+		{Type: "thinking", Thinking: "a", Signature: "s1"},
+		{Type: "thinking", Thinking: "b", Signature: "s2"},
+	}
+	s := appendClaudeReasoningRoundTrip("sum", blocks)
+	if !strings.Contains(s, claudeReasoningRoundTripSep) {
+		t.Fatal("missing sep")
+	}
+	display, back := parseClaudeReasoningAssistantBlocks(s)
+	if display != "sum" || len(back) != 2 {
+		t.Fatalf("display=%q len=%d", display, len(back))
+	}
+	if back[0].Signature != "s1" || back[1].Thinking != "b" {
+		t.Fatalf("%+v", back)
+	}
+}
+
+func TestConvertOpenAIToClaude_AssistantReasoningReplay(t *testing.T) {
+	rc := appendClaudeReasoningRoundTrip("vis", []claudeContentBlock{
+		{Type: "thinking", Thinking: "t1", Signature: "sig1"},
+	})
+	payload := map[string]interface{}{
+		"model": "claude-3-5-sonnet-latest",
+		"messages": []interface{}{
+			map[string]interface{}{
+				"role":              "assistant",
+				"content":           "out",
+				"reasoning_content": rc,
+			},
+		},
+	}
+	req, err := convertOpenAIToClaude(payload)
+	if err != nil {
+		t.Fatal(err)
+	}
+	if len(req.Messages) != 1 {
+		t.Fatalf("messages=%d", len(req.Messages))
+	}
+	blocks := req.Messages[0].Content.Blocks
+	if len(blocks) < 2 {
+		t.Fatalf("blocks=%d", len(blocks))
+	}
+	if blocks[0].Type != "thinking" || blocks[0].Signature != "sig1" {
+		t.Fatalf("first block %+v", blocks[0])
+	}
+	foundText := false
+	for _, b := range blocks {
+		if b.Type == "text" && b.Text == "out" {
+			foundText = true
+		}
+	}
+	if !foundText {
+		t.Fatalf("blocks=%+v", blocks)
+	}
+}
+
+func TestClaudeToOpenAIResponseJSON_Thinking(t *testing.T) {
+	claudeBody := []byte(`{
+		"id":"msg_1","type":"message","role":"assistant","model":"x","stop_reason":"end_turn",
+		"content":[
+			{"type":"thinking","thinking":"step","signature":"sigx"},
+			{"type":"text","text":"hi"}
+		]
+	}`)
+	oai, err := claudeToOpenAIResponseJSON(claudeBody)
+	if err != nil {
+		t.Fatal(err)
+	}
+	var wrap map[string]interface{}
+	if err := json.Unmarshal(oai, &wrap); err != nil {
+		t.Fatal(err)
+	}
+	choices := wrap["choices"].([]interface{})
+	ch0 := choices[0].(map[string]interface{})
+	msg := ch0["message"].(map[string]interface{})
+	rc, _ := msg["reasoning_content"].(string)
+	if !strings.Contains(rc, "step") || !strings.Contains(rc, claudeReasoningRoundTripSep) {
+		t.Fatalf("reasoning_content=%q", rc)
+	}
+	if msg["content"] != "hi" {
+		t.Fatal()
+	}
+}

internal/reasoning/eino.go

@@ -0,0 +1,250 @@
+// Package reasoning maps user/config intent to CloudWeGo Eino OpenAI ChatModel fields
+// (ReasoningEffort, ExtraFields such as thinking / reasoning_effort / output_config).
+package reasoning
+
+import (
+	"strings"
+
+	"cyberstrike-ai/internal/config"
+
+	einoopenai "github.com/cloudwego/eino-ext/components/model/openai"
+)
+
+// ClientIntent is optional per-request override from ChatRequest.reasoning.
+type ClientIntent struct {
+	Mode   string
+	Effort string
+}
+
+type wireProfile int
+
+const (
+	wireNone wireProfile = iota
+	wireClaude
+	wireDeepseek
+	wireOpenAI
+	wireOutputConfig
+)
+
+// ApplyToEinoChatModelConfig merges reasoning-related options into cfg.
+// Precondition: cfg already has APIKey, BaseURL, Model, HTTPClient set.
+func ApplyToEinoChatModelConfig(cfg *einoopenai.ChatModelConfig, oa *config.OpenAIConfig, client *ClientIntent) {
+	if cfg == nil || oa == nil {
+		return
+	}
+	sr := &oa.Reasoning
+	allowClient := sr.AllowClientReasoningEffective()
+	mode := effectiveMode(sr, client, allowClient)
+
+	// Claude (Anthropic): merge admin extras first; optional extended thinking maps to top-level `thinking`
+	// (see internal/openai convertOpenAIToClaude). DeepSeek/OpenAI-style fields are not sent.
+	if strings.EqualFold(strings.TrimSpace(oa.Provider), "claude") ||
+		strings.EqualFold(strings.TrimSpace(oa.Provider), "anthropic") {
+		if len(sr.ExtraRequestFields) > 0 {
+			if cfg.ExtraFields == nil {
+				cfg.ExtraFields = make(map[string]any)
+			}
+			for k, v := range sr.ExtraRequestFields {
+				cfg.ExtraFields[k] = v
+			}
+		}
+		if mode == "off" {
+			return
+		}
+		applyClaudeExtendedThinking(cfg, mode, effectiveEffort(sr, client, allowClient), oa.Model)
+		return
+	}
+
+	if mode == "off" {
+		return
+	}
+	effort := effectiveEffort(sr, client, allowClient)
+	prof := resolveWireProfile(oa, sr)
+
+	// Admin-defined extra root fields (merged first; automatic keys may follow).
+	if len(sr.ExtraRequestFields) > 0 {
+		if cfg.ExtraFields == nil {
+			cfg.ExtraFields = make(map[string]any)
+		}
+		for k, v := range sr.ExtraRequestFields {
+			cfg.ExtraFields[k] = v
+		}
+	}
+
+	switch prof {
+	case wireClaude, wireNone:
+		return
+	case wireDeepseek:
+		applyDeepseek(cfg, mode, effort)
+	case wireOutputConfig:
+		applyOutputConfigEffort(cfg, mode, effort)
+	default: // wireOpenAI
+		applyOpenAICompat(cfg, mode, effort)
+	}
+}
+
+// applyClaudeExtendedThinking sets Anthropic Messages API `thinking` when absent from ExtraRequestFields.
+// Uses adaptive + summarized display by default (per Anthropic guidance for Claude 4.x); Sonnet 3.7 uses enabled+budget.
+func applyClaudeExtendedThinking(cfg *einoopenai.ChatModelConfig, mode, effort, model string) {
+	if cfg == nil || mode == "off" {
+		return
+	}
+	if cfg.ExtraFields == nil {
+		cfg.ExtraFields = make(map[string]any)
+	}
+	if _, exists := cfg.ExtraFields["thinking"]; exists {
+		return
+	}
+	m := strings.ToLower(strings.TrimSpace(model))
+	thinking := map[string]any{
+		"type":    "adaptive",
+		"display": "summarized",
+	}
+	// Sonnet 3.7: manual extended thinking is the documented path.
+	if strings.Contains(m, "claude-3-7-sonnet") || strings.Contains(m, "3-7-sonnet") || strings.Contains(m, "sonnet-3.7") {
+		thinking = map[string]any{
+			"type":          "enabled",
+			"budget_tokens": 10000,
+			"display":       "summarized",
+		}
+	}
+	// Opus 4.7+: manual enabled+budget rejected — keep adaptive only.
+	if strings.Contains(m, "opus-4-7") || strings.Contains(m, "opus-4.7") {
+		thinking = map[string]any{
+			"type":    "adaptive",
+			"display": "summarized",
+		}
+	}
+	_ = effort // reserved: map to Anthropic effort / output_config when API stabilizes in one place
+	cfg.ExtraFields["thinking"] = thinking
+}
+
+func effectiveMode(sr *config.OpenAIReasoningConfig, client *ClientIntent, allowClient bool) string {
+	server := strings.ToLower(strings.TrimSpace(sr.ModeEffective()))
+	if server == "" || server == "default" {
+		server = "auto"
+	}
+	if !allowClient || client == nil {
+		return server
+	}
+	cm := strings.ToLower(strings.TrimSpace(client.Mode))
+	if cm == "" || cm == "default" {
+		return server
+	}
+	return cm
+}
+
+func effectiveEffort(sr *config.OpenAIReasoningConfig, client *ClientIntent, allowClient bool) string {
+	se := normalizeEffort(sr.Effort)
+	if !allowClient || client == nil {
+		return se
+	}
+	ce := normalizeEffort(client.Effort)
+	if ce != "" {
+		return ce
+	}
+	return se
+}
+
+func normalizeEffort(s string) string {
+	e := strings.ToLower(strings.TrimSpace(s))
+	switch e {
+	case "low", "medium", "high", "max":
+		return e
+	default:
+		return ""
+	}
+}
+
+func resolveWireProfile(oa *config.OpenAIConfig, sr *config.OpenAIReasoningConfig) wireProfile {
+	if strings.EqualFold(strings.TrimSpace(oa.Provider), "claude") {
+		return wireClaude
+	}
+	p := strings.ToLower(strings.TrimSpace(sr.ProfileEffective()))
+	switch p {
+	case "output_config", "output_config_effort":
+		return wireOutputConfig
+	case "openai", "openai_compat":
+		return wireOpenAI
+	case "deepseek", "deepseek_compat":
+		return wireDeepseek
+	case "auto", "":
+		bu := strings.ToLower(oa.BaseURL)
+		mo := strings.ToLower(oa.Model)
+		if strings.Contains(bu, "deepseek") || strings.Contains(mo, "deepseek") {
+			return wireDeepseek
+		}
+		return wireOpenAI
+	default:
+		return wireOpenAI
+	}
+}
+
+func applyDeepseek(cfg *einoopenai.ChatModelConfig, mode, effort string) {
+	// auto: enable thinking for DeepSeek line; on: same; auto without effort still opens thinking.
+	if mode == "off" {
+		return
+	}
+	if mode == "auto" || mode == "on" {
+		if cfg.ExtraFields == nil {
+			cfg.ExtraFields = make(map[string]any)
+		}
+		cfg.ExtraFields["thinking"] = map[string]any{"type": "enabled"}
+	}
+	if effort != "" {
+		if cfg.ExtraFields == nil {
+			cfg.ExtraFields = make(map[string]any)
+		}
+		cfg.ExtraFields["reasoning_effort"] = effortStringForAPI(effort)
+	}
+}
+
+func applyOpenAICompat(cfg *einoopenai.ChatModelConfig, mode, effort string) {
+	if mode == "auto" && effort == "" {
+		return
+	}
+	e := effort
+	if mode == "on" && e == "" {
+		e = "medium"
+	}
+	if e == "" {
+		return
+	}
+	if e == "max" {
+		if cfg.ExtraFields == nil {
+			cfg.ExtraFields = make(map[string]any)
+		}
+		cfg.ExtraFields["reasoning_effort"] = "max"
+		return
+	}
+	switch e {
+	case "low":
+		cfg.ReasoningEffort = einoopenai.ReasoningEffortLevelLow
+	case "medium":
+		cfg.ReasoningEffort = einoopenai.ReasoningEffortLevelMedium
+	case "high":
+		cfg.ReasoningEffort = einoopenai.ReasoningEffortLevelHigh
+	}
+}
+
+func applyOutputConfigEffort(cfg *einoopenai.ChatModelConfig, mode, effort string) {
+	if mode == "auto" && effort == "" {
+		return
+	}
+	e := effort
+	if mode == "on" && e == "" {
+		e = "high"
+	}
+	if e == "" {
+		return
+	}
+	if cfg.ExtraFields == nil {
+		cfg.ExtraFields = make(map[string]any)
+	}
+	cfg.ExtraFields["output_config"] = map[string]any{"effort": effortStringForAPI(e)}
+}
+
+func effortStringForAPI(e string) string {
+	// Gateways expect lowercase strings; "max" kept as max.
+	return strings.ToLower(strings.TrimSpace(e))
+}

