Merge main and preserve task navigation with severity filtering

agent/finding_cases.go

@@ -45,13 +45,13 @@ func findingCaseWorkflow(key string, tools []actool.CoreTool) ([]actool.CoreTool
 	}
 	return tools, `
 
-【同一漏洞的多次上报整理】保留本次独立报告，不覆盖其他原报告。默认报告完成即停止的旧指引在此扩展：先保存本次报告，然后 search_finding_duplicates(finding_id=独立记录ID) 分页读取同任务候选摘要，按需 get_finding_record 查看证据。仅凭漏洞类型/标题相似或高中低等级不能归并；须核对目标、接口、参数、权限边界、根因及实际验证证据。
-【判断准则】接口、路径、后端、读/写影响、评级不同，都不能单独作为不重复的理由；同一登录入口、同一令牌或同一漏洞类型也不能单独证明同一根因。比较实际缺失的鉴权检查、对象权限边界、会话门或共同组件，并引用每条记录中的具体证据。读和写可能是同一权限检查缺失的不同影响；也可能是独立 handler 的独立缺陷。只有证据证明共同实际根因才自动归并，证据证明不同根因才判独立；存在关联但无法确认时必须 suggest_finding_merge，说明已知共同点、缺失证据和待确认问题，不用“路径不同/影响不同”代替根因判断。明确同一缺陷时 merge_finding_records(task_id,finding_ids,title,reason)，不确定时 suggest_finding_merge 留待人工确认，不能强行归并。不重复上报、不删除记录、不重新探测目标。人工选择整理时只能判断所选记录，不能加入未选记录（既有组成员可作为报告上下文）。
-归并成功或已有case_id时，get_finding_case 记录实际version，分页读取全部成员，按需get_finding_record、get_finding_traffic读取原报告、PoC及已保存流量。update_finding_case_report保存完整统一报告（包含有内容的概述、影响、前置条件、复现步骤、证据、PoC、根因分析、修复建议章节，不得半篇截断）和基于已验证影响的severity及severity_reason；不能直接取成员最高等级。集中给出前置条件、完整执行顺序、输入输出如何衔接、可运行的完整脚本及命令，不把必要步骤分散到原报告；缺失条件、未验证步骤如实标明。版本冲突重新读取材料并生成，不能仅换version重试。各成员仍有独立报告、评级和证据。没有明确重复则保留独立记录。工具被关闭时不绕过配置。`
+【同一漏洞的多次上报整理】保留本次独立报告，不覆盖其他原报告。默认报告完成即停止的旧指引在此扩展：先保存本次报告，然后 search_finding_duplicates(finding_id=独立记录ID) 分页读取同任务候选摘要（按 score 排序；strong=同漏洞类型且资产重叠，优先核对；weak=未同时具备同漏洞类型和资产重叠，结合 hit_reasons 核对，仅作参考），按需 get_finding_record 查看证据。仅凭漏洞类型、标题、令牌、登录入口相似或高中低等级不能归并；须核对目标、接口、参数、权限边界、根因及实际验证证据。只能处理同一任务，尊重人工否决和 finding_case_blocks。
+【判断准则】若证据表明端点/路由、处理器或后端组件、易受影响参数，或缺失检查属于不同代码路径，可以据此判定 independent；必须说明各自证据。仅凭路径、接口、登录入口、令牌、标题或漏洞类型相同/不同，都不能单独证明同一根因或不同根因。只有根因明确相同才直接调用 merge_finding_records(task_id,finding_ids,title,reason)。只有存在具体共同点（两条记录都证据化地指向同一端点、参数、组件或缺失检查），但缺少确认共同根因的关键证据时，才调用 suggest_finding_merge；一次提交覆盖2..20条同任务记录的一个分组，不要拆成两两建议。仅共享漏洞类型或资产没有具体共同点，不得提交建议；没有具体共同点时保持 independent，不要因为不确定就 suggest。不得重复上报、删除记录或重新探测目标。人工选择整理时只能判断所选记录，不能加入未选记录（既有组成员可作为报告上下文）。
+归并成功或已有case_id时，get_finding_case 记录实际version，分页读取全部成员，按需 get_finding_record、get_finding_traffic 读取原报告、PoC及已保存流量。根因明确时归并不被统一报告阻塞，应随后尽力用 update_finding_case_report 保存统一报告；如果本次无法完成，文件夹显示“统一报告待更新”，由后续 reporter 运行或人工补齐。统一报告在保存时仍须包含有内容的概述、影响、前置条件、复现步骤、证据、PoC、根因分析、修复建议章节，不得半篇截断；评级及 severity_reason 必须基于已验证影响，不能直接取成员最高等级。集中给出前置条件、完整执行顺序、输入输出如何衔接、可运行的完整脚本及命令，不把必要步骤分散到原报告；缺失条件、未验证步骤如实标明。版本冲突重新读取材料并生成，不能仅换version重试。各成员仍有独立报告、评级和证据。工具被关闭时不绕过配置。`
 }
 
 const historicalFindingReviewGuidance = `
 
 【历史整理模式，优先于旧报告撰写及自动绑流量指引】本次只整理已经存在的上报。原始报告即使为空或待更新，也只记录缺失情况，绝不调用 update_finding_report、report_finding、bind_finding_traffic，不修改原始标题、评级、状态、证据或报告，不执行 Bash、写文件或重新探测目标。
-先逐条 get_finding_record 读取所选记录，已有原报告直接作为判断材料复用，不重复撰写。只在所选范围内检查同一实际缺陷，可按需读取已存在的证据。按上述根因准则形成明确归并/独立/疑似建议；无法确认时保存建议供人工确认。
-已有或新建文件夹必须读取全部成员并写入完整、最新的统一报告；统一报告是单独的材料，不覆盖原报告。最后必须调用 complete_finding_case_review，为每条所选 finding_id 提交 verdict（independent/merged/suggested）、reason 和 root_cause_evidence，并提供总结。根因证据应说明具体相同/不同的检查及证据来源；疑似建议明确待补证据。服务端验收通过后才可宣布完成。工具不可用或验收失败时如实报告，不能只用自然语言声称完成。`
+先逐条 get_finding_record 读取所选记录，已有原报告直接作为判断材料复用，不重复撰写。只在所选范围内检查同一实际缺陷，可按需读取已存在的证据。结构化证据显示不同端点/路由、处理器或后端组件、参数或缺失检查属于不同代码路径时，可判 independent；仅类型、资产、标题、令牌或登录入口相似不够。只有具体共同点明确且关键根因证据缺失时才保存一个覆盖2..20条记录的疑似分组，没有具体共同点就保持 independent。历史整理只能处理 selected scope，必须遵守同一任务限制和人工否决。
+历史模式禁止调用 update_finding_report、report_finding、bind_finding_traffic，不修改原始标题、评级、状态、证据或报告，不执行 Bash、写文件或重新探测目标。已有或新建文件夹应读取全部成员并尽力用 update_finding_case_report 写入完整、最新的统一报告；归并不因报告暂缺而阻塞，未完成时文件夹显示“统一报告待更新”，可由后续运行或人工补齐。统一报告是单独的材料，不覆盖原报告。最后必须调用 complete_finding_case_review，为每条所选 finding_id 提交 verdict（independent/merged/suggested）、reason 和 root_cause_evidence，并提供总结。根因证据应说明具体相同/不同的检查及证据来源；疑似建议明确待补证据。服务端验收通过后才可宣布完成。人工整理只能判断 selected scope，不能加入未选记录。工具不可用或验收失败时如实报告，不能只用自然语言声称完成。`

agent/finding_cases_test.go

@@ -46,7 +46,7 @@ func TestHistoricalFindingReviewFiltersMutationAndProbeTools(t *testing.T) {
 			t.Fatal("historical mutation capability", tool.Name())
 		}
 	}
-	if len(out) != 6 || def.FindingGuidance != "" || !strings.Contains(def.CaseGuidance, "历史整理模式") || !strings.Contains(def.CaseGuidance, "接口、路径、后端、读/写影响、评级不同") {
+	if len(out) != 6 || def.FindingGuidance != "" || !strings.Contains(def.CaseGuidance, "历史整理模式") || !strings.Contains(def.CaseGuidance, "端点/路由") || !strings.Contains(def.CaseGuidance, "2..20") {
 		t.Fatal("missing historical isolation", def)
 	}
 	normal, _, cl := AugmentTools(t.Context(), "reporter", tools)

agent/goals.go

@@ -82,7 +82,7 @@ type GoalSpec struct {
 // shares the rate limiter, gets recorded by llmrec, and participates in LLM
 // failover instead of quietly bypassing all three.
 //
-// desc is the task's free-text description (背景：靶标范围/flag 数量/交战说明等).
+// desc is the task's free-text description (背景：靶标范围/数量要求/交战说明等).
 // It is fed alongside the goal so the decomposer no longer splits blind — the
 // prompt still forbids inventing anything the two texts don't state.
 //
@@ -140,7 +140,7 @@ func DecomposeGoalsWithProvider(ctx context.Context, prov llm.Provider, dataDir,
 	}
 	userMsg := "任务目标：\n" + goalText
 	if d := strings.TrimSpace(desc); d != "" {
-		userMsg += "\n\n任务描述（背景信息，可能含靶标范围/flag 数量/交战说明；仅供参考，不要臆造其中未提及的内容）：\n" + d
+		userMsg += "\n\n任务描述（背景信息，可能含靶标范围/数量要求/交战说明；仅供参考，不要臆造其中未提及的内容）：\n" + d
 	}
 	// Use captureRun so every LLM step is emitted as an activity record (visible in
 	// the plan tab under the round-0 marker). Falls back gracefully when emit is nil.

agent/planner.go

@@ -313,7 +313,7 @@ const plannerDefaultTmpl = `你是一个网络安全平台授权渗透测试系
    - **要更深细节才按需调**：list_facts（分页，最新在前，默认 20，可 q 过滤、before 翻页，带 total/has_more）、list_findings（全部漏洞）、node_detail(id)（完整证据/详情；列表/recent_facts 只给摘要）、list_assets（pull：q 搜索、type/company_id/task_id 过滤、分页，或 id/ids 直取）、asset_neighbors。资产全局共享，别默认拉全量。
 
 2. **判目标（核心职责）**：goals 字段已含目标与状态；对已被某发现/事实证明的未达成目标，调 prove_goal(goal_id, evidence_id, reason) 标 met。**当你标记的恰是最后一个未完成目标时，系统自动判定整个任务完成**——收官只由逐个 prove_goal 驱动，没有别的"一键完成"手段。
-   - ⚠️ **量化验收核对（严禁提前盖章）**：目标含可量化条件（覆盖度达 X%、拿 N 个 flag、获得某权限）时，prove_goal 前【必须】核对上方 graph_overview 的实测值（coverage.pct、findings_total 计数等）：未达标就【禁止】prove_goal，改派意图补差；不得以"大体达成/核心已拿下"为由提前标 met。例：要求覆盖度 100% 而实测 coverage.pct=40% → 未达成，继续派补测意图。
+   - ⚠️ **量化验收核对（严禁提前盖章）**：目标含可量化条件（覆盖度达 X%、达到指定数量、获得某权限）时，prove_goal 前【必须】核对上方 graph_overview 的实测值（coverage.pct、findings_total 计数等）：未达标就【禁止】prove_goal，改派意图补差；不得以"大体达成/核心已拿下"为由提前标 met。例：要求覆盖度 100% 而实测 coverage.pct=40% → 未达成，继续派补测意图。
 
 3. **（可选，仅开局、极轻量）探测理解**：仅当图里几乎还没有 fact（recent_facts 基本为空、任务刚开始）、仅凭态势无法把初始意图说具体时，才用 Bash 等对目标做极少量、只读的探测（如 1–2 次 curl 看首页/指纹）。**唯一合法产物是一句更精准的意图描述**——绝不是漏洞的发现/验证/利用，也不是端点/目录/参数的枚举结果（那些是 worker 的活，写成意图派下去）。三条硬边界：
    - 图里已有 worker 产出的 fact（facts>0 / recent_facts 非空）→【禁止】再自己探测，一切判断基于已有 fact，本轮产物只能是"派新意图"或"结束"；想深挖某线索 → 派意图让 worker 去查，不是自己 curl。

agent/promptcatalog.go

@@ -15,7 +15,7 @@ package agent
 const autoDefaultTmpl = `你是 **Auto**，这个渗透测试平台的「操作助手」。你不亲自渗透，而是**用工具操作平台**、按用户指令把事情办好。
 
 你能做的（取决于给你开放了哪些工具）：
-1. **任务操作**：list_tasks 看全局、spawn_task 起子任务、get_task_graph / list_task_findings 读某任务的进展与漏洞(含 flag)、get_task_worker_trace 看某个 work 的执行过程、pause_task 暂停、add_task_hint 给任务注入提示。
+1. **任务操作**：list_tasks 看全局、spawn_task 起子任务、get_task_graph / list_task_findings 读某任务的进展与漏洞、get_task_worker_trace 看某个 work 的执行过程、pause_task 暂停、add_task_hint 给任务注入提示。
 2. **平台管理**：create_skill / update_skill 建改技能；create_custom_tool / update_custom_tool 建改自定义工具(command/script/http)；create_mcp / update_mcp 建改 MCP 服务器。
 
 原则：

agent/tools.go

@@ -1208,7 +1208,7 @@ func (t *ToolSet) proveGoal() actool.CoreTool {
 }
 
 func (t *ToolSet) goalMet() actool.CoreTool {
-	return writeTool("goal_met", "【立即结束整个任务】——仅当你确认任务的【全部目标都已真正达成、整体收官】时才调（注意是任务【整体】完成；仅仅达成了其中某一个目标/某一个 flag/某一个漏洞【不算】——那种情况用 prove_goal 标记该目标即可）。⚠️它不是用来“结束本轮规划”的：本轮没有新意图要派、或在等 worker 产出，都【直接结束本轮即可，不要调本工具】（0 个意图是完全正常的）。正常判定优先用 prove_goal 逐个证明目标；goal_met 只是绕过逐个证明、直接从全局收官的手段。",
+	return writeTool("goal_met", "【立即结束整个任务】——仅当你确认任务的【全部目标都已真正达成、整体收官】时才调（注意是任务【整体】完成；仅仅达成了其中某一个目标/某一个漏洞【不算】——那种情况用 prove_goal 标记该目标即可）。⚠️它不是用来“结束本轮规划”的：本轮没有新意图要派、或在等 worker 产出，都【直接结束本轮即可，不要调本工具】（0 个意图是完全正常的）。正常判定优先用 prove_goal 逐个证明目标；goal_met 只是绕过逐个证明、直接从全局收官的手段。",
 		obj(map[string]any{"reason": str("达成理由（必须是目标真正达成的证据，不能是“本轮无新方向”这类结束本轮的理由）")}, "reason"),
 		func(_ context.Context, in json.RawMessage) (actool.Result, error) {
 			var a struct{ Reason string }

db/finding_cases.go

@@ -41,13 +41,14 @@ type FindingCaseEvent struct {
 	CreatedAt  time.Time       `json:"created_at"`
 }
 type FindingCaseSuggestion struct {
-	ID      int64  `json:"id,string"`
-	TaskID  int64  `json:"task_id,string"`
-	LeftID  int64  `json:"left_id,string"`
-	RightID int64  `json:"right_id,string"`
-	Title   string `json:"title"`
-	Reason  string `json:"reason"`
-	State   string `json:"state"`
+	ID         int64    `json:"id,string"`
+	TaskID     int64    `json:"task_id,string"`
+	LeftID     int64    `json:"left_id,string"`
+	RightID    int64    `json:"right_id,string"`
+	Title      string   `json:"title"`
+	Reason     string   `json:"reason"`
+	State      string   `json:"state"`
+	FindingIDs []string `json:"finding_ids"`
 }
 type FindingCaseRow struct {
 	Case       *FindingCase `json:"case,omitempty"`
@@ -198,9 +199,9 @@ func lockCaseTask(tx *sql.Tx, taskID int64) error {
 	return tx.QueryRow(`SELECT id FROM tasks WHERE id=$1 FOR UPDATE`, taskID).Scan(&id)
 }
 func (d *DB) MergeFindingCase(ctx context.Context, taskID int64, ids []int64, title, reason, actor string) (int64, error) {
-	return d.mergeFindingCase(ctx, taskID, ids, title, reason, actor, 0)
+	return d.mergeFindingCase(ctx, taskID, ids, title, reason, actor, 0, nil)
 }
-func (d *DB) mergeFindingCase(ctx context.Context, taskID int64, ids []int64, title, reason, actor string, suggestionID int64) (int64, error) {
+func (d *DB) mergeFindingCase(ctx context.Context, taskID int64, ids []int64, title, reason, actor string, suggestionID int64, dropped []int64) (int64, error) {
 	ids, err := sortedFindingIDs(ids)
 	if err != nil {
 		return 0, err
@@ -208,6 +209,7 @@ func (d *DB) mergeFindingCase(ctx context.Context, taskID int64, ids []int64, ti
 	if strings.TrimSpace(title) == "" || strings.TrimSpace(reason) == "" {
 		return 0, errors.New("必须提供漏洞名称及基于证据的归并理由")
 	}
+	keptIDs := append([]int64(nil), ids...)
 	var cid int64
 	err = d.WithEvidenceTx(ctx, func(tx *sql.Tx) error {
 		if err := lockCaseTask(tx, taskID); err != nil {
@@ -308,11 +310,69 @@ func (d *DB) mergeFindingCase(ctx context.Context, taskID int64, ids []int64, ti
 				return err
 			}
 		}
-		_, err := tx.Exec(`UPDATE finding_case_suggestions SET state='accepted' WHERE state='pending' AND left_id IN(SELECT value::bigint FROM jsonb_array_elements_text($1::jsonb)) AND right_id IN(SELECT value::bigint FROM jsonb_array_elements_text($1::jsonb))`, string(raw))
-		return err
+		for _, excluded := range dropped {
+			if all[excluded] {
+				return fmt.Errorf("排除的记录 #%d 已与所选记录在同一文件夹，请先移出该文件夹后再确认", excluded)
+			}
+			for _, kept := range keptIDs {
+				if _, err := tx.Exec(`INSERT INTO finding_case_blocks(left_id,right_id,reason)
+                    VALUES(LEAST($1::bigint,$2::bigint),GREATEST($1::bigint,$2::bigint),'人工确认分组时排除')
+                    ON CONFLICT DO NOTHING`, excluded, kept); err != nil {
+					return err
+				}
+			}
+		}
+		if suggestionID > 0 {
+			if _, err := tx.Exec(`UPDATE finding_case_suggestions SET state='accepted' WHERE id=$1 AND state='pending'`, suggestionID); err != nil {
+				return err
+			}
+		}
+		return sweepFindingCaseSuggestions(tx, taskID)
 	})
 	return cid, err
 }
+
+// sweepFindingCaseSuggestions reconciles pending cards after a merge or veto.
+// Cross-task member rows are ignored so a malformed legacy row cannot leak into
+// a task's suggestions.
+func sweepFindingCaseSuggestions(tx *sql.Tx, taskID int64) error {
+	if _, err := tx.Exec(`UPDATE finding_case_suggestions s SET state='accepted'
+        WHERE s.task_id=$1 AND s.state='pending'
+          AND (SELECT count(*) FROM finding_case_suggestion_members sm
+               WHERE sm.suggestion_id=s.id
+                 AND EXISTS (SELECT 1 FROM findings f WHERE f.id=sm.finding_id AND f.task_id=s.task_id)) >= 2
+          AND (SELECT count(*) FROM finding_case_suggestion_members sm
+               JOIN findings f ON f.id=sm.finding_id AND f.task_id=s.task_id
+               JOIN finding_case_members cm ON cm.finding_id=sm.finding_id
+               JOIN finding_cases c ON c.id=cm.case_id AND c.active
+               WHERE sm.suggestion_id=s.id) >= 2
+          AND (SELECT count(DISTINCT cm.case_id) FROM finding_case_suggestion_members sm
+               JOIN findings f ON f.id=sm.finding_id AND f.task_id=s.task_id
+               JOIN finding_case_members cm ON cm.finding_id=sm.finding_id
+               JOIN finding_cases c ON c.id=cm.case_id AND c.active
+               WHERE sm.suggestion_id=s.id) = 1
+          AND NOT EXISTS (SELECT 1 FROM finding_case_suggestion_members sm
+                          JOIN findings f ON f.id=sm.finding_id AND f.task_id=s.task_id
+                          LEFT JOIN finding_case_members cm ON cm.finding_id=sm.finding_id
+                          LEFT JOIN finding_cases c ON c.id=cm.case_id AND c.active
+                          WHERE sm.suggestion_id=s.id AND c.id IS NULL)`, taskID); err != nil {
+		return err
+	}
+	_, err := tx.Exec(`UPDATE finding_case_suggestions s SET state='obsolete'
+        WHERE s.task_id=$1 AND s.state='pending' AND (
+          (SELECT count(*) FROM finding_case_suggestion_members sm
+           JOIN findings f ON f.id=sm.finding_id AND f.task_id=s.task_id
+           WHERE sm.suggestion_id=s.id) < 2
+          OR EXISTS (SELECT 1
+                    FROM finding_case_suggestion_members a
+                    JOIN finding_case_suggestion_members b
+                      ON b.suggestion_id=a.suggestion_id AND a.finding_id < b.finding_id
+                    JOIN findings fa ON fa.id=a.finding_id AND fa.task_id=s.task_id
+                    JOIN findings fb ON fb.id=b.finding_id AND fb.task_id=s.task_id
+                    JOIN finding_case_blocks bl ON bl.left_id=a.finding_id AND bl.right_id=b.finding_id
+                    WHERE a.suggestion_id=s.id))`, taskID)
+	return err
+}
 func (d *DB) RemoveFindingCaseMember(ctx context.Context, cid, fid int64, reason string) error {
 	c, err := d.GetFindingCase(cid)
 	if err != nil {
@@ -355,8 +415,10 @@ func (d *DB) RemoveFindingCaseMember(ctx context.Context, cid, fid int64, reason
 		if _, err := tx.Exec(`INSERT INTO finding_case_events(case_id,action,actor,reason,finding_ids) VALUES($1,'detach','human',$2,jsonb_build_array($3::bigint))`, cid, reason, fid); err != nil {
 			return err
 		}
-		_, err := tx.Exec(`DELETE FROM finding_case_members WHERE case_id=$1 AND finding_id=$2`, cid, fid)
-		return err
+		if _, err := tx.Exec(`DELETE FROM finding_case_members WHERE case_id=$1 AND finding_id=$2`, cid, fid); err != nil {
+			return err
+		}
+		return sweepFindingCaseSuggestions(tx, origin)
 	})
 }
 func (d *DB) UpdateFindingCaseReport(ctx context.Context, cid, version int64, title, report, severity, reason string) error {
@@ -382,81 +444,345 @@ func (d *DB) UpdateFindingCaseReport(ctx context.Context, cid, version int64, ti
 		return err
 	})
 }
-func (d *DB) SuggestFindingCase(ctx context.Context, taskID, left, right int64, title, reason string) (int64, error) {
+
+type suggestionQuery interface {
+	Query(string, ...any) (*sql.Rows, error)
+}
+
+// suggestionMembers reads only surviving members in the suggestion's task. The
+// anchor fallback keeps rows created by older binaries resolvable if a database
+// was upgraded before the idempotent backfill ran.
+func suggestionMembers(q suggestionQuery, suggestionID, taskID int64) ([]int64, error) {
+	rows, err := q.Query(`SELECT sm.finding_id
+        FROM finding_case_suggestion_members sm
+        JOIN findings f ON f.id=sm.finding_id AND f.task_id=$2
+        WHERE sm.suggestion_id=$1 ORDER BY sm.finding_id`, suggestionID, taskID)
+	if err != nil {
+		return nil, err
+	}
+	ids := []int64{}
+	for rows.Next() {
+		var id int64
+		if err := rows.Scan(&id); err != nil {
+			rows.Close()
+			return nil, err
+		}
+		ids = append(ids, id)
+	}
+	err = rows.Err()
+	rows.Close()
+	if err != nil {
+		return nil, err
+	}
+	if len(ids) > 0 {
+		return ids, nil
+	}
+	rows, err = q.Query(`SELECT x.finding_id FROM (
+            SELECT s.left_id AS finding_id FROM finding_case_suggestions s WHERE s.id=$1
+            UNION SELECT s.right_id FROM finding_case_suggestions s WHERE s.id=$1
+        ) x JOIN findings f ON f.id=x.finding_id AND f.task_id=$2 ORDER BY x.finding_id`, suggestionID, taskID)
+	if err != nil {
+		return nil, err
+	}
+	defer rows.Close()
+	for rows.Next() {
+		var id int64
+		if err := rows.Scan(&id); err != nil {
+			return nil, err
+		}
+		ids = append(ids, id)
+	}
+	return ids, rows.Err()
+}
+
+// SuggestFindingCase stores a pending group of 2..20 same-task findings. It
+// refreshes an exact group, reuses a pending superset, and supersedes strict
+// subsets while holding the task advisory lock.
+func (d *DB) SuggestFindingCase(ctx context.Context, taskID int64, findingIDs []int64, title, reason string) (int64, error) {
 	if strings.TrimSpace(title) == "" || strings.TrimSpace(reason) == "" {
 		return 0, errors.New("建议需要名称和具体证据理由")
 	}
-	ids, err := sortedFindingIDs([]int64{left, right})
+	if len(findingIDs) > 20 {
+		return 0, errors.New("疑似重复分组最多20条记录")
+	}
+	seen := make(map[int64]bool, len(findingIDs))
+	for _, fid := range findingIDs {
+		if seen[fid] {
+			return 0, errors.New("疑似重复分组中的记录编号必须唯一")
+		}
+		seen[fid] = true
+	}
+	ids, err := sortedFindingIDs(findingIDs)
 	if err != nil {
 		return 0, err
 	}
-	var id int64
+	var suggestionID int64
 	err = d.WithEvidenceTx(ctx, func(tx *sql.Tx) error {
 		if err := lockCaseTask(tx, taskID); err != nil {
 			return err
 		}
+		caseIDs := map[int64]bool{}
+		sameActiveCase := true
+		var firstCase int64
 		for _, fid := range ids {
 			var tid sql.NullInt64
-			if err := tx.QueryRow(`SELECT task_id FROM findings WHERE id=$1`, fid).Scan(&tid); err != nil {
+			var cid int64
+			if err := tx.QueryRow(`SELECT f.task_id,COALESCE(CASE WHEN c.active THEN m.case_id ELSE 0 END,0)
+				FROM findings f LEFT JOIN finding_case_members m ON m.finding_id=f.id
+                LEFT JOIN finding_cases c ON c.id=m.case_id WHERE f.id=$1`, fid).Scan(&tid, &cid); err != nil {
 				return err
 			}
 			if !tid.Valid || tid.Int64 != taskID {
 				return errors.New("只能建议同一任务的归并")
 			}
+			if cid == 0 {
+				sameActiveCase = false
+			} else if firstCase == 0 {
+				firstCase = cid
+				caseIDs[cid] = true
+			} else if firstCase != cid {
+				sameActiveCase = false
+			}
 		}
-		var blocked bool
-		if err := tx.QueryRow(`SELECT EXISTS(SELECT 1 FROM finding_case_blocks WHERE left_id=$1 AND right_id=$2)`, ids[0], ids[1]).Scan(&blocked); err != nil {
+		if sameActiveCase && firstCase > 0 && len(caseIDs) == 1 {
+			return errors.New("已在同一文件夹")
+		}
+
+		raw, _ := json.Marshal(ids)
+		rows, err := tx.Query(`SELECT left_id,right_id FROM finding_case_blocks
+            WHERE left_id IN (SELECT value::bigint FROM jsonb_array_elements_text($1::jsonb))
+              AND right_id IN (SELECT value::bigint FROM jsonb_array_elements_text($1::jsonb))
+            ORDER BY left_id,right_id`, string(raw))
+		if err != nil {
 			return err
 		}
-		if blocked {
-			return errors.New("人工已否决该归并")
+		blocked := []string{}
+		for rows.Next() {
+			var left, right int64
+			if err := rows.Scan(&left, &right); err != nil {
+				rows.Close()
+				return err
+			}
+			blocked = append(blocked, fmt.Sprintf("#%d/#%d", left, right))
 		}
-		return tx.QueryRow(`INSERT INTO finding_case_suggestions(task_id,left_id,right_id,title,reason) VALUES($1,$2,$3,$4,$5) ON CONFLICT(left_id,right_id) DO UPDATE SET reason=EXCLUDED.reason WHERE finding_case_suggestions.state='pending' RETURNING id`, taskID, ids[0], ids[1], title, reason).Scan(&id)
+		err = rows.Err()
+		rows.Close()
+		if err != nil {
+			return err
+		}
+		if len(blocked) > 0 {
+			return fmt.Errorf("人工已否决这些记录的归并：%s；请去掉其中一方后重新提交", strings.Join(blocked, "、"))
+		}
+
+		type pendingGroup struct {
+			id      int64
+			members []int64
+			title   string
+			reason  string
+		}
+		rows, err = tx.Query(`SELECT id,title,reason FROM finding_case_suggestions
+            WHERE task_id=$1 AND state='pending' ORDER BY id`, taskID)
+		if err != nil {
+			return err
+		}
+		groups := []pendingGroup{}
+		for rows.Next() {
+			var g pendingGroup
+			if err := rows.Scan(&g.id, &g.title, &g.reason); err != nil {
+				rows.Close()
+				return err
+			}
+			groups = append(groups, g)
+		}
+		err = rows.Err()
+		rows.Close()
+		if err != nil {
+			return err
+		}
+		for i := range groups {
+			groups[i].members, err = suggestionMembers(tx, groups[i].id, taskID)
+			if err != nil {
+				return err
+			}
+		}
+		want := map[int64]bool{}
+		for _, fid := range ids {
+			want[fid] = true
+		}
+		contains := func(group []int64, set map[int64]bool) bool {
+			for _, fid := range group {
+				if !set[fid] {
+					return false
+				}
+			}
+			return true
+		}
+		exact, superset := int64(0), int64(0)
+		supersede := []int64{}
+		for _, g := range groups {
+			gm := map[int64]bool{}
+			for _, fid := range g.members {
+				gm[fid] = true
+			}
+			switch {
+			case len(g.members) == len(ids) && contains(ids, gm) && contains(g.members, want):
+				if exact == 0 {
+					exact = g.id
+				} else {
+					supersede = append(supersede, g.id)
+				}
+			case len(g.members) > len(ids) && contains(ids, gm):
+				if superset == 0 {
+					superset = g.id
+				}
+			case len(g.members) < len(ids) && contains(g.members, want):
+				supersede = append(supersede, g.id)
+			}
+		}
+		for _, sid := range supersede {
+			if _, err := tx.Exec(`UPDATE finding_case_suggestions SET state='superseded' WHERE id=$1 AND state='pending'`, sid); err != nil {
+				return err
+			}
+		}
+		if exact > 0 {
+			suggestionID = exact
+			_, err = tx.Exec(`UPDATE finding_case_suggestions SET title=$2,reason=$3 WHERE id=$1`, exact, title, reason)
+			return err
+		}
+		if superset > 0 {
+			suggestionID = superset
+			return nil
+		}
+		if err := tx.QueryRow(`INSERT INTO finding_case_suggestions(task_id,left_id,right_id,title,reason)
+            VALUES($1,$2,$3,$4,$5) RETURNING id`, taskID, ids[0], ids[1], title, reason).Scan(&suggestionID); err != nil {
+			return err
+		}
+		for _, fid := range ids {
+			if _, err := tx.Exec(`INSERT INTO finding_case_suggestion_members(suggestion_id,finding_id) VALUES($1,$2) ON CONFLICT DO NOTHING`, suggestionID, fid); err != nil {
+				return err
+			}
+		}
+		return nil
 	})
-	return id, err
+	return suggestionID, err
 }
+
 func (d *DB) FindingCaseSuggestions(taskID int64) ([]FindingCaseSuggestion, error) {
-	rows, err := d.Query(`SELECT id,task_id,left_id,right_id,title,reason,state FROM finding_case_suggestions WHERE state='pending' AND ($1::bigint=0 OR task_id=$1) ORDER BY id DESC`, taskID)
+	rows, err := d.Query(`SELECT id,task_id,left_id,right_id,title,reason,state
+        FROM finding_case_suggestions WHERE state='pending' AND ($1::bigint=0 OR task_id=$1) ORDER BY id DESC`, taskID)
 	if err != nil {
 		return nil, err
 	}
-	defer rows.Close()
-	out := []FindingCaseSuggestion{}
+	items := []FindingCaseSuggestion{}
 	for rows.Next() {
 		var s FindingCaseSuggestion
 		if err := rows.Scan(&s.ID, &s.TaskID, &s.LeftID, &s.RightID, &s.Title, &s.Reason, &s.State); err != nil {
+			rows.Close()
 			return nil, err
 		}
+		items = append(items, s)
+	}
+	err = rows.Err()
+	rows.Close()
+	if err != nil {
+		return nil, err
+	}
+	out := []FindingCaseSuggestion{}
+	for _, s := range items {
+		ids, err := suggestionMembers(d, s.ID, s.TaskID)
+		if err != nil {
+			return nil, err
+		}
+		if len(ids) < 2 {
+			continue
+		}
+		s.LeftID, s.RightID = ids[0], ids[1]
+		for _, fid := range ids {
+			s.FindingIDs = append(s.FindingIDs, strconv.FormatInt(fid, 10))
+		}
 		out = append(out, s)
 	}
-	return out, rows.Err()
+	return out, nil
 }
-func (d *DB) ResolveFindingCaseSuggestion(ctx context.Context, id int64, accept bool) (int64, error) {
+
+func (d *DB) ResolveFindingCaseSuggestion(ctx context.Context, id int64, accept bool, exclude []int64) (int64, error) {
 	var s FindingCaseSuggestion
-	if err := d.QueryRow(`SELECT task_id,left_id,right_id,title,reason,state FROM finding_case_suggestions WHERE id=$1`, id).Scan(&s.TaskID, &s.LeftID, &s.RightID, &s.Title, &s.Reason, &s.State); err != nil {
+	if err := d.QueryRow(`SELECT task_id,title,reason,state FROM finding_case_suggestions WHERE id=$1`, id).Scan(&s.TaskID, &s.Title, &s.Reason, &s.State); err != nil {
 		return 0, err
 	}
 	if s.State != "pending" {
 		return 0, ErrFindingCaseConflict
 	}
+	members, err := suggestionMembers(d, id, s.TaskID)
+	if err != nil {
+		return 0, err
+	}
+	if len(members) < 2 {
+		return 0, errors.New("疑似重复分组成员不足")
+	}
+	memberSet := map[int64]bool{}
+	for _, fid := range members {
+		memberSet[fid] = true
+	}
+	excludeSet := map[int64]bool{}
+	for _, fid := range exclude {
+		if !memberSet[fid] {
+			return 0, fmt.Errorf("排除记录 #%d 不属于该分组", fid)
+		}
+		excludeSet[fid] = true
+	}
+	kept := []int64{}
+	for _, fid := range members {
+		if !excludeSet[fid] {
+			kept = append(kept, fid)
+		}
+	}
 	if accept {
-		return d.mergeFindingCase(ctx, s.TaskID, []int64{s.LeftID, s.RightID}, s.Title, s.Reason, "human", id)
+		if len(kept) < 2 {
+			return 0, errors.New("至少保留两条记录")
+		}
+		return d.mergeFindingCase(ctx, s.TaskID, kept, s.Title, s.Reason, "human", id, exclude)
+	}
+	if len(exclude) > 0 {
+		return 0, errors.New("否决分组时不能排除记录")
 	}
 	return 0, d.WithEvidenceTx(ctx, func(tx *sql.Tx) error {
 		if err := lockCaseTask(tx, s.TaskID); err != nil {
 			return err
 		}
-		res, err := tx.Exec(`UPDATE finding_case_suggestions SET state='rejected' WHERE id=$1 AND state='pending'`, id)
+		var state string
+		if err := tx.QueryRow(`SELECT state FROM finding_case_suggestions WHERE id=$1 FOR UPDATE`, id).Scan(&state); err != nil {
+			return err
+		}
+		if state != "pending" {
+			return ErrFindingCaseConflict
+		}
+		members, err := suggestionMembers(tx, id, s.TaskID)
 		if err != nil {
 			return err
 		}
-		if n, _ := res.RowsAffected(); n != 1 {
-			return ErrFindingCaseConflict
+		for i := 0; i < len(members); i++ {
+			for j := i + 1; j < len(members); j++ {
+				var leftCase, rightCase sql.NullInt64
+				if err := tx.QueryRow(`SELECT m.case_id FROM finding_case_members m JOIN finding_cases c ON c.id=m.case_id AND c.active WHERE m.finding_id=$1`, members[i]).Scan(&leftCase); err != nil && !errors.Is(err, sql.ErrNoRows) {
+					return err
+				}
+				if err := tx.QueryRow(`SELECT m.case_id FROM finding_case_members m JOIN finding_cases c ON c.id=m.case_id AND c.active WHERE m.finding_id=$1`, members[j]).Scan(&rightCase); err != nil && !errors.Is(err, sql.ErrNoRows) {
+					return err
+				}
+				if leftCase.Valid && rightCase.Valid && leftCase.Int64 == rightCase.Int64 {
+					continue
+				}
+				if _, err := tx.Exec(`INSERT INTO finding_case_blocks(left_id,right_id,reason)
+                    VALUES(LEAST($1::bigint,$2::bigint),GREATEST($1::bigint,$2::bigint),'人工否决') ON CONFLICT DO NOTHING`, members[i], members[j]); err != nil {
+					return err
+				}
+			}
 		}
-		if _, err := tx.Exec(`INSERT INTO finding_case_blocks(left_id,right_id,reason) VALUES($1,$2,'人工否决') ON CONFLICT DO NOTHING`, s.LeftID, s.RightID); err != nil {
+		if _, err := tx.Exec(`UPDATE finding_case_suggestions SET state='rejected' WHERE id=$1`, id); err != nil {
 			return err
 		}
-		return nil
+		return sweepFindingCaseSuggestions(tx, s.TaskID)
 	})
 }
 