internal/security/executor.go

@@ -153,6 +153,7 @@ func (e *Executor) ExecuteTool(ctx context.Context, toolName string, args map[st
 	// 执行命令
 	cmd := exec.CommandContext(ctx, toolConfig.Command, cmdArgs...)
 	applyDefaultTerminalEnv(cmd)
+	_ = prepareShellCmdSession(cmd)
 
 	e.logger.Info("执行安全工具",
 		zap.String("tool", toolName),
@@ -163,13 +164,14 @@ func (e *Executor) ExecuteTool(ctx context.Context, toolName string, args map[st
 	var err error
 	// 如果上层提供了 stdout/stderr 增量回调，则边执行边读取并回调。
 	if cb, ok := ctx.Value(ToolOutputCallbackCtxKey).(ToolOutputCallback); ok && cb != nil {
-		output, err = streamCommandOutput(cmd, cb)
+		output, err = streamCommandOutput(ctx, cmd, cb)
 		if err != nil && shouldRetryWithPTY(output) {
 			e.logger.Info("检测到工具需要 TTY，使用 PTY 重试",
 				zap.String("tool", toolName),
 			)
 			cmd2 := exec.CommandContext(ctx, toolConfig.Command, cmdArgs...)
 			applyDefaultTerminalEnv(cmd2)
+			_ = prepareShellCmdSession(cmd2)
 			output, err = runCommandWithPTY(ctx, cmd2, cb)
 		}
 	} else {
@@ -182,6 +184,7 @@ func (e *Executor) ExecuteTool(ctx context.Context, toolName string, args map[st
 			)
 			cmd2 := exec.CommandContext(ctx, toolConfig.Command, cmdArgs...)
 			applyDefaultTerminalEnv(cmd2)
+			_ = prepareShellCmdSession(cmd2)
 			output, err = runCommandWithPTY(ctx, cmd2, nil)
 		}
 	}
@@ -837,6 +840,8 @@ func (e *Executor) executeSystemCommand(ctx context.Context, args map[string]int
 	} else {
 		cmd = exec.CommandContext(ctx, shell, "-c", command)
 	}
+	applyDefaultTerminalEnv(cmd)
+	_ = prepareShellCmdSession(cmd)
 
 	// 执行命令
 	e.logger.Info("执行系统命令",
@@ -865,6 +870,8 @@ func (e *Executor) executeSystemCommand(ctx context.Context, args map[string]int
 		} else {
 			pidCmd = exec.CommandContext(ctx, shell, "-c", pidCommand)
 		}
+		applyDefaultTerminalEnv(pidCmd)
+		_ = prepareShellCmdSession(pidCmd)
 
 		// 获取stdout管道
 		stdout, err := pidCmd.StdoutPipe()
@@ -976,7 +983,7 @@ func (e *Executor) executeSystemCommand(ctx context.Context, args map[string]int
 	var err error
 	// 若上层提供工具输出增量回调，则边执行边流式读取。
 	if cb, ok := ctx.Value(ToolOutputCallbackCtxKey).(ToolOutputCallback); ok && cb != nil {
-		output, err = streamCommandOutput(cmd, cb)
+		output, err = streamCommandOutput(ctx, cmd, cb)
 		if err != nil && shouldRetryWithPTY(output) {
 			e.logger.Info("检测到系统命令需要 TTY，使用 PTY 重试")
 			cmd2 := exec.CommandContext(ctx, shell, "-c", command)
@@ -984,6 +991,7 @@ func (e *Executor) executeSystemCommand(ctx context.Context, args map[string]int
 				cmd2.Dir = workDir
 			}
 			applyDefaultTerminalEnv(cmd2)
+			_ = prepareShellCmdSession(cmd2)
 			output, err = runCommandWithPTY(ctx, cmd2, cb)
 		}
 	} else {
@@ -997,6 +1005,7 @@ func (e *Executor) executeSystemCommand(ctx context.Context, args map[string]int
 				cmd2.Dir = workDir
 			}
 			applyDefaultTerminalEnv(cmd2)
+			_ = prepareShellCmdSession(cmd2)
 			output, err = runCommandWithPTY(ctx, cmd2, nil)
 		}
 	}
@@ -1034,8 +1043,11 @@ func (e *Executor) executeSystemCommand(ctx context.Context, args map[string]int
 }
 
 // streamCommandOutput 以“边读边回调”的方式读取命令 stdout/stderr。
-// 保持输出内容完整拼接返回，并用 cb(chunk) 向上层持续推送。
-func streamCommandOutput(cmd *exec.Cmd, cb ToolOutputCallback) (string, error) {
+// 使用定长块读取，避免按行读取在无换行输出时永久阻塞；ctx 取消时终止进程树。
+func streamCommandOutput(ctx context.Context, cmd *exec.Cmd, cb ToolOutputCallback) (string, error) {
+	if err := prepareShellCmdSession(cmd); err != nil {
+		return "", err
+	}
 	stdoutPipe, err := cmd.StdoutPipe()
 	if err != nil {
 		return "", err
@@ -1051,18 +1063,27 @@ func streamCommandOutput(cmd *exec.Cmd, cb ToolOutputCallback) (string, error) {
 		return "", err
 	}
 
+	stopWatch := make(chan struct{})
+	go func() {
+		select {
+		case <-ctx.Done():
+			terminateCmdTree(cmd)
+		case <-stopWatch:
+		}
+	}()
+	defer close(stopWatch)
+
 	chunks := make(chan string, 64)
 	var wg sync.WaitGroup
 	readFn := func(r io.Reader) {
 		defer wg.Done()
-		br := bufio.NewReader(r)
+		buf := make([]byte, 8192)
 		for {
-			s, readErr := br.ReadString('\n')
-			if s != "" {
-				chunks <- s
+			n, readErr := r.Read(buf)
+			if n > 0 {
+				chunks <- string(buf[:n])
 			}
 			if readErr != nil {
-				// EOF 正常结束
 				return
 			}
 		}
@@ -1158,12 +1179,14 @@ func runCommandWithPTY(ctx context.Context, cmd *exec.Cmd, cb ToolOutputCallback
 	if runtime.GOOS == "windows" {
 		// PTY 方案为类 Unix；Windows 走原逻辑
 		if cb != nil {
-			return streamCommandOutput(cmd, cb)
+			return streamCommandOutput(ctx, cmd, cb)
 		}
+		_ = prepareShellCmdSession(cmd)
 		out, err := cmd.CombinedOutput()
 		return string(out), err
 	}
 
+	_ = prepareShellCmdSession(cmd)
 	ptmx, err := pty.Start(cmd)
 	if err != nil {
 		return "", err
@@ -1176,9 +1199,7 @@ func runCommandWithPTY(ctx context.Context, cmd *exec.Cmd, cb ToolOutputCallback
 		select {
 		case <-ctx.Done():
 			_ = ptmx.Close() // 触发读退出
-			if cmd.Process != nil {
-				_ = cmd.Process.Kill()
-			}
+			terminateCmdTree(cmd)
 		case <-done:
 		}
 	}()

internal/security/procattr_unix.go

@@ -0,0 +1,31 @@
+//go:build !windows
+
+package security
+
+import (
+	"os/exec"
+	"syscall"
+)
+
+// prepareShellCmdSession 让 shell 子进程在独立会话中运行，便于超时/取消时整组 SIGKILL（含子进程）。
+func prepareShellCmdSession(cmd *exec.Cmd) error {
+	if cmd == nil {
+		return nil
+	}
+	if cmd.SysProcAttr == nil {
+		cmd.SysProcAttr = &syscall.SysProcAttr{}
+	}
+	cmd.SysProcAttr.Setsid = true
+	return nil
+}
+
+// terminateCmdTree 尽力终止 cmd 及其进程组（Unix 下 Setsid 后 PGID == 首进程 PID）。
+func terminateCmdTree(cmd *exec.Cmd) {
+	if cmd == nil || cmd.Process == nil {
+		return
+	}
+	pid := cmd.Process.Pid
+	if err := syscall.Kill(-pid, syscall.SIGKILL); err != nil {
+		_ = cmd.Process.Kill()
+	}
+}

internal/security/procattr_windows.go

@@ -0,0 +1,17 @@
+//go:build windows
+
+package security
+
+import "os/exec"
+
+func prepareShellCmdSession(cmd *exec.Cmd) error {
+	_ = cmd
+	return nil
+}
+
+func terminateCmdTree(cmd *exec.Cmd) {
+	if cmd == nil || cmd.Process == nil {
+		return
+	}
+	_ = cmd.Process.Kill()
+}

upgrade.sh

@@ -8,11 +8,8 @@ set -euo pipefail
 # - data/
 # - venv/ (disabled with --no-venv)
 # - tools/ (user extensions; never overwritten by upgrade)
-#
-# Optional preserves (may overwrite upstream updates):
 # - roles/
 # - skills/
-# Enable with --preserve-custom
 
 ROOT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
 cd "$ROOT_DIR"
@@ -28,7 +25,6 @@ BACKUP_BASE_DIR="$ROOT_DIR/.upgrade-backup"
 GITHUB_REPO="Ed1s0nZ/CyberStrikeAI"
 
 TAG=""
-PRESERVE_CUSTOM=0
 PRESERVE_VENV=1
 STOP_SERVICE=1
 FORCE_STOP=0
@@ -37,14 +33,12 @@ YES=0
 usage() {
   cat <<EOF
 Usage:
-  ./upgrade.sh [--tag vX.Y.Z] [--preserve-custom] [--no-venv] [--no-stop]
+  ./upgrade.sh [--tag vX.Y.Z] [--no-venv] [--no-stop]
                 [--force-stop] [--yes]
 
 Options:
   --tag <tag>          Specify GitHub Release tag (e.g. v1.3.28).
                         If omitted, the script uses the latest release.
-  --preserve-custom    Preserve roles/skills (may overwrite upstream files).
-                        tools/ is always preserved. Use with caution.
   --no-venv             Do not preserve venv/ (Python deps will be re-installed).
   --no-stop             Do not try to stop the running service.
   --force-stop         If no process matching current directory is found, also stop
@@ -52,7 +46,7 @@ Options:
   --yes                 Do not ask for confirmation.
 
 Description:
-  The script backs up config.yaml/data/tools/ (and optionally venv/roles/skills) to
+  The script backs up config.yaml/data/tools/roles/skills/ (and optionally venv/) to
   .upgrade-backup/
 EOF
 }
@@ -177,11 +171,7 @@ confirm_or_exit() {
     info " - Preserve venv/: no (will remove old venv and re-install deps)"
   fi
   info " - Preserve tools/: yes (always)"
-  if [[ "$PRESERVE_CUSTOM" -eq 1 ]]; then
-    info " - Preserve roles/skills: yes (may overwrite upstream updates)"
-  else
-    info " - Preserve roles/skills: no (will use upstream versions)"
-  fi
+  info " - Preserve roles/skills: yes (always)"
   info " - Stop service: ${STOP_SERVICE}"
   echo ""
   read -r -p "Continue? (y/N) " ans
@@ -299,11 +289,8 @@ sync_code() {
 
   # User tool extensions: never replace or delete during upgrade.
   rsync_excludes+=( "--exclude=tools/" )
-
-  if [[ "$PRESERVE_CUSTOM" -eq 1 ]]; then
-    rsync_excludes+=( "--exclude=roles/" )
-    rsync_excludes+=( "--exclude=skills/" )
-  fi
+  rsync_excludes+=( "--exclude=roles/" )
+  rsync_excludes+=( "--exclude=skills/" )
 
   # Ensure this upgrade script itself is not deleted.
   rsync_excludes+=( "--exclude=upgrade.sh" )
@@ -324,10 +311,6 @@ main() {
         TAG="${2:-}"
         shift 2
         ;;
-      --preserve-custom)
-        PRESERVE_CUSTOM=1
-        shift 1
-        ;;
       --no-venv)
         PRESERVE_VENV=0
         shift 1
@@ -384,8 +367,10 @@ main() {
   if [[ -d "$ROOT_DIR/tools" ]]; then
     backup_dir_tgz "tools" "$ROOT_DIR/tools"
   fi
-  if [[ "$PRESERVE_CUSTOM" -eq 1 ]]; then
+  if [[ -d "$ROOT_DIR/roles" ]]; then
     backup_dir_tgz "roles" "$ROOT_DIR/roles"
+  fi
+  if [[ -d "$ROOT_DIR/skills" ]]; then
     backup_dir_tgz "skills" "$ROOT_DIR/skills"
   fi
 

web/static/css/style.css

@@ -536,6 +536,10 @@ body {
     display: none;
 }
 
+.conversation-sidebar.collapsed .conversation-reasoning-card {
+    display: none;
+}
+
 .conversation-sidebar.collapsed .conversation-sidebar-header {
     flex-direction: column;
     align-items: center;
@@ -1136,13 +1140,13 @@ header {
 .hitl-sidebar-card {
     border-top: 1px solid var(--border-color);
     background: linear-gradient(165deg, #f8fafc 0%, #f1f5f9 55%, #eef2f7 100%);
-    padding: 14px 12px 16px;
+    padding: 11px 12px;
     flex-shrink: 0;
 }
 
 .hitl-sidebar-card-header {
     display: flex;
-    align-items: flex-start;
+    align-items: center;
     justify-content: space-between;
     gap: 10px;
     cursor: pointer;
@@ -1160,7 +1164,7 @@ header {
     overflow: hidden;
     max-height: 500px;
     opacity: 1;
-    margin-top: 10px;
+    margin-top: 8px;
     transition: max-height 0.3s ease, opacity 0.2s ease, margin-top 0.3s ease;
 }
 
@@ -1176,19 +1180,19 @@ header {
 
 .hitl-sidebar-heading {
     display: flex;
-    align-items: flex-start;
-    gap: 10px;
+    align-items: center;
+    gap: 8px;
     min-width: 0;
 }
 
 .hitl-sidebar-icon {
     flex-shrink: 0;
-    width: 36px;
-    height: 36px;
+    width: 32px;
+    height: 32px;
     display: flex;
     align-items: center;
     justify-content: center;
-    border-radius: 10px;
+    border-radius: 9px;
     background: linear-gradient(145deg, rgba(0, 102, 255, 0.12), rgba(0, 102, 255, 0.06));
     color: var(--accent-color);
     border: 1px solid rgba(0, 102, 255, 0.18);
@@ -1197,27 +1201,27 @@ header {
 .hitl-sidebar-heading-text {
     display: flex;
     flex-direction: column;
-    gap: 2px;
+    gap: 1px;
     min-width: 0;
 }
 
 .hitl-sidebar-title {
-    font-size: 15px;
+    font-size: 14px;
     font-weight: 700;
     letter-spacing: -0.02em;
     color: var(--text-primary);
-    line-height: 1.25;
+    line-height: 1.2;
 }
 
 .hitl-sidebar-subtitle {
     font-size: 11px;
     font-weight: 500;
     color: var(--text-secondary);
-    line-height: 1.3;
+    line-height: 1.25;
 }
 
 .hitl-apply-btn {
-    padding: 8px 14px;
+    padding: 6px 12px;
     border-radius: 10px;
     font-size: 12px;
     font-weight: 600;
@@ -2391,7 +2395,168 @@ header {
     box-sizing: border-box;
 }
 
-.chat-input-container > .chat-input-with-files {
+.chat-input-primary-row {
+    display: flex;
+    flex-direction: row;
+    align-items: flex-end;
+    gap: 8px;
+    flex: 1;
+    min-width: 0;
+    width: 100%;
+}
+
+.chat-input-leading {
+    display: flex;
+    flex-direction: row;
+    align-items: flex-end;
+    gap: 8px;
+    flex-shrink: 0;
+}
+
+/* Eino：模型推理在对话列表侧栏底部，默认折叠 */
+.conversation-sidebar .chat-reasoning-wrapper {
+    width: 100%;
+    box-sizing: border-box;
+    flex-shrink: 0;
+}
+
+.conversation-reasoning-card {
+    border-top: 1px solid var(--border-color);
+    background: linear-gradient(165deg, #f8fafc 0%, #f1f5f9 55%, #eef2f7 100%);
+    padding: 11px 12px;
+    flex-shrink: 0;
+}
+
+.conversation-reasoning-card-header {
+    display: flex;
+    align-items: center;
+    justify-content: flex-start;
+    gap: 0;
+    width: 100%;
+    padding: 0;
+    margin: 0;
+    border: none;
+    background: transparent;
+    cursor: pointer;
+    text-align: left;
+    font: inherit;
+    color: inherit;
+    -webkit-appearance: none;
+    appearance: none;
+    border-radius: 0;
+}
+
+.conversation-reasoning-card-header:hover .conversation-reasoning-title {
+    color: var(--accent-color);
+}
+
+.conversation-reasoning-heading {
+    display: flex;
+    align-items: center;
+    gap: 8px;
+    min-width: 0;
+    flex: 1;
+}
+
+.conversation-reasoning-icon {
+    flex-shrink: 0;
+    width: 32px;
+    height: 32px;
+    display: flex;
+    align-items: center;
+    justify-content: center;
+    border-radius: 9px;
+    background: linear-gradient(145deg, rgba(0, 102, 255, 0.12), rgba(0, 102, 255, 0.06));
+    color: var(--accent-color);
+    border: 1px solid rgba(0, 102, 255, 0.18);
+}
+
+.conversation-reasoning-icon svg {
+    display: block;
+}
+
+.conversation-reasoning-heading-text {
+    flex: 1;
+    min-width: 0;
+    display: flex;
+    flex-direction: column;
+    gap: 1px;
+}
+
+.conversation-reasoning-title {
+    font-size: 14px;
+    font-weight: 700;
+    letter-spacing: -0.02em;
+    color: var(--text-primary);
+    line-height: 1.2;
+}
+
+.conversation-reasoning-summary {
+    font-size: 11px;
+    font-weight: 500;
+    color: var(--text-secondary);
+    line-height: 1.25;
+    overflow: hidden;
+    text-overflow: ellipsis;
+    white-space: nowrap;
+}
+
+.conversation-reasoning-body {
+    overflow: hidden;
+    max-height: 280px;
+    opacity: 1;
+    margin-top: 8px;
+    padding-bottom: 0;
+    transition: max-height 0.3s ease, opacity 0.2s ease, margin-top 0.3s ease;
+}
+
+.conversation-reasoning-card.conversation-reasoning-collapsed .conversation-reasoning-body {
+    max-height: 0;
+    opacity: 0;
+    margin-top: 0;
+    padding-bottom: 0;
+    pointer-events: none;
+}
+
+.conversation-reasoning-body .chat-reasoning-panel-hint {
+    margin-top: 0;
+    margin-bottom: 8px;
+}
+
+.chat-reasoning-panel-hint {
+    font-size: 0.75rem;
+    color: var(--text-muted, #718096);
+    margin: 0;
+    line-height: 1.45;
+}
+
+.chat-reasoning-fields {
+    display: flex;
+    flex-direction: column;
+    gap: 12px;
+}
+
+.chat-reasoning-field-label {
+    display: block;
+    font-size: 0.75rem;
+    font-weight: 600;
+    color: var(--text-muted, #718096);
+    margin-bottom: 6px;
+}
+
+.chat-reasoning-select {
+    width: 100%;
+    box-sizing: border-box;
+    padding: 0.45rem 0.6rem;
+    font-size: 0.8125rem;
+    border: 1px solid var(--border-color, #e2e8f0);
+    border-radius: 8px;
+    background: var(--card-bg, #fff);
+    color: var(--text-color, #2d3748);
+}
+
+.chat-input-container .chat-input-with-files,
+.chat-input-primary-row .chat-input-with-files {
     flex: 1;
     display: flex;
     flex-direction: column;
@@ -2399,7 +2564,8 @@ header {
     gap: 6px;
 }
 
-.chat-input-container > .chat-input-field {
+.chat-input-container > .chat-input-field,
+.chat-input-primary-row .chat-input-field {
     flex: 1;
     display: flex;
     min-width: 0;
@@ -3542,18 +3708,14 @@ header {
 
 .timeline-item-iteration {
     border-left-color: var(--accent-color);
-    background: rgba(0, 102, 255, 0.05);
+    background: rgba(0, 102, 255, 0.06);
 }
 
-/* Eino 多代理：主编排器 vs 子代理时间线区分 */
-.timeline-eino-role-orchestrator {
-    border-left-color: #5c6bc0 !important;
-    background: rgba(92, 107, 192, 0.09) !important;
-}
-.timeline-eino-role-sub {
-    border-left-color: #00897b !important;
-    background: rgba(0, 137, 123, 0.08) !important;
-}
+/*
+ * Eino 主/子代理：保留 timeline-eino-role-* class（由 applyEinoTimelineRole 写入），
+ * 但不再在此处整卡铺色 + !important，否则会盖住工具调用/结果/思考的类型色。
+ * 主编排 vs 子代理的区分由「迭代轮次」上的 timeline-eino-scope-* 负责。
+ */
 .timeline-item-iteration.timeline-eino-scope-main {
     border-left-color: #3949ab !important;
     background: rgba(57, 73, 171, 0.1) !important;
@@ -3563,24 +3725,72 @@ header {
     background: rgba(0, 105, 92, 0.09) !important;
 }
 
+/* 模型内部思考：弱化灰紫，避免与「助手输出」抢视觉 */
 .timeline-item-thinking {
-    border-left-color: #9c27b0;
-    background: rgba(156, 39, 176, 0.05);
+    border-left-color: #7e57c2;
+    background: rgba(103, 58, 183, 0.06);
 }
 