@@ -479,13 +805,13 @@ func (d *DB) DistinctFindingStats(taskID string) (*FindingDistinctStats, error)
 }
 
 // Compact candidate pages exclude PoC/report bodies and do not treat severity as identity.
-func (d *DB) FindingCaseCandidates(fid int64, page, size int, scope ...[]int64) ([]map[string]any, int, error) {
+func (d *DB) FindingCaseCandidates(fid int64, page, size int, scope ...[]int64) ([]map[string]any, int, int, error) {
 	f, err := d.FindingCaseSummary(fid)
 	if err != nil {
-		return nil, 0, err
+		return nil, 0, 0, err
 	}
 	if f == nil || f.TaskID == nil {
-		return nil, 0, errors.New("漏洞任务不可用")
+		return nil, 0, 0, errors.New("漏洞任务不可用")
 	}
 	if page < 1 {
 		page = 1
@@ -505,32 +831,58 @@ func (d *DB) FindingCaseCandidates(fid int64, page, size int, scope ...[]int64)
 	if len(scope) > 0 {
 		raw, err := json.Marshal(scope[0])
 		if err != nil {
-			return nil, 0, err
+			return nil, 0, 0, err
 		}
 		where += ` AND f.id IN (SELECT jsonb_array_elements_text($6::jsonb)::bigint)`
 		args = append(args, string(raw))
 	}
-	var total int
-	if err := d.QueryRow(`SELECT count(*) FROM findings f`+where, args...).Scan(&total); err != nil {
-		return nil, 0, err
+	classHit := `(trim(f.vulnclass)<>'' AND lower(f.vulnclass)=lower($3))`
+	assetHit := `EXISTS(SELECT 1 FROM jsonb_array_elements(COALESCE(NULLIF(f.asset_ids,'null'::jsonb),'[]'::jsonb)) a WHERE a.value IN (SELECT value FROM jsonb_array_elements($5::jsonb)))`
+	nameHit := `(trim(f.name)<>'' AND lower(f.name)=lower($4))`
+	scoreExpr := fmt.Sprintf(`(CASE WHEN %s THEN 2 ELSE 0 END + CASE WHEN %s THEN 2 ELSE 0 END + CASE WHEN %s THEN 1 ELSE 0 END + CASE WHEN (%s AND %s) THEN 3 ELSE 0 END)`, classHit, assetHit, nameHit, classHit, assetHit)
+	strongExpr := fmt.Sprintf(`(%s AND %s)`, classHit, assetHit)
+	base := `WITH candidates AS (SELECT f.id,f.name,f.vulnclass,f.severity,left(f.summary,400) AS summary,f.asset_ids,
+        COALESCE(m.case_id,0) AS case_id,
+        ` + scoreExpr + ` AS score,
+        ` + strongExpr + ` AS strong,
+        ARRAY_REMOVE(ARRAY[
+            CASE WHEN ` + classHit + ` THEN '同漏洞类型' END,
+            CASE WHEN ` + assetHit + ` THEN '资产重叠' END,
+            CASE WHEN ` + nameHit + ` THEN '同名称' END
+        ]::text[], NULL) AS hit_reasons
+        FROM findings f LEFT JOIN finding_case_members m ON m.finding_id=f.id` + where + `)`
+	var total, strongTotal int
+	if err := d.QueryRow(base+` SELECT count(*),count(*) FILTER(WHERE strong) FROM candidates`, args...).Scan(&total, &strongTotal); err != nil {
+		return nil, 0, 0, err
 	}
 	args = append(args, size, (page-1)*size)
-	rows, err := d.Query(`SELECT f.id,f.name,f.vulnclass,f.severity,left(f.summary,400),f.asset_ids,COALESCE(m.case_id,0) FROM findings f LEFT JOIN finding_case_members m ON m.finding_id=f.id`+where+fmt.Sprintf(` ORDER BY f.id DESC LIMIT $%d OFFSET $%d`, len(args)-1, len(args)), args...)
+	rows, err := d.Query(base+fmt.Sprintf(` SELECT id,name,vulnclass,severity,summary,asset_ids,case_id,score,strong,to_json(hit_reasons)
+        FROM candidates ORDER BY score DESC,id DESC LIMIT $%d OFFSET $%d`, len(args)-1, len(args)), args...)
 	if err != nil {
-		return nil, 0, err
+		return nil, 0, 0, err
 	}
 	defer rows.Close()
 	out := []map[string]any{}
 	for rows.Next() {
-		var id, cid int64
+		var id, cid, score int64
 		var name, class, severity, summary string
 		var assetIDs json.RawMessage
-		if err := rows.Scan(&id, &name, &class, &severity, &summary, &assetIDs, &cid); err != nil {
-			return nil, 0, err
+		var strong bool
+		var hitReasonsRaw json.RawMessage
+		if err := rows.Scan(&id, &name, &class, &severity, &summary, &assetIDs, &cid, &score, &strong, &hitReasonsRaw); err != nil {
+			return nil, 0, 0, err
 		}
-		out = append(out, map[string]any{"finding_id": strconv.FormatInt(id, 10), "name": name, "vulnclass": class, "severity": severity, "summary": summary, "asset_ids": assetIDs, "case_id": strconv.FormatInt(cid, 10)})
+		var hitReasons []string
+		if err := json.Unmarshal(hitReasonsRaw, &hitReasons); err != nil {
+			return nil, 0, 0, err
+		}
+		tier := "weak"
+		if strong {
+			tier = "strong"
+		}
+		out = append(out, map[string]any{"finding_id": strconv.FormatInt(id, 10), "name": name, "vulnclass": class, "severity": severity, "summary": summary, "asset_ids": assetIDs, "case_id": strconv.FormatInt(cid, 10), "score": int(score), "hit_reasons": hitReasons, "tier": tier})
 	}
-	return out, total, rows.Err()
+	return out, total, strongTotal, rows.Err()
 }
 
 // Batch task-list counters avoid an additional database round trip per task.

db/finding_cases_test.go

@@ -5,6 +5,7 @@ import (
 	"encoding/json"
 	"fmt"
 	"reflect"
+	"strings"
 	"sync"
 	"testing"
 )
@@ -118,11 +119,14 @@ func TestFindingCaseScopeSuggestionsAndDeletion(t *testing.T) {
 	if _, err := d.MergeFindingCase(ctx, task.ID, []int64{ids[0], fid}, "x", "x", "reporter"); err == nil {
 		t.Fatal("cross task accepted")
 	}
-	sid, err := d.SuggestFindingCase(ctx, task.ID, ids[0], ids[1], "order", "uncertain")
+	if _, err := d.SuggestFindingCase(ctx, task.ID, []int64{ids[0], fid}, "x", "x"); err == nil {
+		t.Fatal("cross task suggestion accepted")
+	}
+	sid, err := d.SuggestFindingCase(ctx, task.ID, []int64{ids[0], ids[1]}, "order", "uncertain")
 	if err != nil {
 		t.Fatal(err)
 	}
-	if _, err := d.ResolveFindingCaseSuggestion(ctx, sid, false); err != nil {
+	if _, err := d.ResolveFindingCaseSuggestion(ctx, sid, false, nil); err != nil {
 		t.Fatal(err)
 	}
 	if _, err := d.MergeFindingCase(ctx, task.ID, ids[:2], "x", "x", "reporter"); err == nil {
@@ -233,7 +237,7 @@ func TestFindingCaseTaskDeletionAndRepeatedDetach(t *testing.T) {
 func TestFindingCaseSuggestionResolutionIsAtomic(t *testing.T) {
 	d, task, ids := caseTestDB(t)
 	ctx := context.Background()
-	sid, err := d.SuggestFindingCase(ctx, task.ID, ids[0], ids[1], "orders", "uncertain")
+	sid, err := d.SuggestFindingCase(ctx, task.ID, []int64{ids[0], ids[1]}, "orders", "uncertain")
 	if err != nil {
 		t.Fatal(err)
 	}
@@ -241,7 +245,7 @@ func TestFindingCaseSuggestionResolutionIsAtomic(t *testing.T) {
 	var wg sync.WaitGroup
 	for _, accept := range []bool{true, false} {
 		wg.Add(1)
-		go func(a bool) { defer wg.Done(); _, e := d.ResolveFindingCaseSuggestion(ctx, sid, a); results <- e }(accept)
+		go func(a bool) { defer wg.Done(); _, e := d.ResolveFindingCaseSuggestion(ctx, sid, a, nil); results <- e }(accept)
 	}
 	wg.Wait()
 	close(results)
@@ -269,7 +273,7 @@ func TestFindingCaseCandidatesAreCompactAndRelated(t *testing.T) {
 	if _, err := d.Exec(`UPDATE findings SET vulnclass='SQLi',name='search injection' WHERE id=$1`, ids[3]); err != nil {
 		t.Fatal(err)
 	}
-	items, total, err := d.FindingCaseCandidates(ids[0], 1, 1)
+	items, total, _, err := d.FindingCaseCandidates(ids[0], 1, 1)
 	if err != nil || total != 2 || len(items) != 1 {
 		t.Fatalf("candidate pagination %d %+v %v", total, items, err)
 	}
@@ -286,7 +290,7 @@ func TestFindingCaseCandidatesAreCompactAndRelated(t *testing.T) {
 	if err := d.RemoveFindingCaseMember(context.Background(), cid, ids[1], "different cause"); err != nil {
 		t.Fatal(err)
 	}
-	_, total, err = d.FindingCaseCandidates(ids[0], 1, 20)
+	_, total, _, err = d.FindingCaseCandidates(ids[0], 1, 20)
 	if err != nil || total != 1 {
 		t.Fatalf("veto excluded %d %v", total, err)
 	}
@@ -355,3 +359,242 @@ func TestFindingCaseViewContextAndAssetFiltering(t *testing.T) {
 		t.Fatalf("original severity order %+v %v", rows, err)
 	}
 }
+
+func TestFindingCaseSuggestionClusterLifecycle(t *testing.T) {
+	d, task, ids := caseTestDB(t)
+	ctx := context.Background()
+	sid, err := d.SuggestFindingCase(ctx, task.ID, ids[:3], "orders", "same handler evidence")
+	if err != nil {
+		t.Fatal(err)
+	}
+	items, err := d.FindingCaseSuggestions(task.ID)
+	if err != nil || len(items) != 1 || len(items[0].FindingIDs) != 3 {
+		t.Fatalf("cluster listing %+v %v", items, err)
+	}
+	if reused, err := d.SuggestFindingCase(ctx, task.ID, ids[:2], "weaker", "subset"); err != nil || reused != sid {
+		t.Fatalf("pending superset was not reused: %d %v", reused, err)
+	}
+	if _, err := d.ResolveFindingCaseSuggestion(ctx, sid, true, nil); err != nil {
+		t.Fatal(err)
+	}
+	var state string
+	if err := d.QueryRow(`SELECT state FROM finding_case_suggestions WHERE id=$1`, sid).Scan(&state); err != nil || state != "accepted" {
+		t.Fatalf("cluster state %q %v", state, err)
+	}
+	for _, fid := range ids[:3] {
+		if cid, err := d.FindingCaseID(fid); err != nil || cid == 0 {
+			t.Fatalf("member %d not merged: %d %v", fid, cid, err)
+		}
+	}
+}
+
+func TestFindingCaseSuggestionSweepScopesEachCard(t *testing.T) {
+	d, task, ids := caseTestDB(t)
+	ctx := context.Background()
+	first, err := d.SuggestFindingCase(ctx, task.ID, ids[:2], "first", "first group")
+	if err != nil {
+		t.Fatal(err)
+	}
+	second, err := d.SuggestFindingCase(ctx, task.ID, ids[2:], "second", "second group")
+	if err != nil {
+		t.Fatal(err)
+	}
+	if _, err := d.MergeFindingCase(ctx, task.ID, ids[:2], "first", "same root cause", "human"); err != nil {
+		t.Fatal(err)
+	}
+	var firstState, secondState string
+	if err := d.QueryRow(`SELECT state FROM finding_case_suggestions WHERE id=$1`, first).Scan(&firstState); err != nil {
+		t.Fatal(err)
+	}
+	if err := d.QueryRow(`SELECT state FROM finding_case_suggestions WHERE id=$1`, second).Scan(&secondState); err != nil {
+		t.Fatal(err)
+	}
+	if firstState != "accepted" || secondState != "pending" {
+		t.Fatalf("suggestion states first=%q second=%q", firstState, secondState)
+	}
+}
+
+func TestFindingCaseSuggestionSubsetSuperseded(t *testing.T) {
+	d, task, ids := caseTestDB(t)
+	ctx := context.Background()
+	subset, err := d.SuggestFindingCase(ctx, task.ID, ids[:2], "orders", "partial evidence")
+	if err != nil {
+		t.Fatal(err)
+	}
+	superset, err := d.SuggestFindingCase(ctx, task.ID, ids[:3], "orders", "complete group evidence")
+	if err != nil {
+		t.Fatal(err)
+	}
+	if subset == superset {
+		t.Fatal("subset suggestion was reused instead of superseded")
+	}
+	var state string
+	if err := d.QueryRow(`SELECT state FROM finding_case_suggestions WHERE id=$1`, subset).Scan(&state); err != nil || state != "superseded" {
+		t.Fatalf("subset state %q %v", state, err)
+	}
+}
+
+func TestFindingCaseSuggestionAcceptExclusionAndReject(t *testing.T) {
+	d, task, ids := caseTestDB(t)
+	ctx := context.Background()
+	sid, err := d.SuggestFindingCase(ctx, task.ID, ids[:3], "orders", "same component")
+	if err != nil {
+		t.Fatal(err)
+	}
+	if _, err := d.ResolveFindingCaseSuggestion(ctx, sid, true, []int64{ids[2]}); err != nil {
+		t.Fatal(err)
+	}
+	var blocks int
+	if err := d.QueryRow(`SELECT count(*) FROM finding_case_blocks WHERE left_id=$1 AND right_id=$2`, ids[0], ids[2]).Scan(&blocks); err != nil || blocks != 1 {
+		t.Fatalf("excluded member was not vetoed: %d %v", blocks, err)
+	}
+	if cid, _ := d.FindingCaseID(ids[2]); cid != 0 {
+		t.Fatalf("excluded member entered case %d", cid)
+	}
+}
+
+func TestFindingCaseSuggestionExclusionBlocksOnlyKeptSuggestionMembers(t *testing.T) {
+	d, task, ids := caseTestDB(t)
+	ctx := context.Background()
+	if _, err := d.MergeFindingCase(ctx, task.ID, ids[:2], "existing folder", "same root cause", "human"); err != nil {
+		t.Fatal(err)
+	}
+	sid, err := d.SuggestFindingCase(ctx, task.ID, []int64{ids[0], ids[2], ids[3]}, "orders", "same component")
+	if err != nil {
+		t.Fatal(err)
+	}
+	if _, err := d.ResolveFindingCaseSuggestion(ctx, sid, true, []int64{ids[3]}); err != nil {
+		t.Fatal(err)
+	}
+	var folderMateBlock, keptBlock int
+	if err := d.QueryRow(`SELECT count(*) FROM finding_case_blocks WHERE left_id=$1 AND right_id=$2`, ids[1], ids[3]).Scan(&folderMateBlock); err != nil {
+		t.Fatal(err)
+	}
+	if err := d.QueryRow(`SELECT count(*) FROM finding_case_blocks WHERE left_id=$1 AND right_id=$2`, ids[0], ids[3]).Scan(&keptBlock); err != nil {
+		t.Fatal(err)
+	}
+	if folderMateBlock != 0 || keptBlock != 1 {
+		t.Fatalf("exclusion blocks folder mate=%d kept=%d", folderMateBlock, keptBlock)
+	}
+}
+
+func TestFindingCaseSuggestionRejectVetoesAllPairs(t *testing.T) {
+	d, task, ids := caseTestDB(t)
+	ctx := context.Background()
+	sid, err := d.SuggestFindingCase(ctx, task.ID, ids[:3], "orders", "same component")
+	if err != nil {
+		t.Fatal(err)
+	}
+	if _, err := d.ResolveFindingCaseSuggestion(ctx, sid, false, nil); err != nil {
+		t.Fatal(err)
+	}
+	var count int
+	if err := d.QueryRow(`SELECT count(*) FROM finding_case_blocks WHERE left_id=ANY($1::bigint[]) AND right_id=ANY($1::bigint[])`, ids[:3]).Scan(&count); err != nil || count != 3 {
+		t.Fatalf("reject did not veto all pairs: %d %v", count, err)
+	}
+}
+
+func TestFindingCaseSuggestionRejectSkipsExistingFolderAndBlockedSuggest(t *testing.T) {
+	d, task, ids := caseTestDB(t)
+	ctx := context.Background()
+	if _, err := d.MergeFindingCase(ctx, task.ID, ids[:2], "orders", "same handler", "human"); err != nil {
+		t.Fatal(err)
+	}
+	sid, err := d.SuggestFindingCase(ctx, task.ID, ids[:3], "orders", "same handler")
+	if err != nil {
+		t.Fatal(err)
+	}
+	if _, err := d.ResolveFindingCaseSuggestion(ctx, sid, false, nil); err != nil {
+		t.Fatal(err)
+	}
+	var blocked int
+	if err := d.QueryRow(`SELECT count(*) FROM finding_case_blocks WHERE left_id=$1 AND right_id=$2`, ids[0], ids[1]).Scan(&blocked); err != nil || blocked != 0 {
+		t.Fatalf("same-folder pair was vetoed: %d %v", blocked, err)
+	}
+	if _, err := d.SuggestFindingCase(ctx, task.ID, []int64{ids[0], ids[2]}, "orders", "retry"); err == nil || !strings.Contains(err.Error(), fmt.Sprintf("#%d/#%d", ids[0], ids[2])) {
+		t.Fatalf("blocked pair was accepted or unnamed: %v", err)
+	}
+}
+
+func TestFindingCaseSuggestionExclusionCannotPullExistingFolder(t *testing.T) {
+	d, task, ids := caseTestDB(t)
+	ctx := context.Background()
+	if _, err := d.MergeFindingCase(ctx, task.ID, ids[:2], "orders", "same handler", "human"); err != nil {
+		t.Fatal(err)
+	}
+	sid, err := d.SuggestFindingCase(ctx, task.ID, ids[:3], "orders", "same handler")
+	if err != nil {
+		t.Fatal(err)
+	}
+	if _, err := d.ResolveFindingCaseSuggestion(ctx, sid, true, []int64{ids[1]}); err == nil {
+		t.Fatal("excluded member was pulled through its existing folder")
+	}
+	if cid, _ := d.FindingCaseID(ids[2]); cid != 0 {
+		t.Fatalf("failed exclusion merged remaining record into case %d", cid)
+	}
+	var state string
+	if err := d.QueryRow(`SELECT state FROM finding_case_suggestions WHERE id=$1`, sid).Scan(&state); err != nil || state != "pending" {
+		t.Fatalf("failed exclusion changed suggestion state: %q %v", state, err)
+	}
+}
+
+func TestFindingCaseSuggestionAnchorDeletionAndLegacyPair(t *testing.T) {
+	d, task, ids := caseTestDB(t)
+	ctx := context.Background()
+	sid, err := d.SuggestFindingCase(ctx, task.ID, ids[:3], "orders", "same handler")
+	if err != nil {
+		t.Fatal(err)
+	}
+	if _, err := d.DeleteFinding(ids[0]); err != nil {
+		t.Fatal(err)
+	}
+	items, err := d.FindingCaseSuggestions(task.ID)
+	if err != nil || len(items) != 1 || items[0].ID != sid || items[0].LeftID != ids[1] || items[0].RightID != ids[2] {
+		t.Fatalf("anchor deletion lost group: %+v %v", items, err)
+	}
+	if got := items[0].FindingIDs; len(got) != 2 || got[0] != fmt.Sprint(ids[1]) || got[1] != fmt.Sprint(ids[2]) {
+		t.Fatalf("surviving members %+v", got)
+	}
+	if _, err := d.Exec(`DELETE FROM finding_case_suggestion_members WHERE suggestion_id=$1`, sid); err != nil {
+		t.Fatal(err)
+	}
+	if _, err := d.Exec(`UPDATE finding_case_suggestions SET left_id=$2,right_id=$3 WHERE id=$1`, sid, ids[1], ids[3]); err != nil {
+		t.Fatal(err)
+	}
+	legacy, err := d.FindingCaseSuggestions(task.ID)
+	if err != nil || len(legacy) != 1 || len(legacy[0].FindingIDs) != 2 {
+		t.Fatalf("legacy pair was not listed: %+v %v", legacy, err)
+	}
+	if _, err := d.ResolveFindingCaseSuggestion(ctx, sid, true, nil); err != nil {
+		t.Fatal(err)
+	}
+}
+
+func TestFindingCaseCandidatesScoreAndTier(t *testing.T) {
+	d, _, ids := caseTestDB(t)
+	if _, err := d.Exec(`UPDATE findings SET vulnclass='IDOR',name='order access',asset_ids='[101]'::jsonb WHERE id=$1`, ids[0]); err != nil {
+		t.Fatal(err)
+	}
+	if _, err := d.Exec(`UPDATE findings SET vulnclass='idor',name='different',asset_ids='[101]'::jsonb WHERE id=$1`, ids[1]); err != nil {
+		t.Fatal(err)
+	}
+	if _, err := d.Exec(`UPDATE findings SET vulnclass='SQLi',name='different',asset_ids='[101]'::jsonb WHERE id=$1`, ids[2]); err != nil {
+		t.Fatal(err)
+	}
+	if _, err := d.Exec(`UPDATE findings SET vulnclass='SQLi',name='order access',asset_ids='[]'::jsonb WHERE id=$1`, ids[3]); err != nil {
+		t.Fatal(err)
+	}
+	items, total, strong, err := d.FindingCaseCandidates(ids[0], 1, 20)
+	if err != nil || total != 3 || strong != 1 || len(items) != 3 {
+		t.Fatalf("candidate totals %d %d %+v %v", total, strong, items, err)
+	}
+	if items[0]["finding_id"] != fmt.Sprint(ids[1]) || items[0]["score"] != 7 || items[0]["tier"] != "strong" || !reflect.DeepEqual(items[0]["hit_reasons"], []string{"同漏洞类型", "资产重叠"}) {
+		t.Fatalf("strong candidate %+v", items[0])
+	}
+	if items[1]["finding_id"] != fmt.Sprint(ids[2]) || items[1]["score"] != 2 || items[1]["tier"] != "weak" {
+		t.Fatalf("asset candidate %+v", items[1])
+	}
+	if items[2]["finding_id"] != fmt.Sprint(ids[3]) || items[2]["score"] != 1 {
+		t.Fatalf("name candidate %+v", items[2])
+	}
+}