+/* 迭代中主通道流式正文（标题常为「助手输出」等）：中性底 + 主色条，表示对用户可见的答复流 */
+.timeline-item-thinking[data-response-stream-placeholder="1"] {
+    border-left-color: var(--accent-color);
+    background: rgba(0, 102, 255, 0.04);
+}
+
+.timeline-item-reasoning_chain {
+    border-left-color: #5e35b1;
+    background: rgba(94, 53, 177, 0.07);
+}
+
+.timeline-item-planning {
+    border-left-color: #00838f;
+    background: rgba(0, 131, 143, 0.06);
+}
+
+/* 工具调用：信息色（蓝），与「结果绿/红」分离；完成态不再用绿色以免与成功结果混淆 */
 .timeline-item-tool_call {
-    border-left-color: #ff9800;
-    background: rgba(255, 152, 0, 0.05);
+    border-left-color: #1565c0;
+    background: rgba(21, 101, 192, 0.07);
 }
 
 .timeline-item-tool_result {
+    border-left-color: #78909c;
+    background: rgba(120, 144, 156, 0.06);
+}
+
+.timeline-item-tool_result:has(.tool-result-section.success) {
     border-left-color: var(--success-color);
-    background: rgba(40, 167, 69, 0.05);
+    background: rgba(40, 167, 69, 0.07);
+}
+
+.timeline-item-tool_result:has(.tool-result-section.error) {
+    border-left-color: var(--error-color);
+    background: rgba(220, 53, 69, 0.07);
 }
 
 .timeline-item-tool_result.error {
     border-left-color: var(--error-color);
-    background: rgba(220, 53, 69, 0.05);
+    background: rgba(220, 53, 69, 0.07);
+}
+
+.timeline-item-eino_agent_reply {
+    border-left-color: #6a1b9a;
+    background: rgba(106, 27, 154, 0.07);
+}
+
+.timeline-item-progress {
+    border-left-color: #607d8b;
+    background: rgba(96, 125, 139, 0.08);
+}
+
+.timeline-item-warning {
+    border-left-color: #f57c00;
+    background: rgba(245, 124, 0, 0.09);
+}
+
+.timeline-item-tool_calls_detected {
+    border-left-color: #0277bd;
+    background: rgba(2, 119, 189, 0.06);
 }
 
 .timeline-item-error {
@@ -3770,20 +3980,36 @@ header {
     border: 1px solid rgba(220, 53, 69, 0.3);
 }
 
-/* 工具调用项状态样式 */
+/* 工具调用项状态：全程保持「信息蓝」系，完成态不用绿色（避免与工具成功结果混淆） */
 .timeline-item-tool_call.tool-call-running {
-    border-left-color: var(--accent-color);
-    background: rgba(0, 102, 255, 0.08);
+    border-left-color: #42a5f5;
+    background: rgba(66, 165, 245, 0.1);
 }
 
 .timeline-item-tool_call.tool-call-completed {
-    border-left-color: var(--success-color);
-    background: rgba(40, 167, 69, 0.08);
+    border-left-color: #0d47a1;
+    background: rgba(13, 71, 161, 0.08);
 }
 
 .timeline-item-tool_call.tool-call-failed {
     border-left-color: var(--error-color);
-    background: rgba(220, 53, 69, 0.08);
+    background: rgba(220, 53, 69, 0.1);
+}
+
+/* 参数块与卡片类型色弱对齐，扫读时一眼归到「调用」 */
+.timeline-item-tool_call .tool-args {
+    background: rgba(21, 101, 192, 0.06);
+    border-color: rgba(21, 101, 192, 0.22);
+}
+
+.timeline-item-tool_result:has(.tool-result-section.success) .tool-result {
+    background: rgba(40, 167, 69, 0.08);
+    border-color: rgba(40, 167, 69, 0.35);
+}
+
+.timeline-item-tool_result:has(.tool-result-section.error) .tool-result {
+    background: rgba(220, 53, 69, 0.1);
+    border-color: rgba(220, 53, 69, 0.45);
 }
 
 /* 活跃任务栏 */
@@ -12294,6 +12520,9 @@ header {
 .webshell-ai-process-block .webshell-ai-timeline-thinking {
     border-left-color: #9c27b0;
 }
+.webshell-ai-process-block .webshell-ai-timeline-reasoning_chain {
+    border-left-color: #5c6bc0;
+}
 .webshell-ai-process-block .webshell-ai-timeline-tool_call,
 .webshell-ai-process-block .webshell-ai-timeline-tool_calls_detected {
     border-left-color: #ff9800;

web/static/i18n/en-US.json

@@ -277,6 +277,7 @@
     "planExecuteStreamPhase": "Phase output",
     "einoSubAgentStep": "Sub-agent {{agent}} · step {{n}}",
     "aiThinking": "AI thinking",
+    "reasoningChain": "Reasoning process",
     "planning": "Planning",
     "assistantStreamPhase": "Assistant output",
     "toolCallsDetected": "Detected {{count}} tool call(s)",
@@ -329,6 +330,19 @@
     "agentModeMulti": "Multi-agent",
     "agentModeSingleHint": "Single-model ReAct loop for chat and tool use",
     "agentModeMultiHint": "Eino prebuilt orchestration (deep / plan_execute / supervisor) for complex tasks",
+    "reasoningModeLabel": "Model reasoning",
+    "reasoningEffortLabel": "Reasoning effort",
+    "reasoningModeDefault": "Use system default",
+    "reasoningModeOff": "Off",
+    "reasoningModeOn": "On",
+    "reasoningModeAuto": "Auto",
+    "reasoningEffortUnset": "Unspecified",
+    "reasoningCompactLabel": "Reasoning",
+    "reasoningCompactAria": "Open model reasoning options",
+    "reasoningPanelTitle": "Model reasoning",
+    "reasoningPanelHint": "Only Eino single- and multi-agent requests use these; merged with defaults in Settings.",
+    "reasoningSummaryFollow": "System",
+    "reasoningSummaryDash": "—",
     "agentModeOrchPlanExecute": "Plan-Exec",
     "agentModeOrchSupervisor": "Supervisor",
     "hitlTitle": "Human-in-the-loop",
@@ -1592,6 +1606,10 @@
     "maxTotalTokens": "Max Context Tokens",
     "maxTotalTokensPlaceholder": "120000",
     "maxTotalTokensHint": "Shared by memory compression and attack chain building. Default: 120000",
+    "openaiReasoningTitle": "Model reasoning (Eino)",
+    "openaiReasoningHint": "Applies to Eino single-agent and multi-agent only; works with chat-page reasoning controls.",
+    "openaiReasoningProfile": "Wire profile",
+    "openaiReasoningAllowClient": "Allow chat page to override reasoning options",
     "fofaBaseUrlPlaceholder": "https://fofa.info/api/v1/search/all (optional)",
     "fofaBaseUrlHint": "Leave empty for default.",
     "email": "Email",

web/static/i18n/zh-CN.json

@@ -266,6 +266,7 @@
     "planExecuteStreamPhase": "阶段输出",
     "einoSubAgentStep": "子代理 {{agent}} · 第 {{n}} 步",
     "aiThinking": "AI思考",
+    "reasoningChain": "推理过程",
     "planning": "规划中",
     "assistantStreamPhase": "助手输出",
     "toolCallsDetected": "检测到 {{count}} 个工具调用",
@@ -318,6 +319,19 @@
     "agentModeMulti": "多代理",
     "agentModeSingleHint": "单模型 ReAct 循环，适合常规对话与工具调用",
     "agentModeMultiHint": "Eino 预置编排（deep / plan_execute / supervisor），适合复杂任务",
+    "reasoningModeLabel": "模型推理",
+    "reasoningEffortLabel": "推理强度",
+    "reasoningModeDefault": "跟随系统",
+    "reasoningModeOff": "关闭",
+    "reasoningModeOn": "开启",
+    "reasoningModeAuto": "自动",
+    "reasoningEffortUnset": "不指定",
+    "reasoningCompactLabel": "推理",
+    "reasoningCompactAria": "打开模型推理选项",
+    "reasoningPanelTitle": "模型推理",
+    "reasoningPanelHint": "仅 Eino 单代理与多代理请求会带上这些参数；与系统设置中的默认值合并。",
+    "reasoningSummaryFollow": "系统",
+    "reasoningSummaryDash": "—",
     "agentModeOrchPlanExecute": "Plan-Exec",
     "agentModeOrchSupervisor": "Supervisor",
     "hitlTitle": "人机协同",
@@ -1581,6 +1595,10 @@
     "maxTotalTokens": "最大上下文 Token 数",
     "maxTotalTokensPlaceholder": "120000",
     "maxTotalTokensHint": "内存压缩和攻击链构建共用此配置，默认 120000",
+    "openaiReasoningTitle": "模型推理（Eino）",
+    "openaiReasoningHint": "仅 Eino 单代理与多代理请求生效；与对话页「模型推理」下拉配合使用。",
+    "openaiReasoningProfile": "线路 profile",
+    "openaiReasoningAllowClient": "允许对话页覆盖推理选项",
     "fofaBaseUrlPlaceholder": "https://fofa.info/api/v1/search/all（可选）",
     "fofaBaseUrlHint": "留空则使用默认地址。",
     "email": "Email",

web/static/js/auth.js

@@ -342,22 +342,27 @@ function formatMarkdown(text) {
         ALLOWED_ATTR: ['href', 'title', 'alt', 'src', 'class'],
         ALLOW_DATA_ATTR: false,
     };
-    
+
+    const raw = text == null ? '' : String(text);
+    const src = typeof window.normalizeAssistantMarkdownSource === 'function'
+        ? window.normalizeAssistantMarkdownSource(raw)
+        : raw;
+
     if (typeof DOMPurify !== 'undefined') {
-        if (typeof marked !== 'undefined' && !/<[a-z][\s\S]*>/i.test(text)) {
+        if (typeof marked !== 'undefined' && !/<[a-z][\s\S]*>/i.test(src)) {
             try {
                 marked.setOptions({
                     breaks: true,
                     gfm: true,
                 });
-                let parsedContent = marked.parse(text);
+                const parsedContent = marked.parse(src, { async: false });
                 return DOMPurify.sanitize(parsedContent, sanitizeConfig);
             } catch (e) {
                 console.error('Markdown 解析失败:', e);
-                return DOMPurify.sanitize(text, sanitizeConfig);
+                return DOMPurify.sanitize(src, sanitizeConfig);
             }
         } else {
-            return DOMPurify.sanitize(text, sanitizeConfig);
+            return DOMPurify.sanitize(src, sanitizeConfig);
         }
     } else if (typeof marked !== 'undefined') {
         try {
@@ -365,13 +370,13 @@ function formatMarkdown(text) {
                 breaks: true,
                 gfm: true,
             });
-            return marked.parse(text);
+            return marked.parse(src, { async: false });
         } catch (e) {
             console.error('Markdown 解析失败:', e);
-            return escapeHtml(text).replace(/\n/g, '<br>');
+            return escapeHtml(src).replace(/\n/g, '<br>');
         }
     } else {
-        return escapeHtml(text).replace(/\n/g, '<br>');
+        return escapeHtml(src).replace(/\n/g, '<br>');
     }
 }
 

web/static/js/chat.js

@@ -40,6 +40,8 @@ let chatAttachmentSeq = 0;
 