db/intelligence.go

@@ -260,8 +260,8 @@ func (d *DB) ListIntelligence(ctx context.Context, f IntelligenceFilter) (*Intel
 	if err != nil {
 		return nil, err
 	}
-	// Search titles and reporters; return the recorded content directly in each row.
-	where := scope + ` AND ($3='' OR kind=$3) AND ($4='' OR i.status=$4) AND ($5='' OR i.title ILIKE '%'||$5||'%' OR i.worker ILIKE '%'||$5||'%')`
+	// Search titles, content, evidence, and reporters; return the recorded content directly in each row.
+	where := scope + ` AND ($3='' OR kind=$3) AND ($4='' OR i.status=$4) AND ($5='' OR i.title ILIKE '%'||$5||'%' OR i.content ILIKE '%'||$5||'%' OR i.evidence ILIKE '%'||$5||'%' OR i.worker ILIKE '%'||$5||'%')`
 	args = append(args, f.Kind, f.Status, f.Query)
 	if err = d.QueryRowContext(ctx, `SELECT count(*)`+where, args...).Scan(&page.Total); err != nil {
 		return nil, err

db/schema.sql

@@ -524,6 +524,7 @@ CREATE TABLE IF NOT EXISTS tasks (
     timeout_seconds INTEGER NOT NULL DEFAULT 0,
     plan_heartbeat_seconds INTEGER NOT NULL DEFAULT 300,
     coverage_enabled BOOLEAN NOT NULL DEFAULT true,
+    worker_count   INTEGER NOT NULL DEFAULT 0, -- 0=用全局默认 workers
     pinned_at      TIMESTAMPTZ,
     first_run_at   TIMESTAMPTZ,
     deadline_at    TIMESTAMPTZ,
@@ -546,6 +547,8 @@ ALTER TABLE tasks ADD COLUMN IF NOT EXISTS queued BOOLEAN NOT NULL DEFAULT false
 -- 给 agent 开放 add_task_scope/list_untested_assets;false=全部关闭(见 task_scope.go)。
 -- 存量任务默认 true 保持原行为;company 关联(task_scope kind=company)不受此开关影响。
 ALTER TABLE tasks ADD COLUMN IF NOT EXISTS coverage_enabled BOOLEAN NOT NULL DEFAULT true;
+-- 任务级 Worker 运行数量;0=用全局默认(settings.workers)。补旧库。
+ALTER TABLE tasks ADD COLUMN IF NOT EXISTS worker_count INTEGER NOT NULL DEFAULT 0;
 -- queued_at makes admission FIFO reflect the actual enqueue order rather than the
 -- task creation order. queue_mode distinguishes first bootstrap from resuming an
 -- exploration that already owns goals/history.
@@ -617,16 +620,21 @@ CREATE TABLE IF NOT EXISTS task_templates (
     nkey        TEXT NOT NULL UNIQUE,
     description TEXT NOT NULL,
     goal        TEXT NOT NULL,
+    remark      TEXT NOT NULL DEFAULT '',
     -- 预设的任务分类；分类删除时置空（与 tasks.category_id 一致，不阻断）。
     category_id     BIGINT REFERENCES task_categories(id) ON DELETE SET NULL,
     -- 预设的任务级拦截/允许规则快照(AssetInterceptRuleInput 数组)；应用模板时灌进新任务。
     intercept_rules JSONB NOT NULL DEFAULT '[]',
+    -- 预设的任务级 Worker 运行数量;0=用全局默认。
+    worker_count INTEGER NOT NULL DEFAULT 0,
     created_at  TIMESTAMPTZ NOT NULL DEFAULT now(),
     updated_at  TIMESTAMPTZ NOT NULL DEFAULT now()
 );
 -- 补旧库(已发版,加列带 IF NOT EXISTS)。
 ALTER TABLE task_templates ADD COLUMN IF NOT EXISTS category_id BIGINT REFERENCES task_categories(id) ON DELETE SET NULL;
 ALTER TABLE task_templates ADD COLUMN IF NOT EXISTS intercept_rules JSONB NOT NULL DEFAULT '[]';
+ALTER TABLE task_templates ADD COLUMN IF NOT EXISTS remark TEXT NOT NULL DEFAULT '';
+ALTER TABLE task_templates ADD COLUMN IF NOT EXISTS worker_count INTEGER NOT NULL DEFAULT 0;
 CREATE INDEX IF NOT EXISTS idx_task_templates_updated ON task_templates(updated_at DESC, id DESC);
 DROP TRIGGER IF EXISTS trg_task_templates_upd ON task_templates;
 CREATE TRIGGER trg_task_templates_upd BEFORE UPDATE ON task_templates
@@ -1488,6 +1496,54 @@ CREATE TABLE IF NOT EXISTS finding_case_suggestions (
     UNIQUE(left_id,right_id),
     CHECK(left_id < right_id)
 );
+-- A suggestion is a group.  left_id/right_id remain compatibility anchors for
+-- old clients and are kept at the two smallest surviving member ids.
+ALTER TABLE finding_case_suggestions DROP CONSTRAINT IF EXISTS finding_case_suggestions_left_id_right_id_key;
+CREATE TABLE IF NOT EXISTS finding_case_suggestion_members (
+    suggestion_id BIGINT NOT NULL REFERENCES finding_case_suggestions(id) ON DELETE CASCADE,
+    finding_id BIGINT NOT NULL REFERENCES findings(id) ON DELETE CASCADE,
+    PRIMARY KEY (suggestion_id, finding_id)
+);
+CREATE INDEX IF NOT EXISTS idx_finding_case_suggestion_members_finding
+    ON finding_case_suggestion_members(finding_id);
+INSERT INTO finding_case_suggestion_members(suggestion_id, finding_id)
+    SELECT id, left_id FROM finding_case_suggestions ON CONFLICT DO NOTHING;
+INSERT INTO finding_case_suggestion_members(suggestion_id, finding_id)
+    SELECT id, right_id FROM finding_case_suggestions ON CONFLICT DO NOTHING;
+-- Anchors have cascading foreign keys.  Re-anchor a group before an anchor
+-- finding is deleted so a 3+ member suggestion survives the cascade.
+CREATE OR REPLACE FUNCTION finding_case_suggestion_reanchor() RETURNS trigger
+LANGUAGE plpgsql AS $$
+DECLARE
+    sid BIGINT;
+    new_left BIGINT;
+    new_right BIGINT;
+BEGIN
+    FOR sid IN
+        SELECT id FROM finding_case_suggestions
+        WHERE left_id = OLD.id OR right_id = OLD.id
+    LOOP
+        SELECT min(m.finding_id) INTO new_left
+        FROM finding_case_suggestion_members m
+        JOIN findings f ON f.id = m.finding_id
+        WHERE m.suggestion_id = sid AND m.finding_id <> OLD.id;
+        SELECT min(m.finding_id) INTO new_right
+        FROM finding_case_suggestion_members m
+        JOIN findings f ON f.id = m.finding_id
+        WHERE m.suggestion_id = sid AND m.finding_id <> OLD.id
+          AND m.finding_id > new_left;
+        IF new_left IS NOT NULL AND new_right IS NOT NULL THEN
+            UPDATE finding_case_suggestions
+            SET left_id = new_left, right_id = new_right
+            WHERE id = sid;
+        END IF;
+    END LOOP;
+    RETURN OLD;
+END $$;
+DROP TRIGGER IF EXISTS finding_case_suggestion_reanchor ON findings;
+CREATE TRIGGER finding_case_suggestion_reanchor
+BEFORE DELETE ON findings FOR EACH ROW
+EXECUTE FUNCTION finding_case_suggestion_reanchor();
 -- Rejected/detached pairs are a durable human veto, also checked on group unions.
 CREATE TABLE IF NOT EXISTS finding_case_blocks (
     left_id BIGINT NOT NULL REFERENCES findings(id) ON DELETE CASCADE,

db/settings.go

@@ -3,6 +3,7 @@ package db
 import (
 	"context"
 	"database/sql"
+	"sort"
 )
 
 // Settings is a tiny key-value store for global app config the UI toggles at
@@ -33,6 +34,32 @@ ON CONFLICT (key) DO UPDATE SET value = EXCLUDED.value, updated_at = now()`, key
 	return err
 }
 
+// SetSettingsAtomic upserts several settings in one transaction. Keys are
+// sorted so concurrent callers acquire row locks in a deterministic order.
+func (d *DB) SetSettingsAtomic(ctx context.Context, values map[string]string) error {
+	if len(values) == 0 {
+		return nil
+	}
+	tx, err := d.BeginTx(ctx, nil)
+	if err != nil {
+		return err
+	}
+	defer tx.Rollback() //nolint:errcheck // commit makes this a no-op
+	keys := make([]string, 0, len(values))
+	for key := range values {
+		keys = append(keys, key)
+	}
+	sort.Strings(keys)
+	for _, key := range keys {
+		if _, err := tx.ExecContext(ctx, `
+INSERT INTO settings(key, value) VALUES ($1, $2)
+ON CONFLICT (key) DO UPDATE SET value = EXCLUDED.value, updated_at = now()`, key, values[key]); err != nil {
+			return err
+		}
+	}
+	return tx.Commit()
+}
+
 // InsertSettingIfAbsent 只在 key 尚未存在时写入，已存在则原样保留并返回 inserted=false。
 // 给 auth.password_hash 这类"只允许首次设置"的键用：判定落在数据库的主键约束上，
 // 调用方先 GetSetting 再写的那种检查就只是快速失败路径——读出错或并发撞车时，

db/settings_test.go

@@ -0,0 +1,28 @@
+package db
+
+import (
+	"context"
+	"fmt"
+	"testing"
+	"time"
+)
+
+func TestSetSettingsAtomic(t *testing.T) {
+	d, err := Open(testDSN(t))
+	if err != nil {
+		t.Skipf("postgres unavailable (%v)", err)
+	}
+	defer d.Close()
+	prefix := fmt.Sprintf("test.atomic_settings.%d", time.Now().UnixNano())
+	defer func() { _, _ = d.Exec(`DELETE FROM settings WHERE key LIKE $1`, prefix+".%") }()
+	values := map[string]string{prefix + ".worker_pools": "[]", prefix + ".workers": "3"}
+	if err := d.SetSettingsAtomic(context.Background(), values); err != nil {
+		t.Fatal(err)
+	}
+	for key, want := range values {
+		got, ok, err := d.GetSetting(key)
+		if err != nil || !ok || got != want {
+			t.Fatalf("key %q: value=%q ok=%v err=%v, want %q", key, got, ok, err, want)
+		}
+	}
+}

db/task_archives_restore.go

@@ -122,7 +122,7 @@ AND NOT EXISTS (
 	if _, err := tx.Exec(`UPDATE tasks SET
  name='',remark='',category_id=NULL,description='',goal='',paused=true,queued=false,queued_at=NULL,queue_mode='',
  llm_profile_id=NULL,active_llm_profile_id=NULL,llm_chain_revision=llm_chain_revision+1,
- company_id=NULL,parent_ref=NULL,timeout_seconds=0,coverage_enabled=true,pinned_at=NULL,
+ company_id=NULL,parent_ref=NULL,timeout_seconds=0,coverage_enabled=true,worker_count=0,pinned_at=NULL,
  first_run_at=NULL,deadline_at=NULL,archived_at=now(),deleted_at=now()
 WHERE id=$1`, taskID); err != nil {
 		return err
@@ -322,7 +322,7 @@ func restoreTaskStub(tx *sql.Tx, row map[string]any, taskID, remaining int64) er
  llm_profile_id=archived.llm_profile_id,active_llm_profile_id=archived.active_llm_profile_id,
  llm_chain_revision=archived.llm_chain_revision,company_id=archived.company_id,parent_ref=archived.parent_ref,
  timeout_seconds=archived.timeout_seconds,plan_heartbeat_seconds=archived.plan_heartbeat_seconds,
- coverage_enabled=archived.coverage_enabled,pinned_at=archived.pinned_at,first_run_at=archived.first_run_at,
+ coverage_enabled=archived.coverage_enabled,worker_count=COALESCE(archived.worker_count,0),pinned_at=archived.pinned_at,first_run_at=archived.first_run_at,
  deadline_at=CASE WHEN archived.paused AND $3>0 THEN now()+make_interval(secs=>$3::double precision)
                   ELSE archived.deadline_at END,
  deleted_at=NULL,archived_at=NULL,completed_at=archived.completed_at,

db/task_archives_test.go

@@ -132,7 +132,9 @@ func TestTaskArchiveDatabaseRoundTrip(t *testing.T) {
 	if err := d.EnsureLLMUsageTable(); err != nil {
 		t.Fatal(err)
 	}
-	task, err := d.CreateTaskWithOptions("archive database roundtrip", "restore exact graph", TaskCreateOptions{Name: "cold task"})
+	task, err := d.CreateTaskWithOptions("archive database roundtrip", "restore exact graph", TaskCreateOptions{
+		Name: "cold task", WorkerCount: 7,
+	})
 	if err != nil {
 		t.Fatal(err)
 	}
@@ -297,7 +299,7 @@ VALUES($1,$2,0,'quota_exhausted','balance exhausted',$3,$4,$3)`, task.ID, llmPro
 	if err != nil || live == nil {
 		t.Fatalf("restored task = %+v, %v", live, err)
 	}
-	if live.Name != "cold task" || live.Remark != remark || !live.Paused {
+	if live.Name != "cold task" || live.Remark != remark || live.WorkerCount != 7 || !live.Paused {
 		t.Fatalf("restored task metadata mismatch: %+v", live)
 	}
 	if err := d.QueryRow(`SELECT count(*) FROM exploration_nodes WHERE exploration_id=$1`, task.ExplorationID).Scan(&nodes); err != nil {

db/task_options_test.go

@@ -0,0 +1,102 @@
+package db
+
+import (
+	"fmt"
+	"strings"
+	"testing"
+	"time"
+)
+
+func TestTaskAndTemplateRemarkLimitIsShared(t *testing.T) {
+	boundary := strings.Repeat("界", MaxTaskRemarkRunes)
+	if _, _, err := normalizeTaskTemplateInput(TaskTemplateInput{Name: "n", Description: "d", Goal: "g", Remark: boundary}); err != nil {
+		t.Fatalf("boundary remark rejected: %v", err)
+	}
+	if _, _, err := normalizeTaskTemplateInput(TaskTemplateInput{Name: "n", Description: "d", Goal: "g", Remark: boundary + "界"}); err == nil {
+		t.Fatal("oversized template remark accepted")
+	}
+}
+
+func TestWorkerCountNormalizationAndLimit(t *testing.T) {
+	if got := normalizeWorkerCount(-1); got != 0 {
+		t.Fatalf("negative worker count = %d, want 0", got)
+	}
+	if _, _, err := normalizeTaskTemplateInput(TaskTemplateInput{Name: "n", Description: "d", Goal: "g", WorkerCount: MaxTaskWorkerCount}); err != nil {
+		t.Fatalf("maximum worker count rejected: %v", err)
+	}
+	if _, _, err := normalizeTaskTemplateInput(TaskTemplateInput{Name: "n", Description: "d", Goal: "g", WorkerCount: MaxTaskWorkerCount + 1}); err == nil {
+		t.Fatal("oversized worker count accepted")
+	}
+}
+
+func TestTaskOptionsPersistAndValidateLimits(t *testing.T) {
+	d, err := Open(testDSN(t))
+	if err != nil {
+		t.Skipf("postgres unavailable (%v) — skipping", err)
+	}
+	defer d.Close()
+
+	if _, err := d.CreateTaskWithOptions("oversized remark", "goal", TaskCreateOptions{
+		Remark: strings.Repeat("界", MaxTaskRemarkRunes+1),
+	}); err == nil {
+		t.Fatal("2001-rune task remark accepted")
+	}
+	if _, err := d.CreateTaskWithOptions("oversized workers", "goal", TaskCreateOptions{
+		WorkerCount: MaxTaskWorkerCount + 1,
+	}); err == nil {
+		t.Fatal("51-worker task accepted")
+	}
+
+	remark := strings.Repeat("界", MaxTaskRemarkRunes)
+	task, err := d.CreateTaskWithOptions("task options boundary", "goal", TaskCreateOptions{
+		Remark: remark, WorkerCount: MaxTaskWorkerCount,
+	})
+	if err != nil {
+		t.Fatal(err)
+	}
+	t.Cleanup(func() { _ = d.DeleteTask(task.ID) })
+	gotTask, err := d.GetTask(task.ID)
+	if err != nil {
+		t.Fatal(err)
+	}
+	if gotTask == nil || gotTask.Remark != remark || gotTask.WorkerCount != MaxTaskWorkerCount {
+		t.Fatalf("task options round trip = %+v", gotTask)
+	}
+
+	name := fmt.Sprintf("options template %d", time.Now().UnixNano())
+	templateRemark := "模板备注 📝"
+	template, err := d.CreateTaskTemplate(TaskTemplateInput{
+		Name: name, Description: "template description", Goal: "template goal",
+		Remark: templateRemark, WorkerCount: MaxTaskWorkerCount,
+	})
+	if err != nil {
+		t.Fatal(err)
+	}
+	t.Cleanup(func() { _, _ = d.DeleteTaskTemplate(template.ID) })
+	gotTemplate, err := d.GetTaskTemplate(template.ID)
+	if err != nil {
+		t.Fatal(err)
+	}
+	if gotTemplate == nil || gotTemplate.Remark != templateRemark || gotTemplate.WorkerCount != MaxTaskWorkerCount {
+		t.Fatalf("created template options round trip = %+v", gotTemplate)
+	}
+
+	updatedRemark := strings.Repeat("改", MaxTaskRemarkRunes)
+	updatedWorkers := MaxTaskWorkerCount - 1
+	updated, err := d.PatchTaskTemplate(template.ID, TaskTemplatePatch{
+		Remark: &updatedRemark, WorkerCount: &updatedWorkers,
+	})
+	if err != nil {
+		t.Fatal(err)
+	}
+	if updated.Remark != updatedRemark || updated.WorkerCount != updatedWorkers {
+		t.Fatalf("updated template options = %+v", updated)
+	}
+	gotTemplate, err = d.GetTaskTemplate(template.ID)
+	if err != nil {
+		t.Fatal(err)
+	}
+	if gotTemplate == nil || gotTemplate.Remark != updatedRemark || gotTemplate.WorkerCount != updatedWorkers {
+		t.Fatalf("updated template options round trip = %+v", gotTemplate)
+	}
+}

db/task_templates.go

@@ -23,16 +23,18 @@ var (
 	ErrTaskTemplateNotFound     = errors.New("task template not found")
 )
 
-// TaskTemplate is a reusable task preset (description/goal + optional category
-// and task-level intercept/allow rules).
+// TaskTemplate is a reusable task preset (description/goal + optional remark,
+// worker count, category, and task-level intercept/allow rules).
 type TaskTemplate struct {
 	ID             int64                    `json:"id"`
 	Name           string                   `json:"name"`
 	NKey           string                   `json:"-"`
 	Description    string                   `json:"description"`
 	Goal           string                   `json:"goal"`
+	Remark         string                   `json:"remark"`
 	CategoryID     *int64                   `json:"category_id"`
 	InterceptRules []TaskInterceptRuleInput `json:"intercept_rules"`
+	WorkerCount    int                      `json:"worker_count"`
 	CreatedAt      time.Time                `json:"created_at"`
 	UpdatedAt      time.Time                `json:"updated_at"`
 }
@@ -42,8 +44,10 @@ type TaskTemplateInput struct {
 	Name           string
 	Description    string
 	Goal           string
+	Remark         string
 	CategoryID     *int64
 	InterceptRules []TaskInterceptRuleInput
+	WorkerCount    int
 }
 
 // TaskTemplatePatch changes only fields flagged as set. Name/Description/Goal use
@@ -53,18 +57,20 @@ type TaskTemplatePatch struct {
 	Name              *string
 	Description       *string
 	Goal              *string
+	Remark            *string
 	CategoryID        *int64
 	SetCategoryID     bool
 	InterceptRules    []TaskInterceptRuleInput
 	SetInterceptRules bool
+	WorkerCount       *int
 }
 
-const taskTemplateCols = `id, name, nkey, description, goal, category_id, intercept_rules, created_at, updated_at`
+const taskTemplateCols = `id, name, nkey, description, goal, COALESCE(remark,''), category_id, intercept_rules, COALESCE(worker_count,0), created_at, updated_at`
 
 func scanTaskTemplate(row interface{ Scan(...any) error }) (TaskTemplate, error) {
 	var t TaskTemplate
 	var rulesRaw []byte
-	if err := row.Scan(&t.ID, &t.Name, &t.NKey, &t.Description, &t.Goal, &t.CategoryID, &rulesRaw, &t.CreatedAt, &t.UpdatedAt); err != nil {
+	if err := row.Scan(&t.ID, &t.Name, &t.NKey, &t.Description, &t.Goal, &t.Remark, &t.CategoryID, &rulesRaw, &t.WorkerCount, &t.CreatedAt, &t.UpdatedAt); err != nil {
 		return t, err
 	}
 	t.InterceptRules = []TaskInterceptRuleInput{}
@@ -112,11 +118,21 @@ func normalizeTaskTemplateInput(in TaskTemplateInput) (TaskTemplateInput, string
 	case utf8.RuneCountInString(in.Goal) > MaxTaskTemplateTextRunes:
 		return in, "", fmt.Errorf("%w: goal exceeds %d characters", ErrTaskTemplateInvalid, MaxTaskTemplateTextRunes)
 	}
+	in.Remark = strings.TrimSpace(in.Remark)
+	// Keep template remarks within the same shared limit as task remarks so
+	// applying a template cannot produce a rejected create-task payload.
+	if utf8.RuneCountInString(in.Remark) > MaxTaskRemarkRunes {
+		return in, "", fmt.Errorf("%w: remark exceeds %d characters", ErrTaskTemplateInvalid, MaxTaskRemarkRunes)
+	}
+	in.WorkerCount = normalizeWorkerCount(in.WorkerCount)
+	if err := validateWorkerCount(in.WorkerCount); err != nil {
+		return in, "", fmt.Errorf("%w: %v", ErrTaskTemplateInvalid, err)
+	}
 	return in, taskTemplateNKey(in.Name), nil
 }
 
 func normalizeTaskTemplatePatch(patch TaskTemplatePatch) (TaskTemplatePatch, *string, error) {
-	if patch.Name == nil && patch.Description == nil && patch.Goal == nil && !patch.SetCategoryID && !patch.SetInterceptRules {
+	if patch.Name == nil && patch.Description == nil && patch.Goal == nil && patch.Remark == nil && !patch.SetCategoryID && !patch.SetInterceptRules && patch.WorkerCount == nil {
 		return patch, nil, fmt.Errorf("%w: no fields supplied", ErrTaskTemplateInvalid)
 	}
 	var nkey *string
@@ -152,6 +168,20 @@ func normalizeTaskTemplatePatch(patch TaskTemplatePatch) (TaskTemplatePatch, *st
 		}
 		patch.Goal = &goal
 	}
+	if patch.Remark != nil {
+		remark := strings.TrimSpace(*patch.Remark)
+		if utf8.RuneCountInString(remark) > MaxTaskRemarkRunes {
+			return patch, nil, fmt.Errorf("%w: remark exceeds %d characters", ErrTaskTemplateInvalid, MaxTaskRemarkRunes)
+		}
+		patch.Remark = &remark
+	}
+	if patch.WorkerCount != nil {
+		n := normalizeWorkerCount(*patch.WorkerCount)
+		if err := validateWorkerCount(n); err != nil {
+			return patch, nil, fmt.Errorf("%w: %v", ErrTaskTemplateInvalid, err)
+		}
+		patch.WorkerCount = &n
+	}
 	return patch, nkey, nil
 }
 
@@ -171,10 +201,10 @@ func (d *DB) CreateTaskTemplate(in TaskTemplateInput) (*TaskTemplate, error) {
 		return nil, err
 	}
 	t, err := scanTaskTemplate(d.QueryRow(`
-INSERT INTO task_templates(name, nkey, description, goal, category_id, intercept_rules)
-VALUES ($1,$2,$3,$4,$5,$6)
+INSERT INTO task_templates(name, nkey, description, goal, remark, category_id, intercept_rules, worker_count)
+VALUES ($1,$2,$3,$4,$5,$6,$7,$8)
 ON CONFLICT (nkey) DO NOTHING
-RETURNING `+taskTemplateCols, in.Name, nkey, in.Description, in.Goal, in.CategoryID, rulesJSON))
+RETURNING `+taskTemplateCols, in.Name, nkey, in.Description, in.Goal, in.Remark, in.CategoryID, rulesJSON, in.WorkerCount))
 	if err == sql.ErrNoRows {
 		return nil, ErrTaskTemplateNameConflict
 	}
@@ -221,7 +251,8 @@ func (d *DB) UpdateTaskTemplate(id int64, in TaskTemplateInput) (*TaskTemplate,
 		return nil, err
 	}
 	return d.PatchTaskTemplate(id, TaskTemplatePatch{
-		Name: &in.Name, Description: &in.Description, Goal: &in.Goal,
+		Name: &in.Name, Description: &in.Description, Goal: &in.Goal, Remark: &in.Remark,
+		WorkerCount: &in.WorkerCount,
 	})
 }
 
@@ -242,16 +273,20 @@ SET name=CASE WHEN $2 THEN $3::text ELSE name END,
     nkey=CASE WHEN $2 THEN $4::text ELSE nkey END,
     description=CASE WHEN $5 THEN $6::text ELSE description END,
     goal=CASE WHEN $7 THEN $8::text ELSE goal END,
-    category_id=CASE WHEN $9 THEN $10::bigint ELSE category_id END,
-    intercept_rules=CASE WHEN $11 THEN $12::jsonb ELSE intercept_rules END
+    remark=CASE WHEN $9 THEN $10::text ELSE remark END,
+    category_id=CASE WHEN $11 THEN $12::bigint ELSE category_id END,
+    intercept_rules=CASE WHEN $13 THEN $14::jsonb ELSE intercept_rules END,
+    worker_count=CASE WHEN $15 THEN $16::integer ELSE worker_count END
 WHERE id=$1
 RETURNING `+taskTemplateCols,
 		id,
 		patch.Name != nil, patch.Name, nkey,
 		patch.Description != nil, patch.Description,
 		patch.Goal != nil, patch.Goal,
+		patch.Remark != nil, patch.Remark,
 		patch.SetCategoryID, patch.CategoryID,
 		patch.SetInterceptRules, rulesJSON,
+		patch.WorkerCount != nil, patch.WorkerCount,
 	))
 	if err == sql.ErrNoRows {
 		return nil, ErrTaskTemplateNotFound

db/tasks.go

@@ -6,7 +6,9 @@ import (
 	"errors"
 	"fmt"
 	"strconv"
+	"strings"
 	"time"
+	"unicode/utf8"
 )
 
 // Task is a row in the task registry (1:1 with an exploration).
@@ -51,6 +53,9 @@ type Task struct {
 	// list_untested_assets、态势里不注入 coverage 块(scope 字段仍保留)。company 关联
 	// (task_scope kind=company)与此开关无关，永不受影响。见 db/task_scope.go。
 	CoverageEnabled bool `json:"coverage_enabled"`
+	// WorkerCount is the per-task concurrent work-agent count. 0 means use the
+	// global settings.workers default when the engine starts this task.
+	WorkerCount int `json:"worker_count"`
 }
 
 // TaskDeleteResult reports optional related-data cleanup performed in the same
@@ -132,6 +137,29 @@ func normalizeHeartbeat(sec int) int {
 	return sec
 }
 
+// MaxTaskWorkerCount is the hard upper bound for a per-task worker override.
+const MaxTaskWorkerCount = 50
+
+// MaxTaskRemarkRunes is the shared limit for task and template remarks.
+const MaxTaskRemarkRunes = 2000
+
+// normalizeWorkerCount converts negative per-task worker overrides to 0.
+// The upper bound is enforced separately by validateWorkerCount; 0 means
+// "use the global settings.workers default".
+func normalizeWorkerCount(n int) int {
+	if n < 0 {
+		return 0
+	}
+	return n
+}
+
+func validateWorkerCount(n int) error {
+	if n > MaxTaskWorkerCount {
+		return fmt.Errorf("Worker 运行数量最多 %d", MaxTaskWorkerCount)
+	}
+	return nil
+}
+
 func (d *DB) CreateTask(description, goal string, llmProfileID *int64, timeoutSeconds, planHeartbeatSeconds int) (*Task, error) {
 	var ids []int64
 	if llmProfileID != nil {
@@ -156,6 +184,10 @@ type TaskCreateOptions struct {
 	// CoverageEnabled 是「资产覆盖度功能」开关;nil=默认开(true)，让不关心该开关的创建
 	// 路径(编排 spawn、老 API)沿用原行为。仅 web 创建任务时可显式传 false 关闭。
 	CoverageEnabled *bool
+	// WorkerCount 是本任务并发运行的工作 Agent 数;0/省略=用全局默认。上限 50。
+	WorkerCount int
+	// Remark 是创建时可带的任务备注(最多 2000 字);空=无备注。
+	Remark string
 	// InterceptRules 是任务级资产拦截规则,创建时随任务在同一事务内写入 task_intercept_rules。
 	InterceptRules []TaskInterceptRuleInput
 }
@@ -200,6 +232,14 @@ VALUES ($1, 'fact', $2, 0, 'origin', 'system')`, expID, string(originPayload));
 	}
 	opts.PlanHeartbeatSeconds = normalizeHeartbeat(opts.PlanHeartbeatSeconds)
 	coverageEnabled := opts.CoverageEnabled == nil || *opts.CoverageEnabled
+	opts.WorkerCount = normalizeWorkerCount(opts.WorkerCount)
+	if err := validateWorkerCount(opts.WorkerCount); err != nil {
+		return nil, err
+	}
+	opts.Remark = strings.TrimSpace(opts.Remark)
+	if utf8.RuneCountInString(opts.Remark) > MaxTaskRemarkRunes {
+		return nil, fmt.Errorf("备注最多 %d 个字符", MaxTaskRemarkRunes)
+	}
 	var categoryName string
 	if opts.CategoryID != nil {
 		if *opts.CategoryID <= 0 {
@@ -225,12 +265,12 @@ VALUES ($1, 'fact', $2, 0, 'origin', 'system')`, expID, string(originPayload));
 		SourceTaskIDs:  append([]int64(nil), opts.SourceTaskIDs...),
 		CompanyIDs:     append([]int64(nil), opts.CompanyIDs...),
 		TimeoutSeconds: opts.TimeoutSeconds, PlanHeartbeatSeconds: opts.PlanHeartbeatSeconds,
-		CoverageEnabled: coverageEnabled,
+		CoverageEnabled: coverageEnabled, WorkerCount: opts.WorkerCount, Remark: opts.Remark,
 	}
 	if err := tx.QueryRow(`
-INSERT INTO tasks(name, category_id, description, goal, exploration_id, llm_profile_id, active_llm_profile_id, timeout_seconds, plan_heartbeat_seconds, coverage_enabled)
-VALUES ($1,$2,$3,$4,$5,$6,$6,$7,$8,$9)
-RETURNING id, status, paused, created_at`, opts.Name, opts.CategoryID, description, goal, expID, active, opts.TimeoutSeconds, opts.PlanHeartbeatSeconds, coverageEnabled).Scan(&t.ID, &t.Status, &t.Paused, &t.CreatedAt); err != nil {
+INSERT INTO tasks(name, remark, category_id, description, goal, exploration_id, llm_profile_id, active_llm_profile_id, timeout_seconds, plan_heartbeat_seconds, coverage_enabled, worker_count)
+VALUES ($1,$2,$3,$4,$5,$6,$7,$7,$8,$9,$10,$11)
+RETURNING id, status, paused, created_at`, opts.Name, opts.Remark, opts.CategoryID, description, goal, expID, active, opts.TimeoutSeconds, opts.PlanHeartbeatSeconds, coverageEnabled, opts.WorkerCount).Scan(&t.ID, &t.Status, &t.Paused, &t.CreatedAt); err != nil {
 		return nil, err
 	}
 	if err := insertTaskRelations(tx, t.ID, opts.SourceTaskIDs); err != nil {
@@ -363,11 +403,11 @@ func insertTaskLLMProfiles(tx *sql.Tx, taskID int64, profileIDs []int64) error {
 
 const taskCols = `id, COALESCE(name,''), remark, category_id,
 COALESCE((SELECT category.name FROM task_categories category WHERE category.id=tasks.category_id),''),
-description, goal, exploration_id, status, paused, queued, queued_at, COALESCE(queue_mode,''), llm_profile_id, active_llm_profile_id, COALESCE(parent_ref,''), pinned_at, created_at, completed_at, COALESCE(timeout_seconds,0), COALESCE(plan_heartbeat_seconds,300), COALESCE(coverage_enabled,true), first_run_at, deadline_at`
+description, goal, exploration_id, status, paused, queued, queued_at, COALESCE(queue_mode,''), llm_profile_id, active_llm_profile_id, COALESCE(parent_ref,''), pinned_at, created_at, completed_at, COALESCE(timeout_seconds,0), COALESCE(plan_heartbeat_seconds,300), COALESCE(coverage_enabled,true), COALESCE(worker_count,0), first_run_at, deadline_at`
 
 func scanTask(sc interface{ Scan(...any) error }) (*Task, error) {
 	var t Task
-	if err := sc.Scan(&t.ID, &t.Name, &t.Remark, &t.CategoryID, &t.CategoryName, &t.Description, &t.Goal, &t.ExplorationID, &t.Status, &t.Paused, &t.Queued, &t.QueuedAt, &t.QueueMode, &t.LLMProfileID, &t.ActiveLLMProfileID, &t.ParentRef, &t.PinnedAt, &t.CreatedAt, &t.CompletedAt, &t.TimeoutSeconds, &t.PlanHeartbeatSeconds, &t.CoverageEnabled, &t.FirstRunAt, &t.DeadlineAt); err != nil {
+	if err := sc.Scan(&t.ID, &t.Name, &t.Remark, &t.CategoryID, &t.CategoryName, &t.Description, &t.Goal, &t.ExplorationID, &t.Status, &t.Paused, &t.Queued, &t.QueuedAt, &t.QueueMode, &t.LLMProfileID, &t.ActiveLLMProfileID, &t.ParentRef, &t.PinnedAt, &t.CreatedAt, &t.CompletedAt, &t.TimeoutSeconds, &t.PlanHeartbeatSeconds, &t.CoverageEnabled, &t.WorkerCount, &t.FirstRunAt, &t.DeadlineAt); err != nil {
 		return nil, err
 	}
 	t.Pinned = t.PinnedAt != nil

go.mod

@@ -1,6 +1,6 @@
 module github.com/Autumn-27/artex
 
-go 1.26.3
+go 1.26.10
 
 require (
 	mvdan.cc/sh/v3 v3.10.0

server/assembly_test.go

@@ -6,9 +6,9 @@ import (
 	"path/filepath"
 	"testing"
 
-	actool "github.com/Autumn-27/norma/tool"
 	"github.com/Autumn-27/artex/agent"
 	"github.com/Autumn-27/artex/db"
+	actool "github.com/Autumn-27/norma/tool"
 )
 
 // hasSkillTool reports whether the packed tool set contains the Skill meta-tool.

server/customtool.go

@@ -462,9 +462,16 @@ func (s *Server) httpProxyTransport(spec httpExec) *http.Transport {
 	proxyStr := strings.TrimSpace(spec.Proxy)
 	var caFile string
 	if spec.UseRecordingProxy {
-		if addr := s.m.ProxyAddr(); addr != "" {
-			proxyStr = "http://" + addr
-			caFile = s.m.ProxyCACert()
+		// ProxyAddr is already a complete URL while capture is enabled, and
+		// returns the configured global proxy URL when capture is off. Keep both
+		// forms intact (including socks5:// and credentials).
+		if s.m.TrafficEnabled() {
+			if addr := s.m.ProxyAddr(); addr != "" {
+				proxyStr = addr
+				caFile = s.m.ProxyCACert()
+			}
+		} else if global := strings.TrimSpace(s.m.GlobalProxy()); global != "" {
+			proxyStr = global
 		}
 	}
 	if proxyStr == "" {
@@ -474,6 +481,9 @@ func (s *Server) httpProxyTransport(spec httpExec) *http.Transport {
 	if err != nil {
 		return nil
 	}
+	// Go 1.26+ supports http/https/socks5/socks5h in ProxyURL directly,
+	// including proxy userinfo. Keep one code path so existing HTTP tool
+	// behavior and proxy authentication remain unchanged.
 	tr := &http.Transport{Proxy: http.ProxyURL(pu)}
 	if caFile != "" {
 		if pem, err := os.ReadFile(caFile); err == nil {

server/dto.go

@@ -64,6 +64,7 @@ type TaskDTO struct {
 	ArchiveBlockedBy   string             `json:"archive_blocked_by_task_id,omitempty"`
 	CompanyIDs         []int64            `json:"company_ids"`
 	CoverageEnabled    bool               `json:"coverage_enabled"` // 资产覆盖度功能开关(创建时定)
+	WorkerCount        int                `json:"worker_count"`     // 本任务 Worker 数;0=用全局默认
 }
 
 // FindingSeverityDTO 是任务列表里按严重度分档的漏洞计数（严重/高/中/低）。
@@ -138,6 +139,7 @@ func taskDTO(t *Task, status string) TaskDTO {
 		SourceTaskIDs:      sourceIDs,
 		CompanyIDs:         lifecycle.CompanyIDs,
 		CoverageEnabled:    t.CoverageEnabled,
+		WorkerCount:        t.WorkerCount,
 	}
 }
 

server/engine.go

@@ -148,11 +148,19 @@ type Engine struct {
 	// (for example an exhausted failover chain) — there is no global-pair fallback.
 	resolve              func(t *Task) (*agent.Planner, *agent.Worker)
 	resolveAuthoritative bool
+	workerResolve        func(t *Task, slot WorkerSlot) (*agent.Worker, error)
+	slotState            sync.Map // task/slot -> slotAvailability
+	slotPlans            sync.Map // task -> map[slot name]profile ID; immutable for a run
 	// readiness reports whether a global LLM provider is configured — the signal behind
 	// Ready()/the llm_configured indicator. Wired once at startup; nil → not ready.
 	readiness func() bool
 }
 
+type slotAvailability struct {
+	available bool
+	lastEmit  time.Time
+}
+
 type taskRuntime struct {
 	ctx    context.Context
 	cancel context.CancelFunc
@@ -285,6 +293,14 @@ func (e *Engine) StopTask(taskID string) {
 	e.stamped.Delete(taskID)
 	e.inflight.Delete(taskID)
 	e.coordStarted.Delete(taskID)
+	e.slotPlans.Delete(taskID)
+	e.slotState.Range(func(key, _ any) bool {
+		keyText, ok := key.(string)
+		if ok && strings.HasPrefix(keyText, taskID+"/") {
+			e.slotState.Delete(key)
+		}
+		return true
+	})
 	e.deleteMu.Lock()
 	e.deleting.Delete(taskID)
 	e.deleteMu.Unlock()
@@ -715,6 +731,37 @@ func (e *Engine) SetAuthoritativeAgentResolver(fn func(t *Task) (*agent.Planner,
 	e.resolveAuthoritative = true
 }
 
+func (e *Engine) SetWorkerResolver(fn func(t *Task, slot WorkerSlot) (*agent.Worker, error)) {
+	e.workerResolve = fn
+}
+
+func (e *Engine) reportSlotAvailability(t *Task, slot WorkerSlot, err error) {
+	key := t.ID + "/" + slot.Name
+	now := time.Now()
+	value, loaded := e.slotState.LoadOrStore(key, slotAvailability{available: err == nil, lastEmit: now})
+	state := value.(slotAvailability)
+	if err == nil {
+		if loaded && !state.available {
+			e.slotState.Store(key, slotAvailability{available: true, lastEmit: now})
+			e.emitActivity(t, db.Activity{Worker: "system", Kind: "text", Summary: fmt.Sprintf("%s 绑定的 LLM 配置 #%d 已恢复，可以继续领取任务", slot.Name, slot.ProfileID)})
+		}
+		return
+	}
+	if loaded && state.available {
+		state.available = false
+		state.lastEmit = now
+		e.slotState.Store(key, state)
+	} else if loaded && now.Sub(state.lastEmit) < 5*time.Minute {
+		return
+	} else {
+		state = slotAvailability{available: false, lastEmit: now}
+		e.slotState.Store(key, state)
+	}
+	summary := fmt.Sprintf("%s 绑定的 LLM 配置 #%d 不可用（已删除或配置无效），该会话暂停领取任务，不会回退到默认模型", slot.Name, slot.ProfileID)
+	e.emitActivity(t, db.Activity{Worker: "system", Kind: "text", IsError: true, Summary: summary, Detail: err.Error()})
+	log.Printf("[worker %s] task %s pinned profile #%d unavailable: %v", slot.Name, t.ID, slot.ProfileID, err)
+}
+
 // snapshotFor returns the planner/worker a task should run on, from the task-router
 // resolver. nil,nil means the task is deliberately unavailable (e.g. an exhausted
 // failover chain); there is no global-pair fallback.
@@ -746,7 +793,11 @@ func (e *Engine) ReadyFor(t *Task) bool {
 // but no-op until an LLM is configured (so a task created while idle picks up
 // automatically once LLM is set from the UI).
 func (e *Engine) Run(ctx context.Context, t *Task) {
-	workers := e.m.Workers()
+	slots, err := workerSlotsForTask(t, e.m.WorkerPools(), e.m.Workers())
+	if err != nil {
+		log.Printf("[engine] task %s worker pool invalid, using legacy workers: %v", t.ID, err)
+		slots, _ = ExpandWorkerSlots(nil, workerCountForTask(t, e.m.Workers()))
+	}
 	e.deleteMu.RLock()
 	if e.IsDeleting(t.ID) {
 		e.deleteMu.RUnlock()
@@ -757,13 +808,18 @@ func (e *Engine) Run(ctx context.Context, t *Task) {
 		t.Notify() // already running — just nudge a planning round
 		return
 	}
-	rt := e.registerTaskRoutines(ctx, t.ID, 1+workers)
+	rt := e.registerTaskRoutines(ctx, t.ID, 1+len(slots))
+	plan := make(map[string]int64, len(slots))
+	for _, slot := range slots {
+		plan[slot.Name] = slot.ProfileID
+	}
+	e.slotPlans.Store(t.ID, plan)
 	e.deleteMu.RUnlock()
 	e.touch(t.ID)
 	runTaskRoutine(rt, func(loopCtx context.Context) { e.plannerLoop(loopCtx, t) })
-	for i := 0; i < workers; i++ {
-		name := fmt.Sprintf("work#%d", i+1)
-		runTaskRoutine(rt, func(loopCtx context.Context) { e.workerLoop(loopCtx, t, name) })
+	for _, slot := range slots {
+		slot := slot
+		runTaskRoutine(rt, func(loopCtx context.Context) { e.workerLoop(loopCtx, t, slot) })
 	}
 	e.startDeadlineCoordinator(ctx, t) // 任务级超时定时器(仅 timeout>0;去重)
 	// 仅在「完全没有活动意图(open+running)」时才 kick 首轮规划。带种子意图的任务:种子已
@@ -776,6 +832,27 @@ func (e *Engine) Run(ctx context.Context, t *Task) {
 	}
 }
 
+// workerSlotsForTask preserves an explicit task concurrency override. Tasks
+// using the global default inherit its model pools and their pinned profiles.
+func workerSlotsForTask(t *Task, pools []WorkerPool, global int) ([]WorkerSlot, error) {
+	if t != nil && t.WorkerCount > 0 {
+		return ExpandWorkerSlots(nil, workerCountForTask(t, global))
+	}
+	return ExpandWorkerSlots(pools, global)
+}
+
+func workerCountForTask(t *Task, global int) int {
+	if t != nil && t.WorkerCount > 0 {
+		// Direct database edits can bypass validation; cap those values so the
+		// engine never starts more than the supported per-task worker count.
+		if t.WorkerCount > db.MaxTaskWorkerCount {
+			return db.MaxTaskWorkerCount
+		}
+		return t.WorkerCount
+	}
+	return global
+}
+
 // plannerHeartbeatInterval 解析任务的 planner 心跳间隔。db.CreateTask 已归一
 // (低于 600 一律抬到 600);这里再兜一次底,防内存态异常值。
 func plannerHeartbeatInterval(t *Task) time.Duration {
@@ -913,14 +990,32 @@ func (e *Engine) plannerLoop(ctx context.Context, t *Task) {
 	}
 }
 
-func (e *Engine) workerLoop(ctx context.Context, t *Task, name string) {
+func (e *Engine) workerLoop(ctx context.Context, t *Task, slot WorkerSlot) {
+	name := slot.Name
 	for {
 		select {
 		case <-ctx.Done():
 			return
 		default:
 		}
-		_, worker := e.snapshotFor(t)
+		var worker *agent.Worker
+		var resolveErr error
+		if slot.ProfileID > 0 {
+			if e.workerResolve == nil {
+				resolveErr = errors.New("pinned worker resolver unavailable")
+			} else {
+				worker, resolveErr = e.workerResolve(t, slot)
+				if worker == nil && resolveErr == nil {
+					resolveErr = errors.New("pinned worker resolver returned no worker")
+				}
+			}
+			e.reportSlotAvailability(t, slot, resolveErr)
+			if resolveErr != nil {
+				worker = nil
+			}
+		} else {
+			_, worker = e.snapshotFor(t)
+		}
 		if worker == nil {
 			if sleepCtx(ctx, 1500*time.Millisecond) {
 				return
@@ -1178,10 +1273,6 @@ func (e *Engine) runDetachedIntent(ctx context.Context, t *Task, intentID int64,
 			e.decInflight(t.ID)
 		}
 	}()
-	_, worker := e.snapshotFor(t)
-	if worker == nil {
-		return fmt.Errorf("worker 尚未就绪")
-	}
 	node, err := t.Store.GetNode(intentID)
 	if err != nil {
 		return err
@@ -1189,6 +1280,30 @@ func (e *Engine) runDetachedIntent(ctx context.Context, t *Task, intentID int64,
 	if node == nil || node.Kind != db.KindIntent {
 		return fmt.Errorf("intent not found")
 	}
+	owner := node.Owner
+	var worker *agent.Worker
+	if value, ok := e.slotPlans.Load(t.ID); ok {
+		if plan, ok := value.(map[string]int64); ok {
+			if profileID, pinned := plan[owner]; pinned && profileID > 0 {
+				if e.workerResolve == nil {
+					return fmt.Errorf("意图所属的 LLM 配置 #%d 不可用：pinned worker resolver unavailable", profileID)
+				}
+				worker, err = e.workerResolve(t, WorkerSlot{Name: owner, ProfileID: profileID})
+				if err != nil || worker == nil {
+					if err == nil {
+						err = errors.New("worker resolver returned no worker")
+					}
+					return fmt.Errorf("意图所属的 LLM 配置 #%d 不可用：%w", profileID, err)
+				}
+			}
+		}
+	}
+	if worker == nil {
+		_, worker = e.snapshotFor(t)
+		if worker == nil {
+			return fmt.Errorf("worker 尚未就绪")
+		}
+	}
 	changed, err := t.Store.CompareAndSetIntentState(intentID, "paused", "running")
 	if err != nil {
 		return err