 // 对话模式：react = 原生 ReAct（/agent-loop）；eino_single = Eino ADK 单代理（/api/eino-agent/stream）；deep / plan_execute / supervisor = Eino 多代理（/api/multi-agent/stream，请求体 orchestration）
 const AGENT_MODE_STORAGE_KEY = 'cyberstrike-chat-agent-mode';
+const REASONING_MODE_LS = 'cyberstrike-chat-reasoning-mode';
+const REASONING_EFFORT_LS = 'cyberstrike-chat-reasoning-effort';
 const CHAT_AGENT_MODE_REACT = 'react';
 const CHAT_AGENT_MODE_EINO_SINGLE = 'eino_single';
 const CHAT_AGENT_EINO_MODES = ['deep', 'plan_execute', 'supervisor'];
@@ -492,6 +494,131 @@ function syncAgentModeFromValue(value) {
         const v = el.getAttribute('data-value');
         el.classList.toggle('selected', v === value);
     });
+    syncReasoningRowVisibility(value);
+}
+
+function syncReasoningRowVisibility(modeVal) {
+    const wrap = document.getElementById('chat-reasoning-wrapper');
+    if (!wrap) return;
+    const show = modeVal === CHAT_AGENT_MODE_EINO_SINGLE || (multiAgentAPIEnabled && chatAgentModeIsEino(modeVal));
+    wrap.style.display = show ? '' : 'none';
+    if (!show) {
+        closeChatReasoningPanel();
+    } else {
+        updateChatReasoningSummary();
+    }
+}
+
+function reasoningSummaryModeLabel(mode) {
+    const m = (mode || 'default').trim();
+    const t = (typeof window.t === 'function') ? window.t : function (k) { return k; };
+    switch (m) {
+        case 'off': return t('chat.reasoningModeOff');
+        case 'on': return t('chat.reasoningModeOn');
+        case 'auto': return t('chat.reasoningModeAuto');
+        default: return t('chat.reasoningSummaryFollow');
+    }
+}
+
+function updateChatReasoningSummary() {
+    const el = document.getElementById('chat-reasoning-summary');
+    const modeEl = document.getElementById('chat-reasoning-mode');
+    const effEl = document.getElementById('chat-reasoning-effort');
+    if (!el || !modeEl) return;
+    const mode = (modeEl.value || 'default').trim();
+    const effort = effEl && effEl.value ? String(effEl.value).trim() : '';
+    const t = (typeof window.t === 'function') ? window.t : function (k) { return k; };
+    const modePart = reasoningSummaryModeLabel(mode);
+    const effPart = effort || t('chat.reasoningSummaryDash');
+    el.textContent = modePart + ' / ' + effPart;
+}
+
+function closeChatReasoningPanel() {
+    const wrap = document.getElementById('chat-reasoning-wrapper');
+    const toggle = document.getElementById('conversation-reasoning-toggle');
+    if (wrap) wrap.classList.add('conversation-reasoning-collapsed');
+    if (toggle) toggle.setAttribute('aria-expanded', 'false');
+}
+
+function toggleConversationReasoningCard() {
+    const wrap = document.getElementById('chat-reasoning-wrapper');
+    const toggle = document.getElementById('conversation-reasoning-toggle');
+    if (!wrap || !toggle) return;
+    wrap.classList.toggle('conversation-reasoning-collapsed');
+    const collapsed = wrap.classList.contains('conversation-reasoning-collapsed');
+    toggle.setAttribute('aria-expanded', collapsed ? 'false' : 'true');
+    if (!collapsed) {
+        if (typeof closeAgentModePanel === 'function') {
+            closeAgentModePanel();
+        }
+        if (typeof closeRoleSelectionPanel === 'function') {
+            closeRoleSelectionPanel();
+        }
+        updateChatReasoningSummary();
+    }
+}
+
+function toggleChatReasoningPanel() {
+    toggleConversationReasoningCard();
+}
+
+function restoreChatReasoningControlsFromStorage() {
+    try {
+        const m = document.getElementById('chat-reasoning-mode');
+        const e = document.getElementById('chat-reasoning-effort');
+        if (m) {
+            const v = localStorage.getItem(REASONING_MODE_LS);
+            if (v && ['default', 'off', 'on', 'auto'].indexOf(v) !== -1) {
+                m.value = v;
+            }
+        }
+        if (e) {
+            const v = localStorage.getItem(REASONING_EFFORT_LS);
+            if (v !== null && ['', 'low', 'medium', 'high', 'max'].indexOf(v) !== -1) {
+                e.value = v;
+            }
+        }
+        updateChatReasoningSummary();
+    } catch (err) { /* ignore */ }
+}
+
+function persistChatReasoningPrefs() {
+    try {
+        const m = document.getElementById('chat-reasoning-mode');
+        const elEff = document.getElementById('chat-reasoning-effort');
+        if (m) localStorage.setItem(REASONING_MODE_LS, m.value || 'default');
+        if (elEff) localStorage.setItem(REASONING_EFFORT_LS, elEff.value || '');
+        updateChatReasoningSummary();
+    } catch (err) { /* ignore */ }
+}
+
+/** 供 WebShell 等复用：在 Eino 路径下返回 reasoning 请求片段或 undefined */
+function buildReasoningRequestPayload() {
+    const wrap = document.getElementById('chat-reasoning-wrapper');
+    if (!wrap || wrap.style.display === 'none') {
+        return undefined;
+    }
+    const modeEl = document.getElementById('chat-reasoning-mode');
+    const effEl = document.getElementById('chat-reasoning-effort');
+    if (!modeEl) return undefined;
+    const mode = (modeEl.value || 'default').trim();
+    const effort = effEl && effEl.value ? String(effEl.value).trim() : '';
+    if (mode === 'default' && !effort) {
+        return undefined;
+    }
+    const o = {};
+    if (mode !== 'default') o.mode = mode;
+    if (effort) o.effort = effort;
+    return Object.keys(o).length ? o : undefined;
+}
+
+if (typeof window !== 'undefined') {
+    window.persistChatReasoningPrefs = persistChatReasoningPrefs;
+    window.buildReasoningRequestPayload = buildReasoningRequestPayload;
+    window.closeChatReasoningPanel = closeChatReasoningPanel;
+    window.toggleChatReasoningPanel = toggleChatReasoningPanel;
+    window.toggleConversationReasoningCard = toggleConversationReasoningCard;
+    window.updateChatReasoningSummary = updateChatReasoningSummary;
 }
 
 function closeAgentModePanel() {
@@ -513,6 +640,9 @@ function toggleAgentModePanel() {
         closeAgentModePanel();
         return;
     }
+    if (typeof closeChatReasoningPanel === 'function') {
+        closeChatReasoningPanel();
+    }
     if (typeof closeRoleSelectionPanel === 'function') {
         closeRoleSelectionPanel();
     }
@@ -563,6 +693,8 @@ async function initChatAgentModeFromConfig() {
         } catch (e) { /* ignore */ }
         sel.value = stored;
         syncAgentModeFromValue(stored);
+        restoreChatReasoningControlsFromStorage();
+        syncReasoningRowVisibility(stored);
     } catch (e) {
         console.warn('initChatAgentModeFromConfig', e);
     }
@@ -575,6 +707,9 @@ document.addEventListener('languagechange', function () {
     if (v === CHAT_AGENT_MODE_REACT || chatAgentModeIsEinoSingle(v) || chatAgentModeIsEino(v)) {
         syncAgentModeFromValue(v);
     }
+    if (typeof updateChatReasoningSummary === 'function') {
+        updateChatReasoningSummary();
+    }
 });
 
 // 保存输入框草稿到localStorage（防抖版本）
@@ -760,6 +895,10 @@ async function sendMessage() {
             serverPath: a.serverPath
         }));
     }
+    const reasoningPayload = buildReasoningRequestPayload();
+    if (reasoningPayload) {
+        body.reasoning = reasoningPayload;
+    }
     // 发送后清空附件列表
     chatAttachments = [];
     renderChatFileChips();
@@ -1705,7 +1844,10 @@ function refreshSystemReadyMessageBubbles() {
     if (typeof marked !== 'undefined') {
         try {
             marked.setOptions({ breaks: true, gfm: true });
-            const parsed = marked.parse(text);
+            const src = typeof window.normalizeAssistantMarkdownSource === 'function'
+                ? window.normalizeAssistantMarkdownSource(text)
+                : text;
+            const parsed = marked.parse(src, { async: false });
             formattedContent = typeof DOMPurify !== 'undefined'
                 ? DOMPurify.sanitize(parsed, defaultSanitizeConfig)
                 : parsed;
@@ -1796,7 +1938,10 @@ function addMessage(role, content, mcpExecutionIds = null, progressId = null, cr
                 breaks: true,
                 gfm: true,
             });
-            return marked.parse(raw);
+            const src = typeof window.normalizeAssistantMarkdownSource === 'function'
+                ? window.normalizeAssistantMarkdownSource(raw)
+                : raw;
+            return marked.parse(src, { async: false });
         } catch (e) {
             console.error('Markdown 解析失败:', e);
             return null;
@@ -2228,6 +2373,8 @@ function renderProcessDetails(messageId, processDetails) {
             }
         } else if (eventType === 'thinking') {
             itemTitle = agPx + '🤔 ' + (typeof window.t === 'function' ? window.t('chat.aiThinking') : 'AI思考');
+        } else if (eventType === 'reasoning_chain') {
+            itemTitle = agPx + '🔗 ' + (typeof window.t === 'function' ? window.t('chat.reasoningChain') : '推理过程');
         } else if (eventType === 'planning') {
             if (typeof window.einoMainStreamPlanningTitle === 'function') {
                 itemTitle = window.einoMainStreamPlanningTitle(data);
@@ -7233,6 +7380,14 @@ document.addEventListener('click', function(event) {
             closeAgentModePanel();
         }
     }
+
+    const reasoningWrap = document.getElementById('chat-reasoning-wrapper');
+    if (reasoningWrap && reasoningWrap.style.display !== 'none' &&
+        !reasoningWrap.classList.contains('conversation-reasoning-collapsed')) {
+        if (!reasoningWrap.contains(event.target)) {
+            closeChatReasoningPanel();
+        }
+    }
 });
 
 // 创建分组

web/static/js/dashboard.js

@@ -726,8 +726,8 @@ function renderDashboardAlertBanner(stats) {
     try { sessionStorage.setItem(DASH_SESSION_ALERT_LAST_REASONS, reasonPartJoined); } catch (_) {}
 }
 