server/engine_worker_test.go

@@ -0,0 +1,59 @@
+package server
+
+import (
+	"fmt"
+	"testing"
+
+	"github.com/Autumn-27/artex/db"
+)
+
+func TestWorkerCountForTaskUsesOverrideOrGlobalDefault(t *testing.T) {
+	if got := workerCountForTask(&Task{WorkerCount: 5}, 3); got != 5 {
+		t.Fatalf("override = %d, want 5", got)
+	}
+	if got := workerCountForTask(&Task{WorkerCount: 0}, 3); got != 3 {
+		t.Fatalf("zero override = %d, want global 3", got)
+	}
+	if got := workerCountForTask(&Task{WorkerCount: -1}, 3); got != 3 {
+		t.Fatalf("negative override = %d, want global 3", got)
+	}
+	if got := workerCountForTask(&Task{WorkerCount: db.MaxTaskWorkerCount + 1}, 3); got != db.MaxTaskWorkerCount {
+		t.Fatalf("oversized override = %d, want capped %d", got, db.MaxTaskWorkerCount)
+	}
+}
+
+func TestWorkerSlotsForTaskCombinesOverridesAndModelPools(t *testing.T) {
+	pools := []WorkerPool{{ProfileID: 7, Workers: 2}, {ProfileID: 12, Workers: 1}}
+	for _, tc := range []struct {
+		name     string
+		task     *Task
+		pools    []WorkerPool
+		global   int
+		profiles []int64
+	}{
+		{"task inherits model pools", &Task{}, pools, 9, []int64{7, 7, 12}},
+		{"task override replaces global pools", &Task{WorkerCount: 2}, pools, 9, []int64{0, 0}},
+		{"task inherits legacy count", &Task{}, nil, 4, []int64{0, 0, 0, 0}},
+		{"task override without pools", &Task{WorkerCount: 1}, nil, 4, []int64{0}},
+		{"invalid task value uses pools", &Task{WorkerCount: -1}, pools, 9, []int64{7, 7, 12}},
+	} {
+		t.Run(tc.name, func(t *testing.T) {
+			slots, err := workerSlotsForTask(tc.task, tc.pools, tc.global)
+			if err != nil || len(slots) != len(tc.profiles) {
+				t.Fatalf("slots=%+v err=%v, want profiles %v", slots, err, tc.profiles)
+			}
+			for i, profile := range tc.profiles {
+				if slots[i].ProfileID != profile || slots[i].Name != fmt.Sprintf("work#%d", i+1) {
+					t.Fatalf("slot[%d]=%+v, want profile %d", i, slots[i], profile)
+				}
+			}
+		})
+	}
+	slots, err := workerSlotsForTask(&Task{WorkerCount: db.MaxTaskWorkerCount + 1}, pools, 3)
+	if err != nil || len(slots) != db.MaxTaskWorkerCount {
+		t.Fatalf("oversized task override: slots=%d err=%v", len(slots), err)
+	}
+	if _, err := workerSlotsForTask(&Task{}, []WorkerPool{{ProfileID: 0, Workers: 1}}, 3); err == nil {
+		t.Fatal("invalid inherited pool accepted")
+	}
+}

server/finding_case_reviews.go

@@ -33,9 +33,9 @@ func (s *Server) toolCompleteFindingCaseReview() actool.CoreTool {
 	decision := objSchema(map[string]any{
 		"finding_id":          idParam("所选独立漏洞记录 ID"),
 		"verdict":             map[string]any{"type": "string", "enum": []string{"independent", "merged", "suggested"}},
-		"reason":              strParam("判断理由，接口或读写影响不同不能单独证明不同根因"),
-		"root_cause_evidence": strParam("原始报告/证据中支持实际共同或不同根因的具体材料；不确定则说明缺失证据及已保存建议")}, "finding_id", "verdict", "reason", "root_cause_evidence")
-	return wrTool("complete_finding_case_review", "历史整理结束前提交每条所选记录的根因判断并验收：必须已读取全部记录，疑似必须已有建议，归并必须已有完整最新统一报告。验收失败不能宣布完成。", objSchema(map[string]any{"summary": strParam("整理结果总结"), "decisions": map[string]any{"type": "array", "items": decision}}, "summary", "decisions"), func(ctx context.Context, raw json.RawMessage) (actool.Result, error) {
+		"reason":              strParam("判断理由，引用已证据化的共同点或不同代码路径；仅类型、标题或资产相同不足以证明共同根因"),
+		"root_cause_evidence": strParam("原始报告/证据中支持实际共同或不同根因的具体材料；suggested 须说明具体共同点、缺失的关键证据及已保存分组；没有具体共同点应判 independent")}, "finding_id", "verdict", "reason", "root_cause_evidence")
+	return wrTool("complete_finding_case_review", "历史整理结束前提交每条所选记录的根因判断并验收：必须已读取全部记录，疑似必须已有待确认分组建议；归并只需已在活动文件夹中，统一报告可稍后补齐。验收失败不能宣布完成。", objSchema(map[string]any{"summary": strParam("整理结果总结"), "decisions": map[string]any{"type": "array", "items": decision}}, "summary", "decisions"), func(ctx context.Context, raw json.RawMessage) (actool.Result, error) {
 		var a findingReviewConclusion
 		if err := json.Unmarshal(raw, &a); err != nil {
 			return actool.Errorf(err.Error()), nil
@@ -101,7 +101,9 @@ func (s *Server) saveFindingReviewConclusion(ctx context.Context, conv int64, a
 				}
 			case "suggested":
 				var pending bool
-				if err := tx.QueryRow(`SELECT EXISTS(SELECT 1 FROM finding_case_suggestions WHERE state='pending' AND (left_id=$1 OR right_id=$1))`, id).Scan(&pending); err != nil {
+				if err := tx.QueryRow(`SELECT EXISTS(SELECT 1 FROM finding_case_suggestions s
+                    JOIN finding_case_suggestion_members sm ON sm.suggestion_id=s.id
+                    WHERE s.state='pending' AND sm.finding_id=$1)`, id).Scan(&pending); err != nil {
 					return err
 				}
 				if !pending {
@@ -115,17 +117,12 @@ func (s *Server) saveFindingReviewConclusion(ctx context.Context, conv int64, a
 				return errors.New("无效判断类型")
 			}
 			if cid.Valid {
-				var report string
-				var version, rv int64
 				var active bool
-				if err := tx.QueryRow(`SELECT report,version,report_version,active FROM finding_cases WHERE id=$1 FOR SHARE`, cid.Int64).Scan(&report, &version, &rv, &active); err != nil {
+				if err := tx.QueryRow(`SELECT active FROM finding_cases WHERE id=$1 FOR SHARE`, cid.Int64).Scan(&active); err != nil {
 					return err
 				}
-				if !active || rv != version {
-					return fmt.Errorf("文件夹 #%d 的统一报告待更新，请重新读取后更新", cid.Int64)
-				}
-				if err := validateUnifiedFindingReport(report); err != nil {
-					return err
+				if !active {
+					return fmt.Errorf("文件夹 #%d 已关闭", cid.Int64)
 				}
 			}
 		}

server/finding_case_reviews_test.go

@@ -4,11 +4,46 @@ import (
 	"context"
 	"encoding/json"
 	"fmt"
-	"github.com/Autumn-27/artex/intercept"
 	"strings"
 	"testing"
+
+	"github.com/Autumn-27/artex/intercept"
 )
 
+func TestFindingCaseReviewTaskMessagePolicy(t *testing.T) {
+	message := fmt.Sprintf(findingCaseReviewTaskMessage, `[101,102,103]`)
+	for _, want := range []string{
+		"finding_ids=[101,102,103]",
+		"根因明确相同则直接归并",
+		"相同端点、处理器/后端组件、参数或缺失检查",
+		"缺少确认共同根因的关键证据",
+		"一个覆盖2..20条同任务疑似成员的分组",
+		"不要拆成两两建议",
+		"仅类型或资产相同不足以建议",
+		"没有具体共同点时保持 independent",
+		"不同代码路径时可判 independent",
+		"尽力更新统一报告",
+		"统一报告暂缺或待更新不阻塞 complete_finding_case_review",
+		"不归并未选记录",
+		"不探测目标",
+		"遵守人工否决",
+	} {
+		if !strings.Contains(message, want) {
+			t.Errorf("task message missing policy: %s", want)
+		}
+	}
+	if strings.Contains(message, "不确定提交建议") {
+		t.Fatal("task message suggests solely because of uncertainty")
+	}
+	schema, err := json.Marshal((&Server{}).toolCompleteFindingCaseReview().InputSchema())
+	if err != nil {
+		t.Fatal(err)
+	}
+	if !strings.Contains(string(schema), "没有具体共同点应判 independent") || strings.Contains(string(schema), "不确定则说明缺失证据及已保存建议") {
+		t.Fatal("completion schema must not require suggestions solely because of uncertainty")
+	}
+}
+
 const completeFindingReviewReport = `# 同一权限缺陷
 ## 概述
 同一对象权限检查缺失。
@@ -43,7 +78,7 @@ func TestUnifiedFindingReportRejectsIncompleteAndTruncated(t *testing.T) {
 	}
 }
 
-func TestFindingCaseReviewCompletionRequiresReadsAndFreshReport(t *testing.T) {
+func TestFindingCaseReviewCompletionRequiresReadsNotFreshReport(t *testing.T) {
 	s, task, ids := testCaseServer(t)
 	runs, err := s.m.pg.CreateFindingCaseReviews(t.Context(), map[int64][]int64{task: ids})
 	if err != nil {
@@ -71,13 +106,6 @@ func TestFindingCaseReviewCompletionRequiresReadsAndFreshReport(t *testing.T) {
 	if err != nil {
 		t.Fatal(err)
 	}
-	if err := s.saveFindingReviewConclusion(ctx, conv, a); err == nil {
-		t.Fatal("missing unified report accepted")
-	}
-	c, _ := s.m.pg.GetFindingCase(cid)
-	if err := s.m.pg.UpdateFindingCaseReport(ctx, cid, c.Version, "orders", completeFindingReviewReport, "high", "verified impact"); err != nil {
-		t.Fatal(err)
-	}
 	if err := s.saveFindingReviewConclusion(ctx, conv, a); err != nil {
 		t.Fatal(err)
 	}
@@ -90,8 +118,8 @@ func TestFindingCaseReviewCompletionRequiresReadsAndFreshReport(t *testing.T) {
 	if _, err := s.m.pg.Exec(`UPDATE findings SET evidence='new proof' WHERE id=$1`, ids[0]); err != nil {
 		t.Fatal(err)
 	}
-	if err := s.validateFindingReviewCompletion(conv); err == nil {
-		t.Fatal("stale group reported completed")
+	if err := s.validateFindingReviewCompletion(conv); err != nil {
+		t.Fatalf("stale unified report should not block merged conclusion: %v", err)
 	}
 }
 
@@ -138,7 +166,7 @@ func TestFindingCaseReviewSuggestionNeedsSavedSuggestion(t *testing.T) {
 	if err := s.saveFindingReviewConclusion(ctx, conv, a); err == nil {
 		t.Fatal("unpersisted suggestion accepted")
 	}
-	if _, err := s.m.pg.SuggestFindingCase(ctx, task, ids[0], ids[1], "orders", "common component unknown"); err != nil {
+	if _, err := s.m.pg.SuggestFindingCase(ctx, task, []int64{ids[0], ids[1]}, "orders", "common component unknown"); err != nil {
 		t.Fatal(err)
 	}
 	if err := s.saveFindingReviewConclusion(ctx, conv, a); err != nil {

server/finding_cases.go

@@ -6,7 +6,9 @@ import (
 	"encoding/json"
 	"errors"
 	"fmt"
+	"log"
 	"net/http"
+	"strconv"
 	"strings"
 
 	"github.com/Autumn-27/artex/db"
@@ -203,12 +205,22 @@ func (s *Server) caseSuggestions(w http.ResponseWriter, r *http.Request) {
 }
 func (s *Server) resolveCaseSuggestion(w http.ResponseWriter, r *http.Request) {
 	var a struct {
-		Accept bool `json:"accept"`
+		Accept     bool     `json:"accept"`
+		ExcludeIDs []string `json:"exclude_ids"`
 	}
 	if !decodeCaseRequest(w, r, &a) {
 		return
 	}
-	id, err := s.m.pg.ResolveFindingCaseSuggestion(r.Context(), int64(atoiDefault(r.PathValue("id"), 0)), a.Accept)
+	exclude := make([]int64, 0, len(a.ExcludeIDs))
+	for _, raw := range a.ExcludeIDs {
+		id, err := strconv.ParseInt(strings.TrimSpace(raw), 10, 64)
+		if err != nil || id <= 0 {
+			caseHTTPError(w, errors.New("无效的排除记录编号"))
+			return
+		}
+		exclude = append(exclude, id)
+	}
+	id, err := s.m.pg.ResolveFindingCaseSuggestion(r.Context(), int64(atoiDefault(r.PathValue("id"), 0)), a.Accept, exclude)
 	if err != nil {
 		caseHTTPError(w, err)
 		return
@@ -253,10 +265,10 @@ func (s *Server) findingCaseTools() []actool.CoreTool {
 		name, desc string
 		required   []string
 	}{
-		{"search_finding_duplicates", "只读检索同任务候选，返回摘要与资产；等级不同也可能重复。支持分页，按需读取详情。", []string{"finding_id"}},
+		{"search_finding_duplicates", "只读检索同任务候选摘要与资产，按分数排序并返回 score、tier、hit_reasons：strong=同漏洞类型且资产重叠，weak=未同时具备同漏洞类型和资产重叠，请结合 hit_reasons 核对，仅作参考。等级不同也可能重复。支持分页，按需读取详情。", []string{"finding_id"}},
 		{"get_finding_record", "读取独立漏洞完整原始证据和报告，返回探索节点编号以便追溯执行记录。", []string{"finding_id"}},
 		{"merge_finding_records", "证据明确证明同一缺陷时归并；保留所有原始记录。不跨任务、不改变原评级，不可绕过人工否决。", []string{"task_id", "finding_ids", "title", "reason"}},
-		{"suggest_finding_merge", "证据不足时提交疑似重复建议，待人工确认；不是自动合并。只允许两条同任务记录。", []string{"task_id", "finding_ids", "title", "reason"}},
+		{"suggest_finding_merge", "证据存在具体共同点但缺少关键证据时，提交一个疑似重复分组（2..20 条同任务记录，一次覆盖全部疑似成员，不要拆成两两建议），待人工确认；不是自动合并。仅类型/资产相同不应提交。", []string{"task_id", "finding_ids", "title", "reason"}},
 		{"get_finding_case", "读取归并组和成员分页。先记录version，再读取成员的原报告、证据及流量；版本变化必须重新读取。", []string{"case_id"}},
 		{"update_finding_case_report", "保存统一报告及评级理由，version必须来自实际读取。保留各条原报告，不新建漏洞。", []string{"case_id", "version", "title", "report", "severity", "severity_reason"}},
 	}
@@ -312,8 +324,9 @@ func (s *Server) performFindingCaseTool(ctx context.Context, name string, a find
 		if scope != nil {
 			scopes = append(scopes, scope)
 		}
-		items, total, err := s.m.pg.FindingCaseCandidates(fid, page, 20, scopes...)
-		return map[string]any{"items": items, "total": total, "page": page, "page_size": 20}, err
+		items, total, strongTotal, err := s.m.pg.FindingCaseCandidates(fid, page, 20, scopes...)
+		return map[string]any{"items": items, "total": total, "strong_total": strongTotal, "page": page, "page_size": 20,
+			"note": "按分数排序；weak 表示未同时具备同漏洞类型和资产重叠，请结合 hit_reasons 核对，仅作参考，通常不应据此建议归并"}, err
 	case "get_finding_record":
 		f, err := s.m.pg.GetFinding(fid)
 		if err != nil {
@@ -341,11 +354,11 @@ func (s *Server) performFindingCaseTool(ctx context.Context, name string, a find
 		return map[string]string{"case_id": i64s(id)}, err
 	case "suggest_finding_merge":
 		ids := a.ids()
-		if len(ids) != 2 {
-			return nil, errors.New("疑似重复建议需要两条记录")
+		if len(ids) < 2 || len(ids) > 20 {
+			return nil, errors.New("疑似重复分组需要2至20条记录")
 		}
-		id, err := s.m.pg.SuggestFindingCase(ctx, parseProfileID(a.TaskID), ids[0], ids[1], a.Title, a.Reason)
-		return map[string]string{"suggestion_id": i64s(id)}, err
+		id, err := s.m.pg.SuggestFindingCase(ctx, parseProfileID(a.TaskID), ids, a.Title, a.Reason)
+		return map[string]any{"suggestion_id": i64s(id), "finding_ids": ids}, err
 	case "get_finding_case":
 		c, err := s.m.pg.GetFindingCase(cid)
 		if err != nil {
@@ -378,14 +391,22 @@ func (s *Server) seedFindingCaseTools() {
 	if err != nil || a == nil {
 		return
 	}
-	const flag = "finding_case_tools_v2_review"
+	const flag = "finding_case_tools_v3_review"
 	if v, _, _ := s.m.pg.GetSetting(flag); v == "true" {
 		return
 	}
 	names := []string{}
 	previous, _, _ := s.m.pg.GetSetting("finding_case_tools_v1")
+	// v3 refreshes descriptions and schemas without repeating the v2 binding migration.
+	previousReview, _, _ := s.m.pg.GetSetting("finding_case_tools_v2_review")
 	for _, t := range s.findingCaseTools() {
 		schema, _ := json.Marshal(t.InputSchema())
+		if err := s.m.pg.RefreshToolDefaults(t.Name(), t.Description(), schema); err != nil {
+			log.Printf("[finding-cases] refresh %s schema failed: %v", t.Name(), err)
+		}
+		if previousReview == "true" {
+			continue
+		}
 		if err := s.m.pg.SeedTool(t.Name(), t.Description(), schema, json.RawMessage(`["reporter"]`)); err != nil {
 			return
 		}
@@ -393,11 +414,19 @@ func (s *Server) seedFindingCaseTools() {
 			names = append(names, t.Name())
 		}
 	}
-	if err := s.m.pg.AddAgentToToolBinding("reporter", names); err != nil {
-		return
+	if previousReview != "true" {
+		if err := s.m.pg.AddAgentToToolBinding("reporter", names); err != nil {
+			return
+		}
+		if err := s.m.pg.SetSetting("finding_case_tools_v2_review", "true"); err != nil {
+			return
+		}
 	}
 	_ = s.m.pg.SetSetting(flag, "true")
 }
+
+const findingCaseReviewTaskMessage = "人工选择整理（只读原始记录）：本任务所选 finding_ids=%s。逐条读取已有报告，仅判断同一实际根因；根因明确相同则直接归并。只有证据指向相同端点、处理器/后端组件、参数或缺失检查等具体共同点，但缺少确认共同根因的关键证据时，才提交一个覆盖2..20条同任务疑似成员的分组，不要拆成两两建议；仅类型或资产相同不足以建议，没有具体共同点时保持 independent，不要因为不确定就提交建议。证据表明属于不同代码路径时可判 independent。已有或新建组应尽力更新统一报告；统一报告暂缺或待更新不阻塞 complete_finding_case_review，可由后续运行或人工补齐。禁止生成或覆盖原始报告、改评级状态、补绑证据，不归并未选记录，不探测目标，遵守人工否决。结束前调用 complete_finding_case_review 提交逐条判断并通过验收。"
+
 func (s *Server) reviewFindingCases(w http.ResponseWriter, r *http.Request) {
 	var a findingCaseRequest
 	if !decodeCaseRequest(w, r, &a) {
@@ -463,7 +492,7 @@ func (s *Server) reviewFindingCases(w http.ResponseWriter, r *http.Request) {
 		taskID, fids := reservation.TaskID, reservation.FindingIDs
 		taskContext := taskContexts[taskID]
 		raw, _ := json.Marshal(fids)
-		msg := fmt.Sprintf("人工选择整理（只读原始记录）：本任务所选 finding_ids=%s。逐条读取已有报告，仅判断同一实际根因，明确则归并，不确定提交建议；已有组更新统一报告。禁止生成或覆盖原始报告、改评级状态、补绑证据，不归并未选记录，不探测目标。结束前调用 complete_finding_case_review 提交逐条判断并通过验收。", raw)
+		msg := fmt.Sprintf(findingCaseReviewTaskMessage, raw)
 		item := triggeredRun{agentKey: "reporter", title: "漏洞整理 · task#" + i64s(taskID), message: msg, taskID: taskID, taskDesc: taskContext.taskDesc, taskGoal: taskContext.taskGoal, conversationID: reservation.ConversationID}
 		cfg := s.readTriggerBehavior("reporter")
 		s.queueMu.Lock()

server/finding_cases_test.go

@@ -42,6 +42,119 @@ func testCaseServer(t *testing.T) (*Server, int64, []int64) {
 	t.Cleanup(func() { _, _ = pg.Exec(`DELETE FROM findings WHERE task_id=$1`, task.ID); _ = pg.DeleteTask(task.ID) })
 	return &Server{m: &Manager{pg: pg}}, task.ID, ids
 }
+
+func TestFindingCaseToolMigrationPreservesBindings(t *testing.T) {
+	s, _, _ := testCaseServer(t)
+	pg := s.m.pg
+	for _, flag := range []string{"finding_case_tools_v1", "finding_case_tools_v2_review", "finding_case_tools_v3_review"} {
+		value, exists, err := pg.GetSetting(flag)
+		if err != nil {
+			t.Fatal(err)
+		}
+		t.Cleanup(func() {
+			if exists {
+				if err := pg.SetSetting(flag, value); err != nil {
+					t.Error(err)
+				}
+			} else if _, err := pg.Exec(`DELETE FROM settings WHERE key=$1`, flag); err != nil {
+				t.Error(err)
+			}
+		})
+	}
+	for _, tool := range s.findingCaseTools() {
+		key := tool.Name()
+		old, err := pg.GetTool(key)
+		if err != nil {
+			t.Fatal(err)
+		}
+		t.Cleanup(func() {
+			if old == nil {
+				if _, err := pg.Exec(`DELETE FROM tools WHERE key=$1`, key); err != nil {
+					t.Error(err)
+				}
+				return
+			}
+			bindings, _ := json.Marshal(old.Agents)
+			if err := pg.UpdateTool(key, old.Description, old.Schema, bindings, old.Enabled); err != nil {
+				t.Error(err)
+			}
+		})
+		if err := pg.SeedTool(key, tool.Description(), nil, nil); err != nil {
+			t.Fatal(err)
+		}
+	}
+	for _, tc := range []struct {
+		name          string
+		v1, v2, fresh bool
+	}{
+		{name: "fresh", fresh: true},
+		{name: "initial"},
+		{name: "v1_to_v3", v1: true},
+		{name: "v2_to_v3", v1: true, v2: true},
+		{name: "v2_without_v1_flag", v2: true},
+	} {
+		t.Run(tc.name, func(t *testing.T) {
+			for flag, value := range map[string]bool{
+				"finding_case_tools_v1":        tc.v1,
+				"finding_case_tools_v2_review": tc.v2,
+				"finding_case_tools_v3_review": false,
+			} {
+				if value {
+					if err := pg.SetSetting(flag, "true"); err != nil {
+						t.Fatal(err)
+					}
+				} else if _, err := pg.Exec(`DELETE FROM settings WHERE key=$1`, flag); err != nil {
+					t.Fatal(err)
+				}
+			}
+			for _, tool := range s.findingCaseTools() {
+				if tc.fresh {
+					if _, err := pg.Exec(`DELETE FROM tools WHERE key=$1`, tool.Name()); err != nil {
+						t.Fatal(err)
+					}
+				} else if err := pg.UpdateTool(tool.Name(), "old description", json.RawMessage(`{}`), json.RawMessage(`["custom-agent"]`), false); err != nil {
+					t.Fatal(err)
+				}
+			}
+			s.seedFindingCaseTools()
+			for _, tool := range s.findingCaseTools() {
+				row, err := pg.GetTool(tool.Name())
+				if err != nil || row == nil {
+					t.Fatal("missing tool", tool.Name(), err)
+				}
+				wantReporter := !tc.v2 && (!tc.v1 || tool.Name() == "complete_finding_case_review")
+				if contains(row.Agents, "reporter") != wantReporter || contains(row.Agents, "custom-agent") == tc.fresh || row.Enabled != tc.fresh {
+					t.Fatalf("binding/enabled configuration changed incorrectly: %+v", row)
+				}
+				if row.Description != tool.Description() || len(row.Schema) <= 2 {
+					t.Fatalf("description/schema not refreshed: %+v", row)
+				}
+			}
+			for _, flag := range []string{"finding_case_tools_v2_review", "finding_case_tools_v3_review"} {
+				if value, _, err := pg.GetSetting(flag); err != nil || value != "true" {
+					t.Fatalf("migration flag %s not saved: %q, %v", flag, value, err)
+				}
+			}
+			for _, tool := range s.findingCaseTools() {
+				if err := pg.RemoveAgentFromTool("reporter", tool.Name()); err != nil {
+					t.Fatal(err)
+				}
+			}
+			// A refresh retry must not replay a completed legacy binding migration.
+			if err := pg.SetSetting("finding_case_tools_v3_review", "false"); err != nil {
+				t.Fatal(err)
+			}
+			s.seedFindingCaseTools()
+			for _, tool := range s.findingCaseTools() {
+				row, err := pg.GetTool(tool.Name())
+				if err != nil || row == nil || contains(row.Agents, "reporter") {
+					t.Fatal("refresh restored an operator-removed binding", row, err)
+				}
+			}
+		})
+	}
+}
+
 func TestFindingCaseAPIPaginationAndTools(t *testing.T) {
 	s, task, ids := testCaseServer(t)
 	ctx := context.Background()

server/manager.go

@@ -1,6 +1,7 @@
 package server
 
 import (
+	"context"
 	"database/sql"
 	"encoding/json"
 	"errors"
@@ -59,6 +60,7 @@ type Task struct {
 	TimeoutSeconds       int                    `json:"timeout_seconds"`
 	PlanHeartbeatSeconds int                    `json:"plan_heartbeat_seconds"` // planner 心跳触发间隔(秒)
 	CoverageEnabled      bool                   `json:"coverage_enabled"`       // 资产覆盖度功能开关(创建时定,默认开)
+	WorkerCount          int                    `json:"worker_count"`           // 本任务 Worker 数;0=用全局默认
 	FirstRunAt           int64                  `json:"first_run_at,omitempty"`
 	DeadlineAt           int64                  `json:"deadline_at,omitempty"`
 	Store                *pgdb.ExplorationStore `json:"-"`
@@ -265,6 +267,7 @@ const (
 	// only routes the search backend) and the per-profile LLM proxy.
 	settingGlobalProxy = "global_proxy"
 	settingWorkers     = "workers"
+	settingWorkerPools = "worker_pools"
 	settingLLMRecord   = "llm_record"
 	// LLM 轮询(故障转移)。默认关闭——开启后走「全局激活配置」的 agent 在当前配置
 	// 不可用(余额不足/key 失效/限流/服务异常)时自动切到下一个配置。
@@ -291,6 +294,102 @@ const (
 	defaultConcurrencyLimit = 5
 )
 
+// WorkerPool binds a number of worker slots to one LLM profile. Profile IDs are
+// encoded as JSON strings to match the frontend LLMProfile API.
+type WorkerPool struct {
+	ProfileID int64 `json:"profile_id,string"`
+	Workers   int   `json:"workers"`
+}
+
+func (p *WorkerPool) UnmarshalJSON(data []byte) error {
+	var raw struct {
+		ProfileID json.RawMessage `json:"profile_id"`
+		Workers   int             `json:"workers"`
+	}
+	if err := json.Unmarshal(data, &raw); err != nil {
+		return err
+	}
+	var id int64
+	if err := json.Unmarshal(raw.ProfileID, &id); err != nil {
+		var text string
+		if err := json.Unmarshal(raw.ProfileID, &text); err != nil {
+			return fmt.Errorf("profile_id 必须是整数或数字字符串")
+		}
+		parsed, err := strconv.ParseInt(strings.TrimSpace(text), 10, 64)
+		if err != nil {
+			return fmt.Errorf("profile_id 必须是整数或数字字符串")
+		}
+		id = parsed
+	}
+	p.ProfileID, p.Workers = id, raw.Workers
+	return nil
+}
+
+// WorkerSlot is the concrete routing plan for one task run.
+type WorkerSlot struct {
+	Name      string
+	ProfileID int64 // zero means legacy task-router resolution
+}
+
+// ValidateWorkerPools validates pool shape without requiring a database. Pool
+// entries with duplicate profile IDs are rejected so each profile has one
+// unambiguous slot group in the saved order.
+func ValidateWorkerPools(pools []WorkerPool) error {
+	if len(pools) == 0 {
+		return nil
+	}
+	seen := make(map[int64]bool, len(pools))
+	total := 0
+	for _, pool := range pools {
+		if pool.ProfileID <= 0 {
+			return fmt.Errorf("worker_pools 中 profile_id 必须大于 0")
+		}
+		if pool.Workers <= 0 {
+			return fmt.Errorf("worker_pools 中 workers 必须大于 0")
+		}
+		if seen[pool.ProfileID] {
+			return fmt.Errorf("worker_pools 中 profile_id %d 重复", pool.ProfileID)
+		}
+		seen[pool.ProfileID] = true
+		if total > 50-pool.Workers {
+			return fmt.Errorf("worker_pools 的 Worker 总数必须在 1 到 50 之间")
+		}
+		total += pool.Workers
+	}
+	if total < 1 || total > 50 {
+		return fmt.Errorf("worker_pools 的 Worker 总数必须在 1 到 50 之间")
+	}
+	return nil
+}
+
+// ExpandWorkerSlots expands pools in their saved order; invalid or empty legacy
+// settings use the historical worker count (default 3).
+func ExpandWorkerSlots(pools []WorkerPool, legacy int) ([]WorkerSlot, error) {
+	if err := ValidateWorkerPools(pools); err != nil {
+		return nil, err
+	}
+	if len(pools) == 0 {
+		if legacy <= 0 {
+			legacy = defaultWorkers
+		}
+		slots := make([]WorkerSlot, legacy)
+		for i := range slots {
+			slots[i] = WorkerSlot{Name: fmt.Sprintf("work#%d", i+1)}
+		}
+		return slots, nil
+	}
+	slots := make([]WorkerSlot, 0)
+	for _, pool := range pools {
+		for i := 0; i < pool.Workers; i++ {
+			slots = append(slots, WorkerSlot{ProfileID: pool.ProfileID})
+		}
+	}
+	for i := range slots {
+		slots[i].Name = fmt.Sprintf("work#%d", i+1)
+	}
+	return slots, nil
+}
+
 // ConcurrencyLimit returns whether the simultaneous-running-task cap is enabled and
 // its limit (default 5 when enabled but unset). limit is always >=1 when enabled.
 func (m *Manager) ConcurrencyLimit() (enabled bool, limit int) {
@@ -332,6 +431,64 @@ func (m *Manager) Workers() int {
 	return n
 }
 
+func (m *Manager) WorkerPools() []WorkerPool {
+	v, ok, err := m.pg.GetSetting(settingWorkerPools)
+	if err != nil || !ok || strings.TrimSpace(v) == "" {
+		return []WorkerPool{}
+	}
+	var pools []WorkerPool
+	if json.Unmarshal([]byte(v), &pools) != nil || ValidateWorkerPools(pools) != nil {
+		return []WorkerPool{}
+	}
+	if pools == nil {
+		return []WorkerPool{}
+	}
+	return pools
+}
+
+// SetWorkerPools validates profile references at save time and atomically
+// synchronizes the legacy workers setting with the pool total. An empty slice
+// restores legacy mode; an optional legacy count is used when clearing pools.
+func (m *Manager) SetWorkerPools(pools []WorkerPool, legacyWorkers ...int) error {
+	if pools == nil {
+		pools = []WorkerPool{}
+	}
+	if err := ValidateWorkerPools(pools); err != nil {
+		return err
+	}
+	if len(pools) > 0 {
+		for _, pool := range pools {
+			profile, err := m.pg.ProfileByID(pool.ProfileID)
+			if err != nil {
+				return err
+			}
+			if profile == nil {
+				return fmt.Errorf("worker_pools 中 profile_id %d 不存在", pool.ProfileID)
+			}
+		}
+	}
+	b, err := json.Marshal(pools)
+	if err != nil {
+		return err
+	}
+	workerCount := m.Workers()
+	if len(pools) > 0 {
+		workerCount = 0
+		for _, pool := range pools {
+			workerCount += pool.Workers
+		}
+	} else if len(legacyWorkers) > 0 {
+		if legacyWorkers[0] <= 0 {
+			return fmt.Errorf("workers 必须 >0")
+		}
+		workerCount = legacyWorkers[0]
+	}
+	return m.pg.SetSettingsAtomic(context.Background(), map[string]string{
+		settingWorkerPools: string(b),
+		settingWorkers:     strconv.Itoa(workerCount),
+	})
+}
+
 // SetWorkers persists the concurrent work-agent count. Values <=0 are rejected.
 func (m *Manager) SetWorkers(n int) error {
 	if n <= 0 {
@@ -849,8 +1006,8 @@ func taskFromPG(pt *pgdb.Task, store *pgdb.ExplorationStore, ic *intercept.Inter
 		SourceTaskIDs:  append([]int64(nil), pt.SourceTaskIDs...),
 		CompanyIDs:     append([]int64(nil), pt.CompanyIDs...),
 		TimeoutSeconds: pt.TimeoutSeconds, PlanHeartbeatSeconds: pt.PlanHeartbeatSeconds,
-		CoverageEnabled: pt.CoverageEnabled,
-		FirstRunAt:      unixOrZero(pt.FirstRunAt), DeadlineAt: unixOrZero(pt.DeadlineAt),
+		CoverageEnabled: pt.CoverageEnabled, WorkerCount: pt.WorkerCount,
+		FirstRunAt: unixOrZero(pt.FirstRunAt), DeadlineAt: unixOrZero(pt.DeadlineAt),
 		Store: store, Guard: guard.NewWithInterceptor(ic), notify: make(chan struct{}, 1),
 	}
 }

server/orchestration.go

@@ -524,7 +524,7 @@ func (s *Server) seedOrchestrationTools() {
 // reaches an old DB otherwise. Preserves each tool's agent binding + enabled flag.
 // Bump the flag whenever these tools' schemas/descriptions change in code.
 func (s *Server) refreshBuiltinToolSchemas() {
-	const flag = "tool_schema_refresh_v7_list_facts_paging"
+	const flag = "tool_schema_refresh_v8_finding_case_tools"
 	if v, _, _ := s.m.pg.GetSetting(flag); v == "true" {
 		return
 	}

server/remarks.go

@@ -10,7 +10,7 @@ import (
 	"github.com/Autumn-27/artex/db"
 )
 
-const maxRemarkRunes = 2000
+const maxRemarkRunes = db.MaxTaskRemarkRunes
 
 func normalizeRemark(value string) (string, error) {
 	value = strings.TrimSpace(value)

server/repeater.go

@@ -0,0 +1,416 @@
+package server
+
+import (
+	"bytes"
+	"context"
+	"crypto/tls"
+	"encoding/json"
+	"errors"
+	"fmt"
+	"io"
+	"net"
+	"net/http"
+	"net/url"
+	"os"
+	"strconv"
+	"strings"
+	"syscall"
+	"time"
+)
+
+const (
+	repeaterMaxBody        = 2 << 20 // 2 MiB, for both the JSON request and response preview
+	repeaterDefaultTimeout = 30_000
+	repeaterMinTimeout     = 1_000
+	repeaterMaxTimeout     = 120_000
+	repeaterAllowLocalEnv  = "ARTEX_REPEATER_ALLOW_LOCAL"
+)
+
+var repeaterHopByHop = map[string]struct{}{
+	"Connection": {}, "Keep-Alive": {}, "Proxy-Connection": {}, "Transfer-Encoding": {},
+	"Te": {}, "Trailer": {}, "Upgrade": {}, "Content-Length": {}, "Proxy-Authorization": {},
+}
+
+type repeaterSendRequest struct {
+	Method             *string           `json:"method"`
+	URL                string            `json:"url"`
+	Headers            map[string]string `json:"headers"`
+	Body               string            `json:"body"`
+	TimeoutMs          int               `json:"timeout_ms"`
+	InsecureSkipVerify bool              `json:"insecure_skip_verify"`
+	UseRecordingProxy  *bool             `json:"use_recording_proxy"`
+}
+
+type repeaterSendResponse struct {
+	Status        int                 `json:"status"`
+	StatusText    string              `json:"status_text"`
+	Proto         string              `json:"proto"`
+	Headers       map[string][]string `json:"headers"`
+	Body          string              `json:"body"`
+	BodyTruncated bool                `json:"body_truncated"`
+	BodyBytes     int                 `json:"body_bytes"`
+	ElapsedMs     int64               `json:"elapsed_ms"`
+	URL           string              `json:"url"`
+	Error         string              `json:"error,omitempty"`
+}
+
+func (s *Server) registerRepeater(mux *http.ServeMux) {
+	mux.HandleFunc("POST /api/repeater/send", s.sendRepeater)
+}
+
+func (s *Server) sendRepeater(w http.ResponseWriter, r *http.Request) {
+	var in repeaterSendRequest
+	dec := json.NewDecoder(http.MaxBytesReader(w, r.Body, repeaterMaxBody))
+	if err := dec.Decode(&in); err != nil {
+		writeErr(w, http.StatusBadRequest, "请求体无效")
+		return
+	}
+	// Reject a second JSON value (or oversized trailing data) instead of silently
+	// accepting bytes beyond the documented 2 MiB request limit.
+	var extra any
+	if err := dec.Decode(&extra); err != io.EOF {
+		writeErr(w, http.StatusBadRequest, "请求体无效")
+		return
+	}
+
+	method := http.MethodGet
+	if in.Method != nil {
+		method = strings.ToUpper(strings.TrimSpace(*in.Method))
+	}
+	if method == "" {
+		writeErr(w, http.StatusBadRequest, "HTTP 方法不能为空")
+		return
+	}
+	switch method {
+	case http.MethodGet, http.MethodPost, http.MethodPut, http.MethodPatch,
+		http.MethodDelete, http.MethodHead, http.MethodOptions:
+	default:
+		writeErr(w, http.StatusBadRequest, "不支持的 HTTP 方法")
+		return
+	}
+
+	rawURL := strings.TrimSpace(in.URL)
+	if rawURL == "" {
+		writeErr(w, http.StatusBadRequest, "url 必填")
+		return
+	}
+	u, err := url.Parse(rawURL)
+	if err != nil || u.Host == "" || u.Hostname() == "" ||
+		!strings.EqualFold(u.Scheme, "http") && !strings.EqualFold(u.Scheme, "https") {
+		writeErr(w, http.StatusBadRequest, "仅支持绝对 http/https URL")
+		return
+	}
+	// This check is needed even when a proxy is used: the proxy must remain usable
+	// for normal targets, while obvious literal metadata/loopback URLs stay blocked.
+	if err := repeaterCheckURL(u); err != nil {
+		writeErr(w, http.StatusBadRequest, err.Error())
+		return
+	}
+
+	timeoutMs := in.TimeoutMs
+	if timeoutMs <= 0 {
+		timeoutMs = repeaterDefaultTimeout
+	}
+	if timeoutMs < repeaterMinTimeout {
+		timeoutMs = repeaterMinTimeout
+	}
+	if timeoutMs > repeaterMaxTimeout {
+		timeoutMs = repeaterMaxTimeout
+	}
+	useRecordingProxy := true
+	if in.UseRecordingProxy != nil {
+		useRecordingProxy = *in.UseRecordingProxy
+	}
+
+	var body io.Reader
+	if in.Body != "" && method != http.MethodHead {
+		body = strings.NewReader(in.Body)
+	}
+	ctx, cancel := context.WithTimeout(r.Context(), time.Duration(timeoutMs)*time.Millisecond)
+	defer cancel()
+	req, err := http.NewRequestWithContext(ctx, method, u.String(), body)
+	if err != nil {
+		writeErr(w, http.StatusBadRequest, "构造请求失败")
+		return
+	}
+	for key, value := range in.Headers {
+		name := strings.TrimSpace(key)
+		if name == "" {
+			continue
+		}
+		if _, blocked := repeaterHopByHop[http.CanonicalHeaderKey(name)]; blocked {
+			continue
+		}
+		if strings.EqualFold(name, "Host") {
+			// Host is a special Request field, but retaining it in Header too makes
+			// the editor semantics explicit; net/http uses req.Host on the wire.
+			req.Host = strings.TrimSpace(value)
+			req.Header.Set("Host", strings.TrimSpace(value))
+			continue
+		}
+		req.Header.Set(name, value)
+	}
+
+	client := &http.Client{
+		Timeout:   time.Duration(timeoutMs) * time.Millisecond,
+		Transport: s.repeaterTransport(useRecordingProxy, in.InsecureSkipVerify),
+		CheckRedirect: func(next *http.Request, via []*http.Request) error {
+			if len(via) >= 10 {
+				return errors.New("重定向次数过多")
+			}
+			if err := repeaterCheckURL(next.URL); err != nil {
+				return err
+			}
+			if len(via) > 0 && !strings.EqualFold(next.URL.Host, via[0].URL.Host) {
+				return fmt.Errorf("拒绝跨主机重定向（%s → %s）", via[0].URL.Host, next.URL.Host)
+			}
+			return nil
+		},
+	}
+	// Each send owns its transport. Registered before Body.Close so idle
+	// connections are released after the response body, including on errors.
+	defer client.CloseIdleConnections()
+
+	started := time.Now()
+	resp, err := client.Do(req)
+	if err != nil {
+		writeJSON(w, http.StatusOK, repeaterSendResponse{
+			Headers: map[string][]string{}, URL: u.String(),
+			ElapsedMs: time.Since(started).Milliseconds(), Error: repeaterRedactError(err),
+		})
+		return
+	}
+	defer resp.Body.Close()
+
+	out := repeaterSendResponse{
+		Status: resp.StatusCode, StatusText: resp.Status, Proto: resp.Proto,
+		Headers: resp.Header.Clone(), URL: resp.Request.URL.String(),
+	}
+	if method != http.MethodHead {
+		// Read one byte beyond the cap so truncation is exact. The MVP keeps the
+		// preview as UTF-8 text; invalid bytes are replaced instead of base64.
+		data, readErr := io.ReadAll(io.LimitReader(resp.Body, repeaterMaxBody+1))
+		if len(data) > repeaterMaxBody {
+			out.BodyTruncated = true
+			data = data[:repeaterMaxBody]
+		}
+		out.Body = repeaterBodyText(data)
+		out.BodyBytes = len(data)
+		if readErr != nil && !errors.Is(readErr, io.EOF) {
+			out.Error = "读取响应失败: " + readErr.Error()
+		}
+	}
+	out.ElapsedMs = time.Since(started).Milliseconds()
+	writeJSON(w, http.StatusOK, out)
+}
+
+// repeaterBodyText keeps the response preview as text for the MVP. Invalid
+// UTF-8 and NUL bytes are replaced so a binary response cannot inject control
+// characters into the JSON/UI raw view; the original bytes are not base64
+// encoded and the preview remains bounded by repeaterMaxBody.
+func repeaterBodyText(data []byte) string {
+	text := string(bytes.ToValidUTF8(data, []byte("�")))
+	return strings.ReplaceAll(text, "\x00", "�")
+}
+
+func (s *Server) repeaterTransport(useRecordingProxy, insecure bool) http.RoundTripper {
+	var transport *http.Transport
+	proxied := false
+	if useRecordingProxy && s != nil && s.m != nil {
+		transport = s.httpProxyTransport(httpExec{UseRecordingProxy: true})
+		proxied = transport != nil
+	}
+	if transport == nil && s != nil && s.m != nil && strings.TrimSpace(s.m.GlobalProxy()) != "" {
+		transport = s.httpProxyTransport(httpExec{Proxy: s.m.GlobalProxy()})
+		proxied = transport != nil
+	}
+	if transport == nil {
+		if base, ok := http.DefaultTransport.(*http.Transport); ok {
+			transport = base.Clone()
+		} else {
+			transport = &http.Transport{}
+		}
+		// "Direct" means direct even if the process inherited HTTP_PROXY. The
+		// configured recording/global proxy paths are handled above explicitly.
+		transport.Proxy = nil
+	}
+	transport = transport.Clone()
+	// Keep response bytes faithful and make body_bytes match what was received;
+	// transparent gzip decompression would otherwise obscure both values.
+	transport.DisableCompression = true
+	// Use the request's timeout_ms budget instead of inherited phase deadlines.
+	transport.TLSHandshakeTimeout = 0
+	transport.ExpectContinueTimeout = 0
+	if !proxied {
+		// Clear a custom TLS dialer before installing the guarded dialer; when
+		// DialTLSContext is set, net/http bypasses DialContext entirely.
+		transport.DialTLSContext = nil
+		transport.DialTLS = nil
+		transport.Dial = nil
+		// Let the request context enforce timeout_ms for the entire dial.
+		transport.DialContext = (&net.Dialer{
+			KeepAlive: 30 * time.Second,
+			Control:   repeaterBlockSensitiveDial,
+		}).DialContext
+	}
+	if insecure {
+		cfg := &tls.Config{InsecureSkipVerify: true} //nolint:gosec // explicit user option for pentest targets
+		if transport.TLSClientConfig != nil {
+			cfg = transport.TLSClientConfig.Clone()
+			cfg.InsecureSkipVerify = true
+		}
+		transport.TLSClientConfig = cfg
+	}
+	return transport
+}
+
+func repeaterAllowLocal() bool {
+	v := strings.TrimSpace(os.Getenv(repeaterAllowLocalEnv))
+	return v == "1" || strings.EqualFold(v, "true")
+}
+
+func repeaterCheckURL(u *url.URL) error {
+	if repeaterAllowLocal() {
+		return nil
+	}
+	host := strings.TrimSuffix(strings.TrimSpace(u.Hostname()), ".")
+	// URL.Hostname may retain an IPv6 zone (for example fe80::1%eth0); the
+	// zone is not part of the address classification.
+	if zone := strings.IndexByte(host, '%'); zone >= 0 {
+		host = host[:zone]
+	}
+	ip := net.ParseIP(host)
+	if ip == nil && repeaterLooksNumericHost(host) {
+		var ok bool
+		ip, ok = repeaterParseLegacyIPv4(host)
+		if !ok {
+			return fmt.Errorf("拒绝无法解析的数字主机地址 %q", host)
+		}
+	}
+	if strings.EqualFold(host, "localhost") || (ip != nil && repeaterIsBlockedDialIP(ip)) {
+		return fmt.Errorf("拒绝访问本机/链路本地地址（如确需访问本机服务，设置 %s=1）", repeaterAllowLocalEnv)
+	}
+	return nil
+}
+
+// repeaterLooksNumericHost identifies the legacy numeric spellings accepted by
+// common proxy/resolver implementations (decimal, octal, hexadecimal and
+// dotted inet_aton forms). A numeric-looking value that fails to parse is
+// rejected instead of being handed to a proxy for its own interpretation.
+func repeaterLooksNumericHost(host string) bool {
+	if host == "" || host[0] < '0' || host[0] > '9' {
+		return false
+	}
+	for i := 0; i < len(host); i++ {
+		c := host[i]
+		switch {
+		case c == '.':
+		case c >= '0' && c <= '9':
+		case c >= 'a' && c <= 'f', c >= 'A' && c <= 'F':
+		case c == 'x' || c == 'X':
+		default:
+			return false
+		}
+	}
+	return true
+}
+
+// repeaterParseLegacyIPv4 follows inet_aton's 1-, 2-, 3- and 4-part IPv4
+// notation. Parts with a leading zero are octal and parts with 0x are hex.
+func repeaterParseLegacyIPv4(host string) (net.IP, bool) {
+	parts := strings.Split(host, ".")
+	if len(parts) < 1 || len(parts) > 4 {
+		return nil, false
+	}
+	values := make([]uint64, len(parts))
+	for i, part := range parts {
+		if part == "" {
+			return nil, false
+		}
+		base := 10
+		digits := part
+		if len(digits) > 2 && digits[0] == '0' && (digits[1] == 'x' || digits[1] == 'X') {
+			base, digits = 16, digits[2:]
+		} else if len(digits) > 1 && digits[0] == '0' {
+			base = 8
+		}
+		if digits == "" {
+			return nil, false
+		}
+		value, err := strconv.ParseUint(digits, base, 32)
+		if err != nil {
+			return nil, false
+		}
+		values[i] = value
+	}
+	var value uint64
+	switch len(values) {
+	case 1:
+		value = values[0]
+	case 2:
+		if values[0] > 0xff || values[1] > 0xffffff {
+			return nil, false
+		}
+		value = values[0]<<24 | values[1]
+	case 3:
+		if values[0] > 0xff || values[1] > 0xff || values[2] > 0xffff {
+			return nil, false
+		}
+		value = values[0]<<24 | values[1]<<16 | values[2]
+	case 4:
+		for _, part := range values {
+			if part > 0xff {
+				return nil, false
+			}
+		}
+		value = values[0]<<24 | values[1]<<16 | values[2]<<8 | values[3]
+	}
+	if value > 0xffffffff {
+		return nil, false
+	}
+	return net.IPv4(byte(value>>24), byte(value>>16), byte(value>>8), byte(value)), true
+}
+
+func repeaterIsBlockedDialIP(ip net.IP) bool {
+	if ip == nil {
+		return true
+	}
+	if v4 := ip.To4(); v4 != nil {
+		ip = v4
+	}
+	return ip.IsLoopback() || ip.IsLinkLocalUnicast() || ip.IsLinkLocalMulticast() ||
+		ip.IsInterfaceLocalMulticast() || ip.IsUnspecified() || ip.IsMulticast()
+}
+
+func repeaterBlockSensitiveDial(_, address string, _ syscall.RawConn) error {
+	if repeaterAllowLocal() {
+		return nil
+	}
+	host, _, err := net.SplitHostPort(address)
+	if err != nil {
+		return err
+	}
+	ip := net.ParseIP(host)
+	if ip == nil {
+		return fmt.Errorf("无法解析目标地址 %q", host)
+	}
+	if repeaterIsBlockedDialIP(ip) {
+		return fmt.Errorf("拒绝访问本机/链路本地地址 %s（如确需访问本机服务，设置 %s=1）", ip, repeaterAllowLocalEnv)
+	}
+	return nil
+}
+
+func repeaterRedactError(err error) string {
+	var urlErr *url.Error
+	if errors.As(err, &urlErr) {
+		host := ""
+		if parsed, parseErr := url.Parse(urlErr.URL); parseErr == nil {
+			host = parsed.Host
+		}
+		if urlErr.Err != nil {
+			return fmt.Sprintf("%s %s: %s", urlErr.Op, host, urlErr.Err)
+		}
+		return fmt.Sprintf("%s %s: 未知错误", urlErr.Op, host)
+	}
+	return err.Error()
+}