-// External MCP 健康度：从 /api/external-mcp/stats 解析出 running / total / down，
-// 决定是否在「能力总览」第 6 行显示，并把 down 数返回给 alert banner 驱动告警。
+// External MCP 健康度：从 /api/external-mcp/stats 解析（后端字段为 total/enabled/disabled/connected），
+// 决定是否在「能力总览」第 6 行显示，并把「已启用但未连接」的数量返回给 alert banner。
 function renderExternalMcpHealth(stats) {
     var row = document.getElementById('dashboard-resource-external-mcp-row');
     var textEl = document.getElementById('dashboard-resource-external-mcp-text');
@@ -738,22 +738,29 @@ function renderExternalMcpHealth(stats) {
         row.hidden = true;
         return 0;
     }
-    // 兼容多种返回字段：{ total, running, stopped/error }；常见命名都尝试一下
     var total = Number(stats.total ?? stats.Total ?? 0) || 0;
-    var running = Number(stats.running ?? stats.Running ?? 0) || 0;
+    var enabled = Number(stats.enabled ?? stats.Enabled ?? 0) || 0;
+    // 后端用 connected 表示已连接数；兼容旧字段 running
+    var connected = Number(stats.connected ?? stats.Connected ??
+        stats.running ?? stats.Running ?? 0) || 0;
     if (total === 0) {
         row.hidden = true;
         return 0;
     }
-    var down = Math.max(0, total - running);
+    // 未配置任何「已启用」的外部 MCP 时不展示健康行，也不告警（与 MCP 管理页口径一致）
+    if (enabled === 0) {
+        row.hidden = true;
+        return 0;
+    }
+    var down = Math.max(0, enabled - connected);
     row.hidden = false;
-    textEl.textContent = formatNumber(running) + ' / ' + formatNumber(total);
+    textEl.textContent = formatNumber(connected) + ' / ' + formatNumber(enabled);
     if (healthEl) {
         healthEl.classList.remove('is-ok', 'is-warning', 'is-danger');
         if (down === 0) {
             healthEl.classList.add('is-ok');
             healthEl.textContent = dt('dashboard.mcpAllRunning', null, '全部运行');
-        } else if (down < total) {
+        } else if (down < enabled) {
             healthEl.classList.add('is-warning');
             healthEl.textContent = dt('dashboard.mcpPartialDown', { count: down },
                 down + ' 个未运行');

web/static/js/monitor.js

@@ -273,6 +273,116 @@ function escapeHtmlLocal(text) {
     return div.innerHTML;
 }
 
+/** fenced 块占位（BMP 私用区，正文几乎不会出现） */
+const _MD_FENCE_PRE = '\n\uE000CSAI_FENCE_';
+const _MD_FENCE_SUF = '_\uE000\n';
+
+function _maskFencedCodeBlocksForMdPreprocess(md) {
+    const blocks = [];
+    const masked = String(md).replace(/```[\s\S]*?```/g, (m) => {
+        const i = blocks.length;
+        blocks.push(m);
+        return _MD_FENCE_PRE + i + _MD_FENCE_SUF;
+    });
+    return { masked, blocks };
+}
+
+function _unmaskFencedCodeBlocksAfterMdPreprocess(s, blocks) {
+    let out = s;
+    for (let i = 0; i < blocks.length; i++) {
+        out = out.split(_MD_FENCE_PRE + i + _MD_FENCE_SUF).join(blocks[i]);
+    }
+    return out;
+}
+
+/**
+ * 模型/网关偶发把「思考」混进正文，用伪 XML 包裹（如 &lt;redacted_thinking&gt;…&lt;/redacted_thinking&gt;）。
+ * 与 Markdown 列表混排时，结束标签常被吞进 &lt;li&gt;，其后 **、` 等行内语法全部无法解析；成对块整段移除。
+ * @param {string} segment
+ * @returns {string}
+ */
+function _stripXmlReasoningWrappersForMarkdown(segment) {
+    let t = String(segment);
+    const tags = ['redacted_thinking', 'redacted_reasoning'];
+    for (let i = 0; i < tags.length; i++) {
+        const name = tags[i];
+        const re = new RegExp('<\\s*' + name + '\\b[^>]*>[\\s\\S]*?<\\s*/\\s*' + name + '\\s*>', 'gi');
+        t = t.replace(re, '\n\n');
+    }
+    return t.replace(/\n{3,}/g, '\n\n');
+}
+
+/**
+ * 解除 LLM 常用的块级 HTML 外壳（`<div>`、`<p>`、`<section>`、`<article>`、`<main>`）。
+ * 整段包在块级标签里时，CommonMark 不会在块内再解析 Markdown，导致 **、` 原样显示。
+ */
+function _unwrapHtmlBlockWrappersForMarkdown(segment) {
+    let s = segment;
+    let prev;
+    for (let i = 0; i < 30 && s !== prev; i++) {
+        prev = s;
+        s = s.replace(/<div(?:\s[^>]*)?>([\s\S]*?)<\/div>/gi, (_, inner) => String(inner).trim() + '\n\n');
+        s = s.replace(/<p(?:\s[^>]*)?>([\s\S]*?)<\/p>/gi, (_, inner) => String(inner).trim() + '\n\n');
+        s = s.replace(/<section(?:\s[^>]*)?>([\s\S]*?)<\/section>/gi, (_, inner) => String(inner).trim() + '\n\n');
+        s = s.replace(/<article(?:\s[^>]*)?>([\s\S]*?)<\/article>/gi, (_, inner) => String(inner).trim() + '\n\n');
+        s = s.replace(/<main(?:\s[^>]*)?>([\s\S]*?)<\/main>/gi, (_, inner) => String(inner).trim() + '\n\n');
+        s = s.replace(/\n{3,}/g, '\n\n');
+    }
+    return s;
+}
+
+/**
+ * 将 HTML 列表 / 粘连的 `<li>` 还原为 Markdown 列表行，并去掉外层 `<ul>`，便于 marked 解析行内 **、` `
+ * @param {string} segment
+ * @returns {string}
+ */
+function _flattenOrphanHtmlLiInMarkdown(segment) {
+    let s = segment;
+    s = s.replace(/<li(?:\s[^>]*)?>([\s\S]*?)<\/li>/gi, (_, inner) => {
+        const body = String(inner).trim().replace(/\s*\n\s*/g, ' ');
+        return '- ' + body + '\n';
+    });
+    s = s.replace(/<\/?ul(?:\s[^>]*)?>/gi, '\n');
+    s = s.replace(/<\/?ol(?:\s[^>]*)?>/gi, '\n');
+    s = s.replace(/([0-9A-Za-z_\u4e00-\u9fff])\s*<li(?:\s[^>]*)?>\s*/g, (_, ch) => ch + '\n- ');
+    return s.replace(/\n{3,}/g, '\n\n');
+}
+
+/** 行首 Unicode 项目符号 → Markdown 列表 `- `（模型常用 • 而非 `-`） */
+function _normalizeUnicodeBulletMarkersToMdDash(segment) {
+    return segment
+        .replace(/^\s*\u2022\s+/gm, '- ')
+        .replace(/^\s*\u00b7\s+/gm, '- ');
+}
+
+/**
+ * 解析前归一化助手 Markdown：去掉零宽字符，NFKC 将全角 * ` _ 等转为 ASCII，
+ * 避免 marked 无法识别强调/行内代码而原样显示 **、反引号；
+ * 并移除 &lt;redacted_thinking&gt; 等伪 XML 思考块、修正块级 HTML（`<div>`/`<p>`/…、`<ul>`/`<li>`）与 Unicode 项目符号 `•`，避免块级 HTML 吞掉 inline 解析。
+ * @param {string|null|undefined} text
+ * @returns {string}
+ */
+function normalizeAssistantMarkdownSource(text) {
+    if (text == null) return '';
+    let s = String(text);
+    s = s.replace(/[\u200B-\u200D\u200E\u200F\uFEFF\u2060]/g, '');
+    try {
+        s = s.normalize('NFKC');
+    } catch (e) {
+        /* ignore */
+    }
+    s = _stripXmlReasoningWrappersForMarkdown(s);
+    const fb = _maskFencedCodeBlocksForMdPreprocess(s);
+    s = _unwrapHtmlBlockWrappersForMarkdown(fb.masked);
+    s = _flattenOrphanHtmlLiInMarkdown(s);
+    s = _normalizeUnicodeBulletMarkersToMdDash(s);
+    s = _unmaskFencedCodeBlocksAfterMdPreprocess(s, fb.blocks);
+    return s;
+}
+if (typeof window !== 'undefined') {
+    window.normalizeAssistantMarkdownSource = normalizeAssistantMarkdownSource;
+}
+
 /**
  * 与 internal/openai.normalizeStreamingDelta 一致：兼容网关/模型返回「累计全文」或整包重发，
  * 避免前端 buffer += chunk 与后端已归一化的增量叠加导致逐段重复（如「响应中显示了响应中显示了」）。
@@ -316,10 +426,11 @@ function setTimelineItemContentStreamRich(contentEl, html) {
 
 function formatAssistantMarkdownContent(text) {
     const raw = text == null ? '' : String(text);
+    const src = normalizeAssistantMarkdownSource(raw);
     if (typeof marked !== 'undefined') {
         try {
             marked.setOptions({ breaks: true, gfm: true });
-            const parsed = marked.parse(raw);
+            const parsed = marked.parse(src, { async: false });
             if (typeof DOMPurify !== 'undefined') {
                 return DOMPurify.sanitize(parsed, assistantMarkdownSanitizeConfig);
             }
@@ -1222,12 +1333,41 @@ function handleStreamEvent(event, progressElement, progressId,
             });
             break;
         }
+
+        case 'eino_trace_run':
+        case 'eino_trace_start':
+        case 'eino_trace_end':
+        case 'eino_trace_error': {
+            const d = event.data || {};
+            const comp = d.component != null ? String(d.component) : '';
+            const name = d.name != null ? String(d.name) : '';
+            let glyph = '◆';
+            if (event.type === 'eino_trace_run') glyph = '●';
+            else if (event.type === 'eino_trace_start') glyph = '▶';
+            else if (event.type === 'eino_trace_end') glyph = '■';
+            else if (event.type === 'eino_trace_error') glyph = '✖';
+            const title = '[Eino] ' + glyph + ' ' + (comp || 'component') + (name ? '/' + name : '');
+            const parts = [];
+            if (d.runId) parts.push('run=' + String(d.runId));
+            if (d.spanId) parts.push('span=' + String(d.spanId));
+            if (d.parentSpanId) parts.push('parent=' + String(d.parentSpanId));
+            if (d.inputSummary) parts.push(String(d.inputSummary));
+            if (d.outputSummary) parts.push(String(d.outputSummary));
+            if (d.error) parts.push(String(d.error));
+            if (event.message && String(event.message).trim()) parts.push(String(event.message));
+            const body = parts.join(' · ');
+            addTimelineItem(timeline, 'progress', { title, message: body, data: d });
+            break;
+        }
             
-        case 'thinking_stream_start': {
+        case 'thinking_stream_start':
+        case 'reasoning_chain_stream_start': {
             const d = event.data || {};
             const streamId = d.streamId || null;
             if (!streamId) break;
 
+            const timelineType = event.type === 'reasoning_chain_stream_start' ? 'reasoning_chain' : 'thinking';
+
             let state = thinkingStreamStateByProgressId.get(progressId);
             if (!state) {
                 state = new Map();
@@ -1246,9 +1386,12 @@ function handleStreamEvent(event, progressElement, progressId,
                 }
                 break;
             }
-            const thinkBase = typeof window.t === 'function' ? window.t('chat.aiThinking') : 'AI思考';
-            const title = timelineAgentBracketPrefix(d) + '🤔 ' + thinkBase;
-            const itemId = addTimelineItem(timeline, 'thinking', {
+            const labelBase = typeof window.t === 'function'
+                ? window.t(timelineType === 'reasoning_chain' ? 'chat.reasoningChain' : 'chat.aiThinking')
+                : (timelineType === 'reasoning_chain' ? '推理过程' : 'AI思考');
+            const emoji = timelineType === 'reasoning_chain' ? '🔗' : '🤔';
+            const title = timelineAgentBracketPrefix(d) + emoji + ' ' + labelBase;
+            const itemId = addTimelineItem(timeline, timelineType, {
                 title: title,
                 message: ' ',
                 data: d
@@ -1257,7 +1400,8 @@ function handleStreamEvent(event, progressElement, progressId,
             break;
         }
 