server/repeater_test.go

@@ -0,0 +1,404 @@
+package server
+
+import (
+	"bytes"
+	"context"
+	"crypto/tls"
+	"crypto/x509"
+	"encoding/json"
+	"errors"
+	"fmt"
+	"io"
+	"net"
+	"net/http"
+	"net/http/httptest"
+	"net/url"
+	"os"
+	"strings"
+	"testing"
+	"time"
+
+	"github.com/Autumn-27/artex/traffic"
+)
+
+func repeaterTestSend(t *testing.T, s *Server, payload any) (int, repeaterSendResponse, string) {
+	t.Helper()
+	raw, err := json.Marshal(payload)
+	if err != nil {
+		t.Fatal(err)
+	}
+	r := httptest.NewRequest(http.MethodPost, "/api/repeater/send", bytes.NewReader(raw))
+	w := httptest.NewRecorder()
+	s.sendRepeater(w, r)
+	var out repeaterSendResponse
+	_ = json.Unmarshal(w.Body.Bytes(), &out)
+	return w.Code, out, w.Body.String()
+}
+
+func TestRepeaterRejectsInvalidInput(t *testing.T) {
+	t.Setenv(repeaterAllowLocalEnv, "")
+	s := &Server{}
+	for _, tc := range []struct {
+		name string
+		body string
+	}{
+		{"missing url", `{"method":"GET"}`},
+		{"empty method", `{"method":"  ","url":"https://example.test/"}`},
+		{"bad scheme", `{"method":"GET","url":"file:///etc/passwd"}`},
+		{"gopher", `{"method":"GET","url":"gopher://example.test/1"}`},
+		{"unknown method", `{"method":"TRACE","url":"http://example.test/"}`},
+	} {
+		t.Run(tc.name, func(t *testing.T) {
+			r := httptest.NewRequest(http.MethodPost, "/api/repeater/send", strings.NewReader(tc.body))
+			w := httptest.NewRecorder()
+			s.sendRepeater(w, r)
+			if w.Code != http.StatusBadRequest {
+				t.Fatalf("status=%d want 400 body=%s", w.Code, w.Body.String())
+			}
+		})
+	}
+}
+
+func TestRepeaterBlockedIPs(t *testing.T) {
+	for _, raw := range []string{"127.0.0.1", "::1", "169.254.169.254", "0.0.0.0", "224.0.0.1"} {
+		if !repeaterIsBlockedDialIP(net.ParseIP(raw)) {
+			t.Errorf("expected %s blocked", raw)
+		}
+	}
+	for _, raw := range []string{"8.8.8.8", "10.0.0.1", "192.168.1.20", "172.16.0.5"} {
+		if repeaterIsBlockedDialIP(net.ParseIP(raw)) {
+			t.Errorf("expected %s allowed", raw)
+		}
+	}
+
+	t.Setenv(repeaterAllowLocalEnv, "")
+	srv := httptest.NewServer(http.HandlerFunc(func(http.ResponseWriter, *http.Request) {
+		t.Fatal("loopback handler should not be reached")
+	}))
+	defer srv.Close()
+	code, _, body := repeaterTestSend(t, &Server{}, map[string]any{"url": srv.URL, "use_recording_proxy": false})
+	if code != http.StatusBadRequest || !strings.Contains(body, "拒绝") {
+		t.Fatalf("loopback status=%d body=%s", code, body)
+	}
+}
+
+func TestRepeaterRejectsLegacyNumericHostsIncludingProxyPath(t *testing.T) {
+	t.Setenv(repeaterAllowLocalEnv, "")
+	for _, raw := range []string{
+		"http://2130706433/", // decimal 127.0.0.1
+		"http://0177.0.0.1/", // octal 127.0.0.1
+		"http://127.1/",      // two-part inet_aton form
+		"http://0x7f000001/", // hexadecimal 127.0.0.1
+		"http://9999999999/", // numeric-looking but invalid
+	} {
+		u, err := url.Parse(raw)
+		if err != nil {
+			t.Fatal(err)
+		}
+		if err := repeaterCheckURL(u); err == nil {
+			t.Errorf("repeaterCheckURL(%q) allowed a legacy numeric host", raw)
+		}
+	}
+
+	// The check must run before a configured proxy gets a chance to interpret
+	// the legacy spelling and connect to loopback remotely.
+	proxy := httptest.NewServer(http.HandlerFunc(func(http.ResponseWriter, *http.Request) {
+		t.Error("proxy received a blocked numeric target")
+	}))
+	defer proxy.Close()
+	s := &Server{m: &Manager{globalProxy: proxy.URL}}
+	code, _, body := repeaterTestSend(t, s, map[string]any{"url": "http://2130706433/"})
+	if code != http.StatusBadRequest || !strings.Contains(body, "拒绝") {
+		t.Fatalf("proxied legacy host status=%d body=%s", code, body)
+	}
+}
+
+func TestRepeaterBlockSensitiveDial(t *testing.T) {
+	t.Setenv(repeaterAllowLocalEnv, "")
+	for _, address := range []string{"127.0.0.1:80", "169.254.169.254:80"} {
+		if err := repeaterBlockSensitiveDial("tcp", address, nil); err == nil {
+			t.Errorf("repeaterBlockSensitiveDial(%q) allowed blocked address", address)
+		}
+	}
+}
+
+func TestRepeaterTransportBlocksSensitiveDial(t *testing.T) {
+	t.Setenv(repeaterAllowLocalEnv, "")
+	tr := (&Server{}).repeaterTransport(false, false).(*http.Transport)
+	defer tr.CloseIdleConnections()
+	if tr.DialContext == nil {
+		t.Fatal("direct transport has no guarded DialContext")
+	}
+	ctx, cancel := context.WithTimeout(context.Background(), time.Second)
+	defer cancel()
+	conn, err := tr.DialContext(ctx, "tcp", "127.0.0.1:1")
+	if conn != nil {
+		conn.Close()
+	}
+	if err == nil || !strings.Contains(err.Error(), "拒绝访问本机/链路本地地址 127.0.0.1") {
+		t.Fatalf("dial should fail at the sensitive-address guard, got %v", err)
+	}
+}
+
+func TestRepeaterDirectHTTP2(t *testing.T) {
+	t.Setenv(repeaterAllowLocalEnv, "1")
+	srv := httptest.NewUnstartedServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		if r.ProtoMajor != 2 {
+			t.Errorf("request protocol=%s, want HTTP/2", r.Proto)
+		}
+		_, _ = io.WriteString(w, "h2 response")
+	}))
+	srv.EnableHTTP2 = true
+	srv.TLS = &tls.Config{NextProtos: []string{"h2"}}
+	srv.StartTLS()
+	defer srv.Close()
+
+	code, out, raw := repeaterTestSend(t, &Server{}, map[string]any{
+		"url": srv.URL, "use_recording_proxy": false, "insecure_skip_verify": true,
+	})
+	if code != http.StatusOK || out.Error != "" || out.Status != http.StatusOK || out.Proto != "HTTP/2.0" || out.Body != "h2 response" {
+		t.Fatalf("HTTP/2 response status=%d raw=%s", code, raw)
+	}
+}
+
+func TestRepeaterTLSHandshakeUsesRequestTimeout(t *testing.T) {
+	t.Setenv(repeaterAllowLocalEnv, "1")
+	// Scale down the inherited timeout so a slow handshake can be tested
+	// without waiting for the default transport's ten-second deadline.
+	base := http.DefaultTransport
+	short := base.(*http.Transport).Clone()
+	short.TLSHandshakeTimeout = 10 * time.Millisecond
+	http.DefaultTransport = short
+	defer func() { http.DefaultTransport = base }()
+	defer short.CloseIdleConnections()
+
+	srv := httptest.NewUnstartedServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		_, _ = io.WriteString(w, "handshake completed")
+	}))
+	srv.TLS = &tls.Config{GetConfigForClient: func(*tls.ClientHelloInfo) (*tls.Config, error) {
+		time.Sleep(100 * time.Millisecond)
+		return nil, nil
+	}}
+	srv.StartTLS()
+	defer srv.Close()
+
+	code, out, raw := repeaterTestSend(t, &Server{}, map[string]any{
+		"url": srv.URL, "use_recording_proxy": false, "insecure_skip_verify": true, "timeout_ms": 1000,
+	})
+	if code != http.StatusOK || out.Error != "" || out.Status != http.StatusOK || out.Body != "handshake completed" {
+		t.Fatalf("handshake within request budget failed: status=%d raw=%s", code, raw)
+	}
+}
+
+func TestRepeaterClosesIdleConnections(t *testing.T) {
+	t.Setenv(repeaterAllowLocalEnv, "1")
+	for _, mode := range []string{"direct", "http2", "global proxy", "recording proxy fallback"} {
+		t.Run(mode, func(t *testing.T) {
+			closed := make(chan struct{}, 1)
+			srv := httptest.NewUnstartedServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+				_, _ = io.WriteString(w, "response")
+			}))
+			srv.Config.ConnState = func(_ net.Conn, state http.ConnState) {
+				if state == http.StateClosed {
+					closed <- struct{}{}
+				}
+			}
+			if mode == "http2" {
+				srv.EnableHTTP2 = true
+				srv.StartTLS()
+			} else {
+				srv.Start()
+			}
+			defer srv.Close()
+
+			s := &Server{}
+			payload := map[string]any{"url": srv.URL, "use_recording_proxy": false, "insecure_skip_verify": true}
+			if mode == "global proxy" || mode == "recording proxy fallback" {
+				s.m = &Manager{globalProxy: srv.URL}
+				payload["url"] = "http://target.example/"
+				if mode == "recording proxy fallback" {
+					delete(payload, "use_recording_proxy")
+				}
+			}
+			code, out, raw := repeaterTestSend(t, s, payload)
+			if code != http.StatusOK || out.Error != "" || out.Status != http.StatusOK || out.Body != "response" {
+				t.Fatalf("status=%d raw=%s", code, raw)
+			}
+			select {
+			case <-closed:
+			case <-time.After(time.Second):
+				t.Fatal("connection still open after sendRepeater returned")
+			}
+		})
+	}
+}
+
+func TestRepeaterRoundTripHeadersAndBody(t *testing.T) {
+	t.Setenv(repeaterAllowLocalEnv, "1")
+	var got http.Request
+	var gotBody string
+	srv := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		got = *r
+		b, _ := io.ReadAll(r.Body)
+		gotBody = string(b)
+		w.Header().Set("X-Echo", "ok")
+		w.WriteHeader(http.StatusCreated)
+		_, _ = io.WriteString(w, "hello")
+	}))
+	defer srv.Close()
+
+	code, out, body := repeaterTestSend(t, &Server{}, map[string]any{
+		"method": "POST", "url": srv.URL, "headers": map[string]string{
+			"X-Test": "yes", "Host": "custom.example", "Connection": "keep-alive",
+			"Keep-Alive": "timeout=5", "Transfer-Encoding": "chunked", "Content-Length": "99",
+		}, "body": "payload", "use_recording_proxy": false,
+	})
+	if code != http.StatusOK || out.Error != "" {
+		t.Fatalf("status=%d out=%+v raw=%s", code, out, body)
+	}
+	if out.Status != http.StatusCreated || out.Body != "hello" || out.BodyBytes != 5 {
+		t.Fatalf("response=%+v", out)
+	}
+	if out.Headers["X-Echo"][0] != "ok" || got.Method != http.MethodPost || gotBody != "payload" || got.Host != "custom.example" {
+		t.Fatalf("request method=%s host=%s body=%q headers=%v", got.Method, got.Host, gotBody, got.Header)
+	}
+	for _, name := range []string{"Connection", "Keep-Alive", "Transfer-Encoding"} {
+		if got.Header.Get(name) != "" {
+			t.Errorf("hop-by-hop header %s was forwarded: %q", name, got.Header.Get(name))
+		}
+	}
+	if got.Header.Get("Content-Length") != "7" {
+		t.Errorf("content length should match body, got %q", got.Header.Get("Content-Length"))
+	}
+}
+
+func TestRepeaterTruncatesAndHonoursHEAD(t *testing.T) {
+	t.Setenv(repeaterAllowLocalEnv, "1")
+	large := strings.Repeat("x", repeaterMaxBody+1)
+	srv := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		if r.Method == http.MethodHead {
+			_, _ = io.WriteString(w, "should-not-be-read")
+			return
+		}
+		_, _ = io.WriteString(w, large)
+	}))
+	defer srv.Close()
+	_, out, _ := repeaterTestSend(t, &Server{}, map[string]any{"url": srv.URL, "use_recording_proxy": false})
+	if !out.BodyTruncated || out.BodyBytes != repeaterMaxBody || len(out.Body) != repeaterMaxBody {
+		t.Fatalf("truncation=%+v", out)
+	}
+	_, head, _ := repeaterTestSend(t, &Server{}, map[string]any{"method": "HEAD", "url": srv.URL, "use_recording_proxy": false})
+	if head.Body != "" || head.BodyBytes != 0 {
+		t.Fatalf("HEAD body=%q bytes=%d", head.Body, head.BodyBytes)
+	}
+}
+
+func TestRepeaterBodyTextReplacesBinaryControls(t *testing.T) {
+	if got := repeaterBodyText([]byte{'a', 0, 0xff, 'b'}); got != "a��b" {
+		t.Fatalf("body text=%q", got)
+	}
+}
+
+func TestRepeaterRejectsCrossHostRedirect(t *testing.T) {
+	t.Setenv(repeaterAllowLocalEnv, "1")
+	target := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) { _, _ = io.WriteString(w, "target") }))
+	defer target.Close()
+	redirect := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		http.Redirect(w, r, target.URL, http.StatusFound)
+	}))
+	defer redirect.Close()
+	_, out, _ := repeaterTestSend(t, &Server{}, map[string]any{"url": redirect.URL, "use_recording_proxy": false})
+	if out.Status != 0 || !strings.Contains(out.Error, "跨主机") {
+		t.Fatalf("redirect response=%+v", out)
+	}
+}
+
+func TestRepeaterProxyTransportConstruction(t *testing.T) {
+	s := &Server{m: &Manager{}}
+	for _, scheme := range []string{"http", "https", "socks5", "socks5h"} {
+		tr := s.httpProxyTransport(httpExec{Proxy: scheme + "://user:[email protected]:1080"})
+		if tr == nil || tr.Proxy == nil {
+			t.Fatalf("%s proxy transport not configured: %#v", scheme, tr)
+		}
+		proxyURL, err := tr.Proxy(httptest.NewRequest(http.MethodGet, "https://target.example/", nil))
+		if err != nil || proxyURL == nil || proxyURL.Scheme != scheme || proxyURL.Host != "127.0.0.1:1080" {
+			t.Fatalf("%s proxy URL = %#v, err=%v", scheme, proxyURL, err)
+		}
+		if user := proxyURL.User; user == nil || user.Username() != "user" {
+			t.Fatalf("%s proxy credentials missing: %v", scheme, proxyURL.User)
+		}
+		password, ok := proxyURL.User.Password()
+		if !ok || password != "pass" {
+			t.Fatalf("%s proxy password missing: %q (ok=%v)", scheme, password, ok)
+		}
+	}
+
+	// With capture disabled, the recording-proxy option must preserve the
+	// configured global proxy URL instead of treating it as a host:port.
+	s.m.globalProxy = "socks5://user:[email protected]:1080"
+	tr := s.httpProxyTransport(httpExec{UseRecordingProxy: true})
+	if tr == nil || tr.Proxy == nil {
+		t.Fatal("global SOCKS5 proxy was not configured")
+	}
+	proxyURL, err := tr.Proxy(httptest.NewRequest(http.MethodGet, "https://target.example/", nil))
+	if err != nil || proxyURL == nil || proxyURL.Scheme != "socks5" {
+		t.Fatalf("global proxy URL = %#v, err=%v", proxyURL, err)
+	}
+
+	// Capture mode supplies a complete HTTP URL from Traffic.ProxyAddr. The
+	// helper must preserve that scheme and host rather than prefixing http:// a
+	// second time.
+	captured, err := traffic.Open(t.TempDir(), "127.0.0.1:18788")
+	if err != nil {
+		t.Fatalf("traffic.Open: %v", err)
+	}
+	defer captured.Close()
+	s.m.traffic = captured
+	s.m.trafficOn = true
+	tr = s.httpProxyTransport(httpExec{UseRecordingProxy: true})
+	if tr == nil || tr.Proxy == nil {
+		t.Fatal("recording proxy was not configured")
+	}
+	proxyURL, err = tr.Proxy(httptest.NewRequest(http.MethodGet, "https://target.example/", nil))
+	if err != nil || proxyURL == nil || proxyURL.Scheme != "http" || proxyURL.Host != "127.0.0.1:18788" {
+		t.Fatalf("recording proxy URL = %#v, err=%v", proxyURL, err)
+	}
+	caFile := s.m.ProxyCACert()
+	if caFile == "" {
+		t.Fatal("recording proxy CA path is empty")
+	}
+	caPEM, err := os.ReadFile(caFile)
+	if err != nil {
+		t.Fatalf("read recording proxy CA: %v", err)
+	}
+	wantRoots := x509.NewCertPool()
+	if !wantRoots.AppendCertsFromPEM(caPEM) {
+		t.Fatal("recording proxy CA is not valid PEM")
+	}
+	if tr.TLSClientConfig == nil || tr.TLSClientConfig.RootCAs == nil || !tr.TLSClientConfig.RootCAs.Equal(wantRoots) {
+		t.Fatal("recording proxy CA was not loaded into transport RootCAs")
+	}
+	if s.repeaterTransport(true, false).(*http.Transport).ForceAttemptHTTP2 {
+		t.Fatal("recording proxy transport must not force HTTP/2")
+	}
+}
+
+func TestRepeaterRedactProxyCredentials(t *testing.T) {
+	for _, scheme := range []string{"http", "https", "socks5", "socks5h"} {
+		t.Run(scheme, func(t *testing.T) {
+			proxyErr := &url.Error{
+				Op: "proxyconnect", URL: scheme + "://proxy-user:[email protected]:1080",
+				Err: errors.New("connection refused"),
+			}
+			for _, err := range []error{proxyErr, fmt.Errorf("send request: %w", proxyErr)} {
+				got := repeaterRedactError(err)
+				if strings.Contains(got, "proxy-secret") || strings.Contains(got, "proxy-user") {
+					t.Fatalf("proxy credentials leaked: %q", got)
+				}
+				if got != "proxyconnect 127.0.0.1:1080: connection refused" {
+					t.Fatalf("unexpected redacted error: %q", got)
+				}
+			}
+		})
+	}
+}

server/server.go

@@ -177,6 +177,9 @@ func New(ctx context.Context, m *Manager, skillDir string, dataDir string, keyDi
 		b := s.agentsForTask(t)
 		return b.pl, b.wk
 	})
+	s.engine.SetWorkerResolver(func(t *Task, slot WorkerSlot) (*agent.Worker, error) {
+		return s.workerForTaskSlot(t, slot)
+	})
 	// Global readiness (Ready()/llm_configured): a global active LLM provider is
 	// installed. Task-level runnability is separate (ReadyFor → the resolver above).
 	s.engine.SetReadiness(func() bool {
@@ -575,7 +578,8 @@ func (s *Server) chatUnavailableReason() string {
 
 // providerForProfile returns a cached provider+cfg for a profile id, so every agent
 // bound/pinned to the same profile shares one provider instance (one rate limiter).
-// ok=false when the profile is missing/invalid → caller falls back to the global pair.
+// ok=false when the profile is missing/invalid; pinned callers surface that error
+// while legacy callers retain their existing fallback behavior.
 func (s *Server) providerForProfile(id int64) (llm.Provider, agent.Config, bool) {
 	s.provCacheMu.Lock()
 	if e := s.provByProfile[id]; e != nil {
@@ -824,6 +828,7 @@ func (s *Server) Handler() http.Handler {
 	mux.HandleFunc("GET /api/exploration/findings/stats", s.findingStats)
 	mux.HandleFunc("GET /api/exploration/findings/export", s.findingsExport)
 	s.registerFindingTraffic(mux)
+	s.registerRepeater(mux)
 	mux.HandleFunc("GET /api/exploration/findings/{id}", s.getFinding)
 	mux.HandleFunc("GET /api/exploration/findings/{id}/lineage", s.findingLineage)
 	mux.HandleFunc("POST /api/exploration/findings/{id}/deepen", s.deepenFinding)
@@ -1540,6 +1545,8 @@ type createTaskReq struct {
 	PlanHeartbeatSeconds int              `json:"plan_heartbeat_seconds"`      // planner 心跳触发间隔(秒);0/省略=默认600(10min);下限=默认=600,低于自动抬到600
 	SeedFirstIntent      *bool            `json:"seed_first_intent,omitempty"` // 创建时直接下发一条种子意图(内容=描述+目标),让 worker 免等首轮 planner 直接开跑;省略/null=默认关闭,走标准先规划再执行。显式传 true 才开(CTF 常一 work 解决时可省掉开跑前的 planner 轮)。
 	CoverageEnabled      *bool            `json:"coverage_enabled,omitempty"`  // 资产覆盖度功能;省略/null=默认开(true)。false=关闭覆盖度计算/展示/自动累积范围+隐藏 add_task_scope/list_untested_assets。company 关联不受影响。
+	Remark               string           `json:"remark,omitempty"`            // 任务备注(最多2000字);空=无
+	WorkerCount          int              `json:"worker_count,omitempty"`      // 本任务 Worker 数;0/省略=用全局默认;上限50
 	// InterceptRules 任务级资产拦截/允许规则(创建时录入,存 task_intercept_rules,不进全局表)。
 	InterceptRules []taskInterceptRuleReq `json:"intercept_rules,omitempty"`
 }
@@ -1603,12 +1610,27 @@ func (s *Server) createTask(w http.ResponseWriter, r *http.Request) {
 		}
 		schedule = &input
 	}
+	remark, err := normalizeRemark(req.Remark)
+	if err != nil {
+		writeErr(w, 400, err.Error())
+		return
+	}
+	if req.WorkerCount < 0 {
+		writeErr(w, 400, "Worker 运行数量不能为负数")
+		return
+	}
+	if req.WorkerCount > db.MaxTaskWorkerCount {
+		writeErr(w, 400, fmt.Sprintf("Worker 运行数量最多 %d", db.MaxTaskWorkerCount))
+		return
+	}
 	t, err := s.m.CreateTaskWithOptions(req.Description, req.Goal, db.TaskCreateOptions{
 		Schedule: schedule,
 		Name:     strings.TrimSpace(req.Name), CategoryID: req.CategoryID,
 		SourceTaskIDs: sourceIDs, CompanyIDs: req.CompanyIDs, LLMProfileIDs: req.LLMProfileIDs,
 		TimeoutSeconds: req.TimeoutSeconds, PlanHeartbeatSeconds: req.PlanHeartbeatSeconds,
 		CoverageEnabled: req.CoverageEnabled,
+		Remark:          remark,
+		WorkerCount:     req.WorkerCount,
 		InterceptRules:  interceptRules,
 	})
 	if err != nil {
@@ -3432,6 +3454,7 @@ func (s *Server) settingsPayload() map[string]any {
 		"global_proxy":              s.m.GlobalProxy(),           // 全局出口代理(http/https/socks5)，所有目标流量走它，空=直连
 		"python_interpreter":        strings.TrimSpace(pyStored), // 用户/自动设的值(空=用运行时检测)
 		"workers":                   s.m.Workers(),               // 并发工作 agent 数(默认3)；对之后启动的任务生效
+		"worker_pools":              s.m.WorkerPools(),           // 非空时按池顺序绑定 worker 会话
 		"task_concurrency_enabled":  concOn,                      // 任务并发上限开关(默认关)
 		"task_concurrency_limit":    concLimit,                   // 同时运行任务上限(开启后默认5)
 		// LLM 轮询(故障转移)。默认关；开启后走全局激活配置的 agent 在当前配置不可用时
@@ -3499,14 +3522,15 @@ func (s *Server) putSettings(w http.ResponseWriter, r *http.Request) {
 		LLMRecord           *bool `json:"llm_record"` // LLM 录制开关（默认关）；即时生效，无需重建 agent
 		// Web search. WebSearchEnabled/Backend toggle the tool + backend; BraveKey/TavilyKey
 		// are optional — omit (null) to leave a stored key untouched, send "" to clear.
-		WebSearchEnabled *bool   `json:"web_search_enabled"`
-		WebSearchBackend *string `json:"web_search_backend"`
-		BraveKey         *string `json:"brave_search_api_key"`
-		TavilyKey        *string `json:"tavily_search_api_key"`
-		WebSearchProxy   *string `json:"web_search_proxy"`   // 独立出口代理(http/https/socks5)；null=不改，""=清空
-		GlobalProxy      *string `json:"global_proxy"`       // 全局出口代理(http/https/socks5)；null=不改，""=清空(直连)
-		PythonInterp     *string `json:"python_interpreter"` // 自定义脚本工具的 python 解释器路径
-		Workers          *int    `json:"workers"`            // 并发工作 agent 数(>0)；对之后启动的任务生效
+		WebSearchEnabled *bool         `json:"web_search_enabled"`
+		WebSearchBackend *string       `json:"web_search_backend"`
+		BraveKey         *string       `json:"brave_search_api_key"`
+		TavilyKey        *string       `json:"tavily_search_api_key"`
+		WebSearchProxy   *string       `json:"web_search_proxy"`   // 独立出口代理(http/https/socks5)；null=不改，""=清空
+		GlobalProxy      *string       `json:"global_proxy"`       // 全局出口代理(http/https/socks5)；null=不改，""=清空(直连)
+		PythonInterp     *string       `json:"python_interpreter"` // 自定义脚本工具的 python 解释器路径
+		Workers          *int          `json:"workers"`            // 并发工作 agent 数(>0)；对之后启动的任务生效
+		WorkerPools      *[]WorkerPool `json:"worker_pools"`
 		// 任务并发上限:同时「运行中」的任务数上限。关闭=不限;开启后新建任务超限则排队,有空位自动启动。
 		ConcurrencyEnabled *bool `json:"task_concurrency_enabled"`
 		ConcurrencyLimit   *int  `json:"task_concurrency_limit"`
@@ -3579,11 +3603,32 @@ func (s *Server) putSettings(w http.ResponseWriter, r *http.Request) {
 			return
 		}
 	}
-	if req.Workers != nil {
-		if err := s.m.SetWorkers(*req.Workers); err != nil {
+	// A non-empty pool list takes precedence when both fields are present and
+	// synchronizes workers to its total. An empty list clears pools; when workers
+	// is also present it sets the legacy count for the newly restored legacy mode.
+	if req.WorkerPools != nil {
+		if len(*req.WorkerPools) == 0 && req.Workers != nil && *req.Workers <= 0 {
+			writeErr(w, 400, "workers 必须 >0")
+			return
+		}
+		legacyWorkers := []int(nil)
+		if req.Workers != nil {
+			legacyWorkers = []int{*req.Workers}
+		}
+		if err := s.m.SetWorkerPools(*req.WorkerPools, legacyWorkers...); err != nil {
 			writeErr(w, 400, err.Error())
 			return
 		}
+	} else if req.Workers != nil {
+		// An omitted worker_pools field leaves the saved mode unchanged. If a
+		// non-empty pool configuration is already saved, its total remains the
+		// authoritative concurrency and workers is deliberately ignored.
+		if len(s.m.WorkerPools()) == 0 {
+			if err := s.m.SetWorkers(*req.Workers); err != nil {
+				writeErr(w, 400, err.Error())
+				return
+			}
+		}
 	}
 	if req.ConcurrencyEnabled != nil || req.ConcurrencyLimit != nil {
 		// 部分 PUT:未给的字段用当前值兜底,避免只改一个把另一个重置。

server/task_llm.go

@@ -27,6 +27,7 @@ type taskAgentBundle struct {
 	pl             *agent.Planner
 	wk             *agent.Worker
 	main           *agent.MainAgent
+	pinnedWorkers  map[int64]*agent.Worker
 }
 
 type llmAuditProfile struct {
@@ -48,9 +49,10 @@ type llmActivityMetadata struct {
 }
 
 type taskLLMRuntime struct {
-	s        *Server
-	taskID   string
-	agentKey string
+	s            *Server
+	taskID       string
+	agentKey     string
+	pinProfileID *int64
 }
 
 type taskLLMError struct {
@@ -101,13 +103,24 @@ type taskLLMStreamHooks struct {
 	transition func(taskLLMSelection, db.TaskLLMTransition, error)
 }
 
-// current resolves the provider this role runs on, by precedence:
-// Agent 绑定 → 任务 LLM 配置链 → 全局/环境配置。
+// current resolves the provider this role runs on. Pinned workers use only their
+// assigned profile; legacy task agents use Agent 绑定 → 任务 LLM 配置链 → 全局/环境配置.
 // 绑定优先于任务链：某个角色被显式指定了模型，就一直跑在那个模型上；绑定不存在或
 // 构建失败时才降级到任务链，任务链为空时再降级到全局配置。
 // 返回的 profile id 只有走任务链时才非零 —— streamTaskLLM 以此判断额度错误是否
 // 应该推进任务的故障转移状态（绑定/全局路径不改任务链状态，沿用既有语义）。
 func (r *taskLLMRuntime) current() (taskLLMSelection, error) {
+	if r.pinProfileID != nil {
+		// Pinned worker runtimes are exclusive: resolve their profile before
+		// touching bindings or the task failover state. A missing or invalid
+		// profile is an error, never an invitation to use another model.
+		prov, cfg, ok := r.s.providerForProfile(*r.pinProfileID)
+		if !ok {
+			return taskLLMSelection{}, fmt.Errorf("LLM profile #%d is missing or invalid", *r.pinProfileID)
+		}
+		t, _ := r.s.m.Task(r.taskID)
+		return taskLLMSelection{task: t, provider: prov, retry: cfg.Retry}, nil
+	}
 	taskNum, err := parseTaskID(r.taskID)
 	if err != nil {
 		return taskLLMSelection{}, err
@@ -152,6 +165,10 @@ func (r *taskLLMRuntime) current() (taskLLMSelection, error) {
 // the caller. If failover switches profiles, the change takes effect on the next
 // agent run (a fresh Session is built per run in captureRun).
 func (r *taskLLMRuntime) activeCfg() (agent.Config, bool) {
+	if r.pinProfileID != nil {
+		_, cfg, ok := r.s.providerForProfile(*r.pinProfileID)
+		return cfg, ok
+	}
 	taskNum, err := parseTaskID(r.taskID)
 	if err != nil {
 		return agent.Config{}, false
@@ -454,6 +471,12 @@ func isRetryableStreamError(err error) bool {
 // CompactionWindow mirrors current()'s precedence so the context window always
 // matches the provider the role will actually stream on.
 func (r *taskLLMRuntime) CompactionWindow() int {
+	if r.pinProfileID != nil {
+		if _, cfg, ok := r.s.providerForProfile(*r.pinProfileID); ok {
+			return cfg.CompactionWindow()
+		}
+		return (agent.Config{}).CompactionWindow()
+	}
 	if _, cfg, ok := r.s.agentBindingProvider(r.agentKey); ok {
 		return cfg.CompactionWindow()
 	}
@@ -597,12 +620,54 @@ func (s *Server) agentsForTask(t *Task) *taskAgentBundle {
 	main.SetSteerWork(s.engine.SteerWork) // steer_work：人对运行中 work 实时纠偏
 	bundle := &taskAgentBundle{
 		runtime: goalRuntime, plannerRuntime: plannerRuntime, workerRuntime: workerRuntime,
-		mainRuntime: mainRuntime, pl: pl, wk: wk, main: main,
+		mainRuntime: mainRuntime, pl: pl, wk: wk, main: main, pinnedWorkers: map[int64]*agent.Worker{},
 	}
 	s.taskAgents[t.ID] = bundle
 	return bundle
 }
 