-        case 'thinking_stream_delta': {
+        case 'thinking_stream_delta':
+        case 'reasoning_chain_stream_delta': {
             const d = event.data || {};
             const streamId = d.streamId || null;
             if (!streamId) break;
@@ -1281,7 +1425,9 @@ function handleStreamEvent(event, progressElement, progressId,
         }
 
         case 'thinking':
-            // 如果本 thinking 是由 thinking_stream_* 聚合出来的（带 streamId），避免重复创建 timeline item
+        case 'reasoning_chain': {
+            const timelineType = event.type === 'reasoning_chain' ? 'reasoning_chain' : 'thinking';
+            // 若已由 *_stream_* 聚合（带 streamId），避免重复创建 timeline item
             if (event.data && event.data.streamId) {
                 const streamId = event.data.streamId;
                 const state = thinkingStreamStateByProgressId.get(progressId);
@@ -1303,12 +1449,17 @@ function handleStreamEvent(event, progressElement, progressId,
                 }
             }
 
-            addTimelineItem(timeline, 'thinking', {
-                title: timelineAgentBracketPrefix(event.data) + '🤔 ' + (typeof window.t === 'function' ? window.t('chat.aiThinking') : 'AI思考'),
+            const labelBase = typeof window.t === 'function'
+                ? window.t(timelineType === 'reasoning_chain' ? 'chat.reasoningChain' : 'chat.aiThinking')
+                : (timelineType === 'reasoning_chain' ? '推理过程' : 'AI思考');
+            const emoji = timelineType === 'reasoning_chain' ? '🔗' : '🤔';
+            addTimelineItem(timeline, timelineType, {
+                title: timelineAgentBracketPrefix(event.data) + emoji + ' ' + labelBase,
                 message: event.message,
                 data: event.data
             });
             break;
+        }
             
         case 'tool_calls_detected':
             addTimelineItem(timeline, 'tool_calls_detected', {
@@ -2475,7 +2626,7 @@ function addTimelineItem(timeline, type, options) {
     `;
     
     // 根据类型添加详细内容
-    if ((type === 'thinking' || type === 'planning') && options.message) {
+    if ((type === 'thinking' || type === 'reasoning_chain' || type === 'planning') && options.message) {
         const streamBody = typeof formatTimelineStreamBody === 'function'
             ? formatTimelineStreamBody(options.message, options.data)
             : options.message;
@@ -2530,6 +2681,8 @@ function addTimelineItem(timeline, type, options) {
                 ${escapeHtml(options.message || taskCancelledLabel)}
             </div>
         `;
+    } else if (type === 'progress' && options.message) {
+        content += `<div class="timeline-item-content timeline-eino-trace"><pre class="tool-result">${escapeHtml(options.message)}</pre></div>`;
     } else if (type === 'user_interrupt_continue' && options.message) {
         const streamBody = typeof formatTimelineStreamBody === 'function'
             ? formatTimelineStreamBody(options.message, options.data)
@@ -3410,6 +3563,8 @@ function refreshProgressAndTimelineI18n() {
             } else {
                 titleSpan.textContent = ap + '\uD83E\uDD14 ' + _t('chat.aiThinking');
             }
+        } else if (type === 'reasoning_chain') {
+            titleSpan.textContent = ap + '\uD83D\uDD17 ' + _t('chat.reasoningChain');
         } else if (type === 'planning') {
             if (item.dataset.orchestration && typeof einoMainStreamPlanningTitle === 'function') {
                 titleSpan.textContent = einoMainStreamPlanningTitle({

web/static/js/roles.js

@@ -256,6 +256,9 @@ function toggleRoleSelectionPanel() {
         if (typeof closeAgentModePanel === 'function') {
             closeAgentModePanel();
         }
+        if (typeof closeChatReasoningPanel === 'function') {
+            closeChatReasoningPanel();
+        }
         panel.style.display = 'flex'; // 使用flex布局
         // 添加打开状态的视觉反馈
         if (roleSelectorBtn) {

web/static/js/settings.js

@@ -159,6 +159,27 @@ async function loadConfig(loadTools = true) {
         if (maxTokensEl) {
             maxTokensEl.value = currentConfig.openai.max_total_tokens || 120000;
         }
+        const orm = currentConfig.openai && currentConfig.openai.reasoning ? currentConfig.openai.reasoning : {};
+        const orModeEl = document.getElementById('openai-reasoning-mode');
+        if (orModeEl) {
+            const mv = (orm.mode || 'auto').toString().trim().toLowerCase();
+            orModeEl.value = ['auto', 'on', 'off'].includes(mv) ? mv : 'auto';
+        }
+        const orEffEl = document.getElementById('openai-reasoning-effort');
+        if (orEffEl) {
+            const ev = (orm.effort || '').toString().trim().toLowerCase();
+            orEffEl.value = ['', 'low', 'medium', 'high', 'max'].includes(ev) ? ev : '';
+        }
+        const orProfEl = document.getElementById('openai-reasoning-profile');
+        if (orProfEl) {
+            const pv = (orm.profile || 'auto').toString().trim().toLowerCase();
+            const ok = ['auto', 'deepseek_compat', 'openai_compat', 'output_config_effort'];
+            orProfEl.value = ok.includes(pv) ? pv : 'auto';
+        }
+        const orAllowEl = document.getElementById('openai-reasoning-allow-client');
+        if (orAllowEl) {
+            orAllowEl.checked = orm.allow_client_reasoning !== false;
+        }
 
         // 填充FOFA配置
         const fofa = currentConfig.fofa || {};
@@ -1065,13 +1086,22 @@ async function applySettings() {
         };
         
         const wecomAgentIdVal = document.getElementById('robot-wecom-agent-id')?.value.trim();
+        const prevOpenai = (currentConfig && currentConfig.openai) ? currentConfig.openai : {};
         const config = {
             openai: {
+                ...prevOpenai,
                 provider: provider,
                 api_key: apiKey,
                 base_url: baseUrl,
                 model: model,
-                max_total_tokens: parseInt(document.getElementById('openai-max-total-tokens')?.value) || 120000
+                max_total_tokens: parseInt(document.getElementById('openai-max-total-tokens')?.value) || 120000,
+                reasoning: {
+                    ...(prevOpenai.reasoning || {}),
+                    mode: document.getElementById('openai-reasoning-mode')?.value || 'auto',
+                    effort: (document.getElementById('openai-reasoning-effort')?.value || '').trim(),
+                    profile: document.getElementById('openai-reasoning-profile')?.value || 'auto',
+                    allow_client_reasoning: document.getElementById('openai-reasoning-allow-client')?.checked !== false
+                }
             },
             fofa: {
                 email: document.getElementById('fofa-email')?.value.trim() || '',

web/static/js/webshell.js

@@ -1658,6 +1658,8 @@ function buildWebshellTimelineItemFromDetail(detail) {
         title = ap + ((typeof window.t === 'function') ? window.t('chat.iterationRound', { n: data.iteration || 1 }) : ('第 ' + (data.iteration || 1) + ' 轮迭代'));
     } else if (eventType === 'thinking') {
         title = ap + '🤔 ' + ((typeof window.t === 'function') ? window.t('chat.aiThinking') : 'AI 思考');
+    } else if (eventType === 'reasoning_chain') {
+        title = ap + '🔗 ' + ((typeof window.t === 'function') ? window.t('chat.reasoningChain') : '推理过程');
     } else if (eventType === 'tool_calls_detected') {
         title = ap + '🔧 ' + ((typeof window.t === 'function') ? window.t('chat.toolCallsDetected', { count: data.count || 0 }) : ('检测到 ' + (data.count || 0) + ' 个工具调用'));
     } else if (eventType === 'tool_call') {
@@ -2847,6 +2849,12 @@ function runWebshellAiSend(conn, inputEl, sendBtn, messagesContainer) {
         if (info && info.orchestration) {
             body.orchestration = info.orchestration;
         }
+        if (typeof window.buildReasoningRequestPayload === 'function') {
+            var rp = window.buildReasoningRequestPayload();
+            if (rp) {
+                body.reasoning = rp;
+            }
+        }
         return apiFetch(info.path, {
             method: 'POST',
             headers: { 'Content-Type': 'application/json' },
@@ -2953,17 +2961,19 @@ function runWebshellAiSend(conn, inputEl, sendBtn, messagesContainer) {
                         appendTimelineItem('iteration', '🔍 ' + iterTitle, iterMessage, _ed);
                         if (!streamingTarget) assistantDiv.textContent = '…';
 
-                    // ─── Thinking (non-stream + stream) ───
-                    } else if (_et === 'thinking_stream_start' && _ed.streamId) {
+                    // ─── Thinking / reasoning_chain（推理过程，reasoning_content） ───
+                    } else if ((_et === 'thinking_stream_start' || _et === 'reasoning_chain_stream_start') && _ed.streamId) {
+                        var isRcStart = _et === 'reasoning_chain_stream_start';
                         if (wsThinkingStreams.has(_ed.streamId)) {
                             var tsExist = wsThinkingStreams.get(_ed.streamId);
                             tsExist.buf = '';
                             if (tsExist.body) tsExist.body.textContent = '';
                         } else {
-                        var thinkSLabel = wsTOr('chat.aiThinking', 'AI 思考');
+                        var thinkSLabel = wsTOr(isRcStart ? 'chat.reasoningChain' : 'chat.aiThinking', isRcStart ? '推理过程' : 'AI 思考');
+                        var thinkEmoji = isRcStart ? '🔗' : '🤔';
                         var thinkSItem = document.createElement('div');
-                        thinkSItem.className = 'webshell-ai-timeline-item webshell-ai-timeline-thinking';
-                        thinkSItem.innerHTML = '<span class="webshell-ai-timeline-title">' + escapeHtml(webshellAgentPx(_ed) + '🤔 ' + thinkSLabel) + '</span>';
+                        thinkSItem.className = 'webshell-ai-timeline-item webshell-ai-timeline-' + (isRcStart ? 'reasoning_chain' : 'thinking');
+                        thinkSItem.innerHTML = '<span class="webshell-ai-timeline-title">' + escapeHtml(webshellAgentPx(_ed) + thinkEmoji + ' ' + thinkSLabel) + '</span>';
                         var thinkSPre = document.createElement('div');
                         thinkSPre.className = 'webshell-ai-timeline-msg webshell-thinking-stream-body';
                         thinkSItem.appendChild(thinkSPre);
@@ -2972,7 +2982,7 @@ function runWebshellAiSend(conn, inputEl, sendBtn, messagesContainer) {
                         wsThinkingStreams.set(_ed.streamId, { el: thinkSItem, body: thinkSPre, buf: '' });
                         }
                         if (!streamingTarget) assistantDiv.textContent = '…';
-                    } else if (_et === 'thinking_stream_delta' && _ed.streamId) {
+                    } else if ((_et === 'thinking_stream_delta' || _et === 'reasoning_chain_stream_delta') && _ed.streamId) {
                         var tsD = wsThinkingStreams.get(_ed.streamId);
                         if (tsD) {
                             var normT = (typeof window.normalizeStreamingDeltaJs === 'function')
@@ -2985,7 +2995,7 @@ function runWebshellAiSend(conn, inputEl, sendBtn, messagesContainer) {
                             }
                         }
                         if (!streamingTarget) assistantDiv.textContent = '…';
-                    } else if (_et === 'thinking_stream_end' && _ed.streamId) {
+                    } else if ((_et === 'thinking_stream_end' || _et === 'reasoning_chain_stream_end') && _ed.streamId) {
                         var tsE = wsThinkingStreams.get(_ed.streamId);
                         if (tsE) {
                             var fullThink = (_em != null && _em !== '') ? String(_em) : tsE.buf;
@@ -2996,13 +3006,15 @@ function runWebshellAiSend(conn, inputEl, sendBtn, messagesContainer) {
                             }
                             wsThinkingStreams.delete(_ed.streamId);
                         }
-                    } else if (_et === 'thinking' && _em) {
+                    } else if ((_et === 'thinking' || _et === 'reasoning_chain') && _em) {
                         // 如果有 streamId 且已存在流式条目，跳过避免重复
                         if (_ed.streamId && wsThinkingStreams.has(_ed.streamId)) {
-                            // 已由 thinking_stream_* 处理
+                            // 已由 *_stream_* 处理
                         } else {
-                            var thinkLabel = wsTOr('chat.aiThinking', 'AI 思考');
-                            appendTimelineItem('thinking', webshellAgentPx(_ed) + '🤔 ' + thinkLabel, _em, _ed);
+                            var isRc = _et === 'reasoning_chain';
+                            var thinkLabel = wsTOr(isRc ? 'chat.reasoningChain' : 'chat.aiThinking', isRc ? '推理过程' : 'AI 思考');
+                            var thinkEm = isRc ? '🔗' : '🤔';
+                            appendTimelineItem(isRc ? 'reasoning_chain' : 'thinking', webshellAgentPx(_ed) + thinkEm + ' ' + thinkLabel, _em, _ed);
                         }
                         if (!streamingTarget) assistantDiv.textContent = '…';
 

web/templates/index.html

@@ -792,11 +792,51 @@
                                     <div id="conversations-list" class="conversations-list"></div>
                                 </div>
                             </div>
+                            <div id="chat-reasoning-wrapper" class="chat-reasoning-wrapper conversation-reasoning-card conversation-reasoning-collapsed" style="display: none;">
+                                <button type="button" id="conversation-reasoning-toggle" class="conversation-reasoning-card-header" onclick="toggleConversationReasoningCard()" aria-expanded="false" aria-controls="conversation-reasoning-body" data-i18n="chat.reasoningCompactAria" data-i18n-attr="aria-label,title" data-i18n-skip-text="true" aria-label="模型推理选项" title="模型推理选项">
+                                    <div class="conversation-reasoning-heading">
+                                        <span class="conversation-reasoning-icon" aria-hidden="true">
+                                            <svg width="16" height="16" viewBox="0 0 24 24" fill="none" xmlns="http://www.w3.org/2000/svg">
+                                                <circle cx="11" cy="11" r="7" stroke="currentColor" stroke-width="1.75"/>
+                                                <path d="M16 16l5 5" stroke="currentColor" stroke-width="1.75" stroke-linecap="round"/>
+                                            </svg>
+                                        </span>
+                                        <div class="conversation-reasoning-heading-text">
+                                            <span class="conversation-reasoning-title" data-i18n="chat.reasoningPanelTitle">模型推理</span>
+                                            <span id="chat-reasoning-summary" class="conversation-reasoning-summary"></span>
+                                        </div>
+                                    </div>
+                                </button>
+                                <div id="conversation-reasoning-body" class="conversation-reasoning-body" role="region">
+                                    <p class="chat-reasoning-panel-hint" data-i18n="chat.reasoningPanelHint">仅 Eino 请求生效，与系统设置中的默认值合并。</p>
+                                    <div class="chat-reasoning-fields">
+                                        <div class="chat-reasoning-field">
+                                            <label class="chat-reasoning-field-label" for="chat-reasoning-mode"><span data-i18n="chat.reasoningModeLabel">模式</span></label>
+                                            <select id="chat-reasoning-mode" class="chat-reasoning-select" onchange="persistChatReasoningPrefs()">
+                                                <option value="default" data-i18n="chat.reasoningModeDefault">跟随系统</option>
+                                                <option value="off" data-i18n="chat.reasoningModeOff">关闭</option>
+                                                <option value="on" data-i18n="chat.reasoningModeOn">开启</option>
+                                                <option value="auto" data-i18n="chat.reasoningModeAuto">自动</option>
+                                            </select>
+                                        </div>
+                                        <div class="chat-reasoning-field">
+                                            <label class="chat-reasoning-field-label" for="chat-reasoning-effort"><span data-i18n="chat.reasoningEffortLabel">推理强度</span></label>
+                                            <select id="chat-reasoning-effort" class="chat-reasoning-select" onchange="persistChatReasoningPrefs()">
+                                                <option value="" data-i18n="chat.reasoningEffortUnset">不指定</option>
+                                                <option value="low">low</option>
+                                                <option value="medium">medium</option>
+                                                <option value="high">high</option>
+                                                <option value="max">max</option>
+                                            </select>
+                                        </div>
+                                    </div>
+                                </div>
+                            </div>
                             <div class="hitl-sidebar-card hitl-sidebar-collapsed" id="hitl-sidebar-card">
                                 <div class="hitl-sidebar-card-header" onclick="toggleHitlSidebarCard()">
                                     <div class="hitl-sidebar-heading">
                                         <span class="hitl-sidebar-icon" aria-hidden="true">
-                                            <svg width="18" height="18" viewBox="0 0 24 24" fill="none" xmlns="http://www.w3.org/2000/svg">
+                                            <svg width="16" height="16" viewBox="0 0 24 24" fill="none" xmlns="http://www.w3.org/2000/svg">
                                                 <path d="M12 2L4 5v6.09c0 5.05 3.41 9.76 8 10.91 4.59-1.15 8-5.86 8-10.91V5l-8-3z" stroke="currentColor" stroke-width="1.75" stroke-linejoin="round"/>
                                                 <path d="M9.5 12.5l2 2 3-4" stroke="currentColor" stroke-width="1.75" stroke-linecap="round" stroke-linejoin="round"/>
                                             </svg>
@@ -894,6 +934,8 @@
                             <div id="active-tasks-bar" class="active-tasks-bar"></div>
                             <div id="chat-messages" class="chat-messages"></div>
                             <div id="chat-input-container" class="chat-input-container">
+                                <div class="chat-input-primary-row">
+                                <div class="chat-input-leading">
                                 <div class="role-selector-wrapper">
                                     <button id="role-selector-btn" class="role-selector-btn" onclick="toggleRoleSelectionPanel()" data-i18n="chat.selectRole" data-i18n-attr="title" title="选择角色">
                                         <span id="role-selector-icon" class="role-selector-icon">🔵</span>
@@ -979,6 +1021,7 @@
                                     </div>
                                     <input type="hidden" id="agent-mode-select" value="react" autocomplete="off">
                                 </div>
+                                </div>
                                 <div class="chat-input-with-files">
                                     <div id="chat-file-list" class="chat-file-list" aria-label="已选文件列表"></div>
                                     <div id="chat-attachment-progress" class="chat-upload-progress-row" hidden role="status" aria-live="polite">
@@ -1002,6 +1045,7 @@
                                         <path d="M5 12h14M12 5l7 7-7 7" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round"/>
                                     </svg>
                                 </button>
+                                </div>
                             </div>
                         </div>
                     </div>
@@ -1989,6 +2033,38 @@
                                             <input type="number" id="openai-max-total-tokens" data-i18n="settingsBasic.maxTotalTokensPlaceholder" data-i18n-attr="placeholder" placeholder="120000" min="1000" step="1000" />
                                             <small style="color: var(--text-muted, #718096); font-size: 0.75rem;" data-i18n="settingsBasic.maxTotalTokensHint">内存压缩和攻击链构建共用此配置，默认 120000</small>
                                         </div>
+                                        <div class="form-group">
+                                            <label data-i18n="settingsBasic.openaiReasoningTitle">模型推理（Eino）</label>
+                                            <small class="form-hint" data-i18n="settingsBasic.openaiReasoningHint">仅影响 Eino 单代理与多代理；对话页可覆盖（见下方「允许对话覆盖」）。</small>
+                                            <div style="display: flex; flex-wrap: wrap; gap: 10px; margin-top: 8px; align-items: center;">
+                                                <label for="openai-reasoning-mode" style="font-size: 0.8125rem;" data-i18n="chat.reasoningModeLabel">模式</label>
+                                                <select id="openai-reasoning-mode" style="min-width: 120px; padding: 0.35rem 0.5rem; border-radius: 6px; border: 1px solid var(--border-color, #e2e8f0);">
+                                                    <option value="auto" data-i18n="chat.reasoningModeAuto">自动</option>
+                                                    <option value="on" data-i18n="chat.reasoningModeOn">开启</option>
+                                                    <option value="off" data-i18n="chat.reasoningModeOff">关闭</option>
+                                                </select>
+                                                <label for="openai-reasoning-effort" style="font-size: 0.8125rem;" data-i18n="chat.reasoningEffortLabel">强度</label>
+                                                <select id="openai-reasoning-effort" style="min-width: 100px; padding: 0.35rem 0.5rem; border-radius: 6px; border: 1px solid var(--border-color, #e2e8f0);">
+                                                    <option value="" data-i18n="chat.reasoningEffortUnset">不指定</option>
+                                                    <option value="low">low</option>
+                                                    <option value="medium">medium</option>
+                                                    <option value="high">high</option>
+                                                    <option value="max">max</option>
+                                                </select>
+                                                <label for="openai-reasoning-profile" style="font-size: 0.8125rem;" data-i18n="settingsBasic.openaiReasoningProfile">线路</label>
+                                                <select id="openai-reasoning-profile" style="min-width: 140px; padding: 0.35rem 0.5rem; border-radius: 6px; border: 1px solid var(--border-color, #e2e8f0);">
+                                                    <option value="auto">auto</option>
+                                                    <option value="deepseek_compat">deepseek_compat</option>
+                                                    <option value="openai_compat">openai_compat</option>
+                                                    <option value="output_config_effort">output_config_effort</option>
+                                                </select>
+                                            </div>
+                                            <label class="checkbox-label" style="margin-top: 8px;">
+                                                <input type="checkbox" id="openai-reasoning-allow-client" class="modern-checkbox" checked />
+                                                <span class="checkbox-custom"></span>
+                                                <span class="checkbox-text" data-i18n="settingsBasic.openaiReasoningAllowClient">允许对话页覆盖推理选项</span>
+                                            </label>
+                                        </div>
                                         <div style="display: flex; align-items: center; gap: 8px; margin-top: 2px;">
                                             <a href="javascript:void(0)" id="test-openai-btn" onclick="testOpenAIConnection()" style="font-size: 0.8125rem; color: var(--accent-color, #3182ce); text-decoration: none; cursor: pointer; user-select: none;" data-i18n="settingsBasic.testConnection">测试连接</a>
                                             <span id="test-openai-result" style="font-size: 0.8125rem;"></span>