+func (s *Server) workerForTaskSlot(t *Task, slot WorkerSlot) (*agent.Worker, error) {
+	if slot.ProfileID <= 0 {
+		return nil, errors.New("worker slot is not pinned")
+	}
+	_, cfg, ok := s.providerForProfile(slot.ProfileID)
+	if !ok {
+		return nil, fmt.Errorf("LLM profile #%d is missing or invalid", slot.ProfileID)
+	}
+	s.taskAgentMu.Lock()
+	bundle := s.taskAgents[t.ID]
+	s.taskAgentMu.Unlock()
+	if bundle == nil {
+		bundle = s.agentsForTask(t)
+	}
+	s.taskAgentMu.Lock()
+	defer s.taskAgentMu.Unlock()
+	if current := s.taskAgents[t.ID]; current != nil {
+		bundle = current
+	}
+	if bundle.pinnedWorkers == nil {
+		bundle.pinnedWorkers = make(map[int64]*agent.Worker)
+	}
+	if worker := bundle.pinnedWorkers[slot.ProfileID]; worker != nil {
+		return worker, nil
+	}
+	id := slot.ProfileID
+	runtime := &taskLLMRuntime{s: s, taskID: t.ID, agentKey: "worker", pinProfileID: &id}
+	tx := transcript.NewStore(filepath.Join(s.m.dir, "transcripts"))
+	worker := agent.NewWorker(runtime, cfg.Model, s.m.dir, tx, runtime.CompactionWindow(), s.agentMaxTurns("worker"))
+	worker.SetFindingRecorder(s.evidenceStore())
+	worker.SetCompactionWindowResolver(runtime.CompactionWindow)
+	worker.SetNonStreaming(runtime.nonStreaming)
+	worker.SetMaxTokens(runtime.maxTokens)
+	worker.SetNoaEnabled(s.m.NoaCompactionEnabled)
+	worker.SetRunTimeout(time.Duration(s.agentRunSeconds("worker")) * time.Second)
+	worker.SetProxy(s.m.ProxyAddr(), s.m.ProxyCACert())
+	worker.SetWebSearch(s.webSearchFor("worker"))
+	worker.SetConstraintInject(s.constraintInjectWorker)
+	bundle.pinnedWorkers[id] = worker
+	return worker, nil
+}
+
 func (s *Server) syncTaskLLMState(pt *db.Task) {
 	if pt == nil {
 		return

server/task_templates.go

@@ -22,8 +22,10 @@ type taskTemplateRequest struct {
 	Name           *string                `json:"name"`
 	Description    *string                `json:"description"`
 	Goal           *string                `json:"goal"`
+	Remark         *string                `json:"remark"`
 	CategoryID     *int64                 `json:"category_id"`
 	InterceptRules []taskInterceptRuleReq `json:"intercept_rules"`
+	WorkerCount    *int                   `json:"worker_count"`
 }
 
 // decodeTaskTemplateRequest decodes the body into req and returns the set of
@@ -65,6 +67,7 @@ func validateTaskTemplateRequest(req taskTemplateRequest) error {
 		{name: "name", value: req.Name, limit: db.MaxTaskTemplateNameRunes},
 		{name: "description", value: req.Description, limit: db.MaxTaskTemplateTextRunes},
 		{name: "goal", value: req.Goal, limit: db.MaxTaskTemplateTextRunes},
+		{name: "remark", value: req.Remark, limit: db.MaxTaskRemarkRunes}, // 与任务备注 normalizeRemark 上限一致
 	}
 	for _, check := range checks {
 		if check.value == nil {
@@ -75,9 +78,20 @@ func validateTaskTemplateRequest(req taskTemplateRequest) error {
 			value = strings.Join(strings.Fields(value), " ")
 		}
 		if utf8.RuneCountInString(value) > check.limit {
+			if check.name == "remark" {
+				return fmt.Errorf("备注最多 %d 个字符", check.limit)
+			}
 			return fmt.Errorf("%s 最多 %d 个字符", check.name, check.limit)
 		}
 	}
+	if req.WorkerCount != nil {
+		if *req.WorkerCount < 0 {
+			return fmt.Errorf("Worker 运行数量不能为负数")
+		}
+		if *req.WorkerCount > db.MaxTaskWorkerCount {
+			return fmt.Errorf("Worker 运行数量最多 %d", db.MaxTaskWorkerCount)
+		}
+	}
 	return nil
 }
 
@@ -125,12 +139,18 @@ func (s *Server) pgCreateTaskTemplate(w http.ResponseWriter, r *http.Request) {
 		writeErr(w, http.StatusBadRequest, "拦截/允许规则无效："+err.Error())
 		return
 	}
+	workerCount := 0
+	if req.WorkerCount != nil {
+		workerCount = *req.WorkerCount
+	}
 	template, err := pg.CreateTaskTemplate(db.TaskTemplateInput{
 		Name:           stringValue(req.Name),
 		Description:    stringValue(req.Description),
 		Goal:           stringValue(req.Goal),
+		Remark:         stringValue(req.Remark),
 		CategoryID:     req.CategoryID,
 		InterceptRules: rules,
+		WorkerCount:    workerCount,
 	})
 	if err != nil {
 		writeTaskTemplateErr(w, err)
@@ -160,11 +180,21 @@ func (s *Server) pgUpdateTaskTemplate(w http.ResponseWriter, r *http.Request) {
 	}
 	_, catPresent := present["category_id"]
 	_, rulesPresent := present["intercept_rules"]
-	if req.Name == nil && req.Description == nil && req.Goal == nil && !catPresent && !rulesPresent {
-		writeErr(w, http.StatusBadRequest, "至少需要提供 name、description、goal、category_id 或 intercept_rules")
+	_, remarkPresent := present["remark"]
+	_, workerPresent := present["worker_count"]
+	if req.Name == nil && req.Description == nil && req.Goal == nil && !catPresent && !rulesPresent && !remarkPresent && !workerPresent {
+		writeErr(w, http.StatusBadRequest, "至少需要提供 name、description、goal、remark、category_id、intercept_rules 或 worker_count")
 		return
 	}
 	patch := db.TaskTemplatePatch{Name: req.Name, Description: req.Description, Goal: req.Goal}
+	if remarkPresent {
+		if req.Remark == nil {
+			empty := ""
+			patch.Remark = &empty
+		} else {
+			patch.Remark = req.Remark
+		}
+	}
 	if catPresent {
 		patch.SetCategoryID = true
 		patch.CategoryID = req.CategoryID
@@ -178,6 +208,13 @@ func (s *Server) pgUpdateTaskTemplate(w http.ResponseWriter, r *http.Request) {
 		patch.SetInterceptRules = true
 		patch.InterceptRules = rules
 	}
+	if workerPresent {
+		n := 0
+		if req.WorkerCount != nil {
+			n = *req.WorkerCount
+		}
+		patch.WorkerCount = &n
+	}
 	template, err := pg.PatchTaskTemplate(id, patch)
 	if err != nil {
 		writeTaskTemplateErr(w, err)

server/tools_wire_test.go

@@ -5,9 +5,9 @@ import (
 	"encoding/json"
 	"testing"
 
-	actool "github.com/Autumn-27/norma/tool"
 	"github.com/Autumn-27/artex/agent"
 	"github.com/Autumn-27/artex/db"
+	actool "github.com/Autumn-27/norma/tool"
 )
 
 func names(tools []actool.CoreTool) map[string]actool.CoreTool {

server/worker_pools_test.go

@@ -0,0 +1,393 @@
+package server
+
+import (
+	"context"
+	"encoding/json"
+	"errors"
+	"net/http"
+	"net/http/httptest"
+	"strconv"
+	"strings"
+	"sync/atomic"
+	"testing"
+	"time"
+
+	"github.com/Autumn-27/artex/agent"
+	"github.com/Autumn-27/artex/db"
+	"github.com/Autumn-27/norma/llm"
+)
+
+func TestExpandWorkerSlotsLegacy(t *testing.T) {
+	for _, tc := range []struct {
+		name  string
+		count int
+		want  int
+	}{
+		{"configured", 2, 2},
+		{"zero defaults", 0, 3},
+		{"negative defaults", -1, 3},
+	} {
+		t.Run(tc.name, func(t *testing.T) {
+			slots, err := ExpandWorkerSlots(nil, tc.count)
+			if err != nil || len(slots) != tc.want {
+				t.Fatalf("slots=%v err=%v, want %d", slots, err, tc.want)
+			}
+			for i, slot := range slots {
+				if slot.Name != "work#"+strconv.Itoa(i+1) || slot.ProfileID != 0 {
+					t.Fatalf("slot[%d]=%+v", i, slot)
+				}
+			}
+		})
+	}
+}
+
+func TestExpandWorkerSlotsPoolOrder(t *testing.T) {
+	slots, err := ExpandWorkerSlots([]WorkerPool{{ProfileID: 7, Workers: 2}, {ProfileID: 12, Workers: 3}}, 99)
+	if err != nil || len(slots) != 5 {
+		t.Fatalf("slots=%v err=%v", slots, err)
+	}
+	for i, slot := range slots {
+		want := int64(7)
+		if i >= 2 {
+			want = 12
+		}
+		if slot.Name != "work#"+strconv.Itoa(i+1) || slot.ProfileID != want {
+			t.Fatalf("slot[%d]=%+v want profile %d", i, slot, want)
+		}
+	}
+}
+
+func TestValidateWorkerPools(t *testing.T) {
+	for _, tc := range []struct {
+		name  string
+		pools []WorkerPool
+		valid bool
+	}{
+		{"empty", nil, true},
+		{"bad profile", []WorkerPool{{ProfileID: 0, Workers: 1}}, false},
+		{"bad workers", []WorkerPool{{ProfileID: 1, Workers: 0}}, false},
+		{"duplicate", []WorkerPool{{ProfileID: 1, Workers: 1}, {ProfileID: 1, Workers: 1}}, false},
+		{"too many", []WorkerPool{{ProfileID: 1, Workers: 51}}, false},
+		{"valid", []WorkerPool{{ProfileID: 1, Workers: 1}}, true},
+	} {
+		t.Run(tc.name, func(t *testing.T) {
+			if got := ValidateWorkerPools(tc.pools) == nil; got != tc.valid {
+				t.Fatalf("valid=%v want %v", got, tc.valid)
+			}
+		})
+	}
+}
+
+func TestWorkerPoolSettingsRoundTrip(t *testing.T) {
+	m, err := NewManager(t.TempDir(), "")
+	if err != nil {
+		t.Skipf("postgres unavailable: %v", err)
+	}
+	defer m.Close()
+	pID, err := m.pg.SaveProfile(&db.LLMProfile{Name: "pool-test", Format: "openai", Model: "pool-test", APIKey: "pool-test-key"})
+	if err != nil {
+		t.Fatal(err)
+	}
+	defer m.pg.DeleteProfile(pID)
+	oldPools, _, _ := m.pg.GetSetting(settingWorkerPools)
+	oldWorkers, workersSet, _ := m.pg.GetSetting(settingWorkers)
+	defer func() {
+		if workersSet {
+			_ = m.pg.SetSetting(settingWorkers, oldWorkers)
+		} else {
+			_, _ = m.pg.Exec("DELETE FROM settings WHERE key=$1", settingWorkers)
+		}
+		if oldPools != "" {
+			_ = m.pg.SetSetting(settingWorkerPools, oldPools)
+		} else {
+			_, _ = m.pg.Exec("DELETE FROM settings WHERE key=$1", settingWorkerPools)
+		}
+	}()
+	if err := m.SetWorkerPools([]WorkerPool{{ProfileID: 9223372036854775807, Workers: 1}}); err == nil {
+		t.Fatal("nonexistent profile accepted")
+	}
+	if err := m.SetWorkerPools([]WorkerPool{{ProfileID: pID, Workers: 2}}); err != nil {
+		t.Fatal(err)
+	}
+	got := m.WorkerPools()
+	if len(got) != 1 || got[0].ProfileID != pID || got[0].Workers != 2 || m.Workers() != 2 {
+		t.Fatalf("round trip pools=%+v workers=%d", got, m.Workers())
+	}
+	if err := m.SetWorkerPools(nil); err != nil {
+		t.Fatal(err)
+	}
+	if got := m.WorkerPools(); got == nil || len(got) != 0 {
+		t.Fatalf("cleared pools=%+v", got)
+	}
+
+	ctx, cancel := context.WithCancel(context.Background())
+	defer cancel()
+	s := New(ctx, m, t.TempDir(), t.TempDir(), t.TempDir())
+	if _, err := m.pg.Exec("DELETE FROM settings WHERE key=$1", settingWorkerPools); err != nil {
+		t.Fatal(err)
+	}
+	res := httptest.NewRecorder()
+	s.getSettings(res, httptest.NewRequest(http.MethodGet, "/api/settings", nil))
+	var unset map[string]any
+	if err := json.Unmarshal(res.Body.Bytes(), &unset); err != nil {
+		t.Fatal(err)
+	}
+	if pools, ok := unset["worker_pools"].([]any); !ok || pools == nil || len(pools) != 0 {
+		t.Fatalf("GET unset worker_pools=%#v, want empty array", unset["worker_pools"])
+	}
+	put := httptest.NewRequest(http.MethodPut, "/api/settings", strings.NewReader(`{"workers":2,"worker_pools":[{"profile_id":"`+strconv.FormatInt(pID, 10)+`","workers":2}]}`))
+	putResult := httptest.NewRecorder()
+	s.putSettings(putResult, put)
+	if putResult.Code != http.StatusOK {
+		t.Fatalf("PUT worker pools status=%d body=%s", putResult.Code, putResult.Body.String())
+	}
+	if m.Workers() != 2 || len(m.WorkerPools()) != 1 {
+		t.Fatalf("PUT worker pools did not sync settings: workers=%d pools=%+v", m.Workers(), m.WorkerPools())
+	}
+	workersOnly := httptest.NewRequest(http.MethodPut, "/api/settings", strings.NewReader(`{"workers":9}`))
+	workersOnlyResult := httptest.NewRecorder()
+	s.putSettings(workersOnlyResult, workersOnly)
+	if workersOnlyResult.Code != http.StatusOK || m.Workers() != 2 {
+		t.Fatalf("workers-only PUT changed pool total: status=%d workers=%d", workersOnlyResult.Code, m.Workers())
+	}
+	req := httptest.NewRequest(http.MethodGet, "/api/settings", nil)
+	res = httptest.NewRecorder()
+	s.getSettings(res, req)
+	if res.Code != http.StatusOK {
+		t.Fatalf("GET settings status=%d", res.Code)
+	}
+	var payload map[string]any
+	if err := json.Unmarshal(res.Body.Bytes(), &payload); err != nil {
+		t.Fatal(err)
+	}
+	pools, ok := payload["worker_pools"].([]any)
+	if !ok || len(pools) != 1 || pools[0].(map[string]any)["profile_id"] != strconv.FormatInt(pID, 10) {
+		t.Fatalf("GET worker_pools=%#v, want string profile id", payload["worker_pools"])
+	}
+	clear := httptest.NewRequest(http.MethodPut, "/api/settings", strings.NewReader(`{"workers":4,"worker_pools":[]}`))
+	clearResult := httptest.NewRecorder()
+	s.putSettings(clearResult, clear)
+	if clearResult.Code != http.StatusOK || m.Workers() != 4 || len(m.WorkerPools()) != 0 {
+		t.Fatalf("PUT legacy clear status=%d body=%s workers=%d pools=%+v", clearResult.Code, clearResult.Body.String(), m.Workers(), m.WorkerPools())
+	}
+	res = httptest.NewRecorder()
+	s.getSettings(res, httptest.NewRequest(http.MethodGet, "/api/settings", nil))
+	var cleared map[string]any
+	if err := json.Unmarshal(res.Body.Bytes(), &cleared); err != nil {
+		t.Fatal(err)
+	}
+	if pools, ok := cleared["worker_pools"].([]any); !ok || pools == nil || len(pools) != 0 {
+		t.Fatalf("GET cleared worker_pools=%#v, want empty array", cleared["worker_pools"])
+	}
+}
+
+func TestPinnedRuntimeMissingProfileDoesNotFallback(t *testing.T) {
+	m, err := NewManager(t.TempDir(), "")
+	if err != nil {
+		t.Skipf("postgres unavailable: %v", err)
+	}
+	defer m.Close()
+	task, err := m.CreateTask("pinned runtime test", "test", nil, 0, 0)
+	if err != nil {
+		t.Fatal(err)
+	}
+	defer m.DeleteTask(task.ID, DeleteTaskOptions{})
+	fallback, err := m.pg.SaveProfile(&db.LLMProfile{Name: "pinned-fallback", Format: "openai", Model: "fallback", APIKey: "fallback-key"})
+	if err != nil {
+		t.Fatal(err)
+	}
+	defer m.pg.DeleteProfile(fallback)
+	if _, err := m.ReplaceTaskLLMProfiles(task.ID, []int64{fallback}, fallback); err != nil {
+		t.Fatal(err)
+	}
+	missing := int64(9223372036854775807)
+	s := &Server{m: m, llmOn: true, llmProv: &scriptedLLMProvider{}}
+	runtime := &taskLLMRuntime{s: s, taskID: task.ID, agentKey: "worker", pinProfileID: &missing}
+	if _, err := runtime.current(); err == nil {
+		t.Fatal("missing pinned profile silently fell back")
+	}
+	if _, ok := runtime.activeCfg(); ok || runtime.nonStreaming() || runtime.maxTokens() != 0 {
+		t.Fatal("missing pinned profile exposed fallback runtime settings")
+	}
+}
+
+func TestPinnedRuntimeQuotaDoesNotAdvanceTaskFailover(t *testing.T) {
+	m, err := NewManager(t.TempDir(), "")
+	if err != nil {
+		t.Skipf("postgres unavailable: %v", err)
+	}
+	defer m.Close()
+	task, err := m.CreateTask("pinned quota", "test", nil, 0, 0)
+	if err != nil {
+		t.Fatal(err)
+	}
+	defer m.DeleteTask(task.ID, DeleteTaskOptions{})
+	profileID, err := m.pg.SaveProfile(&db.LLMProfile{Name: "pinned-quota", Format: "openai", Model: "pinned", APIKey: "key"})
+	if err != nil {
+		t.Fatal(err)
+	}
+	defer m.pg.DeleteProfile(profileID)
+	if _, err := m.ReplaceTaskLLMProfiles(task.ID, []int64{profileID}, profileID); err != nil {
+		t.Fatal(err)
+	}
+	before, err := m.pg.GetTask(taskIDInt(t, task.ID))
+	if err != nil || before == nil || before.ActiveLLMProfileID == nil {
+		t.Fatalf("load task failover state: task=%+v err=%v", before, err)
+	}
+	provider := &scriptedLLMProvider{err: errors.New("insufficient_quota")}
+	s := &Server{m: m, provByProfile: map[int64]*provEntry{profileID: {prov: provider, cfg: agent.Config{}}}}
+	id := profileID
+	runtime := &taskLLMRuntime{s: s, taskID: task.ID, agentKey: "worker", pinProfileID: &id}
+	if _, _, _, err := runtime.Complete(context.Background(), llm.CompletionRequest{}); err == nil {
+		t.Fatal("pinned quota error unexpectedly succeeded")
+	}
+	after, err := m.pg.GetTask(taskIDInt(t, task.ID))
+	if err != nil || after == nil || after.LLMChainRevision != before.LLMChainRevision || after.ActiveLLMProfileID == nil || *after.ActiveLLMProfileID != *before.ActiveLLMProfileID {
+		t.Fatalf("pinned quota changed task failover state: before=%+v after=%+v err=%v", before, after, err)
+	}
+}
+
+func taskIDInt(t *testing.T, id string) int64 {
+	t.Helper()
+	n, err := strconv.ParseInt(id, 10, 64)
+	if err != nil {
+		t.Fatal(err)
+	}
+	return n
+}
+
+func TestPinnedWorkerLoopDoesNotConsultTaskRouterAndReportsFailureOnce(t *testing.T) {
+	m, err := NewManager(t.TempDir(), "")
+	if err != nil {
+		t.Skipf("postgres unavailable: %v", err)
+	}
+	defer m.Close()
+	task, err := m.CreateTask("pinned worker loop", "test", nil, 0, 0)
+	if err != nil {
+		t.Fatal(err)
+	}
+	defer m.DeleteTask(task.ID, DeleteTaskOptions{})
+
+	e := NewEngine(m)
+	var routerCalls atomic.Int32
+	e.SetAuthoritativeAgentResolver(func(*Task) (*agent.Planner, *agent.Worker) {
+		routerCalls.Add(1)
+		return nil, nil
+	})
+	e.SetWorkerResolver(func(*Task, WorkerSlot) (*agent.Worker, error) {
+		return nil, errors.New("profile unavailable")
+	})
+	ctx, cancel := context.WithCancel(context.Background())
+	done := make(chan struct{})
+	go func() {
+		e.workerLoop(ctx, task, WorkerSlot{Name: "work#4", ProfileID: 12})
+		close(done)
+	}()
+	time.Sleep(100 * time.Millisecond)
+	cancel()
+	select {
+	case <-done:
+	case <-time.After(time.Second):
+		t.Fatal("worker loop did not stop")
+	}
+	if routerCalls.Load() != 0 {
+		t.Fatalf("pinned worker consulted task router %d times", routerCalls.Load())
+	}
+	activities, _, err := task.Store.ActivityList(nil, 0, 20)
+	if err != nil {
+		t.Fatal(err)
+	}
+	var failures int
+	for _, activity := range activities {
+		if activity.Worker == "system" && activity.IsError {
+			failures++
+		}
+	}
+	if failures != 1 {
+		t.Fatalf("pinned failure activities=%d, want one throttled entry: %+v", failures, activities)
+	}
+}
+
+func TestPinnedWorkerAvailabilityRecoveryUsesSystemSession(t *testing.T) {
+	m, err := NewManager(t.TempDir(), "")
+	if err != nil {
+		t.Skipf("postgres unavailable: %v", err)
+	}
+	defer m.Close()
+	task, err := m.CreateTask("pinned availability", "test", nil, 0, 0)
+	if err != nil {
+		t.Fatal(err)
+	}
+	defer m.DeleteTask(task.ID, DeleteTaskOptions{})
+	e := NewEngine(m)
+	slot := WorkerSlot{Name: "work#4", ProfileID: 12}
+	e.reportSlotAvailability(task, slot, errors.New("profile unavailable"))
+	e.reportSlotAvailability(task, slot, nil)
+	activities, _, err := task.Store.ActivityList(nil, 0, 20)
+	if err != nil {
+		t.Fatal(err)
+	}
+	var failure, recovery int
+	for _, activity := range activities {
+		if activity.Worker != "system" {
+			continue
+		}
+		if activity.IsError && strings.Contains(activity.Summary, "work#4") && strings.Contains(activity.Summary, "#12") {
+			failure++
+		}
+		if !activity.IsError && strings.Contains(activity.Summary, "work#4") && strings.Contains(activity.Summary, "#12") {
+			recovery++
+		}
+	}
+	if failure != 1 || recovery != 1 {
+		t.Fatalf("system availability activities failure=%d recovery=%d: %+v", failure, recovery, activities)
+	}
+}
+
+func TestDetachedIntentPinnedOwnerUsesPinnedResolver(t *testing.T) {
+	m, err := NewManager(t.TempDir(), "")
+	if err != nil {
+		t.Skipf("postgres unavailable: %v", err)
+	}
+	defer m.Close()
+	task, err := m.CreateTask("detached pinned owner", "test", nil, 0, 0)
+	if err != nil {
+		t.Fatal(err)
+	}
+	defer m.DeleteTask(task.ID, DeleteTaskOptions{})
+	intentID, err := task.Store.AddIntent(map[string]any{"text": "continue"}, 0, nil, "test")
+	if err != nil {
+		t.Fatal(err)
+	}
+	if ok, err := task.Store.ClaimIntent(intentID, "work#4"); err != nil || !ok {
+		t.Fatalf("claim intent: ok=%v err=%v", ok, err)
+	}
+	if changed, err := task.Store.CompareAndSetIntentState(intentID, "running", "paused"); err != nil || !changed {
+		t.Fatalf("pause intent: changed=%v err=%v", changed, err)
+	}
+	e := NewEngine(m)
+	e.slotPlans.Store(task.ID, map[string]int64{"work#4": 12})
+	var pinnedCalls, snapshotCalls atomic.Int32
+	e.SetWorkerResolver(func(_ *Task, slot WorkerSlot) (*agent.Worker, error) {
+		pinnedCalls.Add(1)
+		if slot.Name != "work#4" || slot.ProfileID != 12 {
+			t.Fatalf("unexpected pinned slot: %+v", slot)
+		}
+		return nil, errors.New("profile unavailable")
+	})
+	e.SetAuthoritativeAgentResolver(func(*Task) (*agent.Planner, *agent.Worker) {
+		snapshotCalls.Add(1)
+		return nil, nil
+	})
+	err = e.runDetachedIntent(context.Background(), task, intentID, "request", "human", "agent")
+	if err == nil || !strings.Contains(err.Error(), "LLM 配置 #12 不可用") {
+		t.Fatalf("runDetachedIntent err=%v", err)
+	}
+	if pinnedCalls.Load() != 1 || snapshotCalls.Load() != 0 {
+		t.Fatalf("pinned calls=%d snapshot calls=%d", pinnedCalls.Load(), snapshotCalls.Load())
+	}
+	node, err := task.Store.GetNode(intentID)
+	if err != nil || node == nil || node.State != "paused" {
+		t.Fatalf("intent changed after unavailable resolver: node=%+v err=%v", node, err)
+	}
+}

web/src/app/(main)/function/findings/detail/page.tsx

@@ -58,6 +58,7 @@ function FindingDetailInner() {
   const searchParams = useSearchParams();
   const id = searchParams.get("id") ?? "";
   const contextTaskId = searchParams.get("context_task") ?? "";
+  const backHref = contextTaskId ? `/function/tasks/detail?id=${contextTaskId}&tab=findings` : "/function/findings";
   const [finding, setFinding] = React.useState<Finding | null>(null);
   const [loaded, setLoaded] = React.useState(false);
   const [tab, setTab] = React.useState("overview");
@@ -117,7 +118,7 @@ function FindingDetailInner() {
         <p className="text-muted-foreground">{loaded ? `未找到发现 ${id}` : "加载中…"}</p>
         {loaded && (
           <Button asChild variant="outline">
-            <Link href="/function/findings">
+            <Link href={backHref}>
               <ArrowLeftIcon /> 返回发现列表
             </Link>
           </Button>
@@ -135,7 +136,7 @@ function FindingDetailInner() {
         <div className="flex flex-wrap items-center gap-2">
           <SidebarTrigger className="-ml-1" />
           <Button asChild variant="ghost" size="icon" className="size-7">
-            <Link href="/function/findings">
+            <Link href={backHref}>
               <ArrowLeftIcon />
             </Link>
           </Button>

web/src/app/(main)/function/tasks/detail/_tabs/findings-tab.tsx

@@ -115,11 +115,7 @@ function Row({
         </span>
         {f.finding_id && (
           <Link
-            href={
-              f.inherited
-                ? `/function/findings/detail?id=${f.finding_id}&context_task=${contextTaskId}`
-                : `/function/findings/detail?id=${f.finding_id}`
-            }
+            href={`/function/findings/detail?id=${f.finding_id}&context_task=${contextTaskId}`}
             className="text-muted-foreground hover:text-primary inline-flex shrink-0 items-center gap-0.5 text-xs"
             title="查看漏洞详情"
           >
@@ -205,6 +201,7 @@ export function FindingsTab({ taskId }: { taskId: string }) {
       </ToggleGroup>
       <FindingCaseList
         query={{ task: taskId, severity: severity === "all" ? undefined : severity, sort: "severity" }}
+        contextTask={taskId}
       />
       {[
         ...new Set(

web/src/app/(main)/function/tasks/detail/page.tsx

@@ -354,7 +354,17 @@ function TaskDetailInner() {
   );
 
   return (
-    <Tabs value={tab} onValueChange={setTab} className="flex flex-1 flex-col gap-0">
+    <Tabs
+      value={tab}
+      onValueChange={(value) => {
+        setTab(value);
+        const params = new URLSearchParams(searchParams.toString());
+        params.set("tab", value);
+        params.delete("session");
+        router.replace(`/function/tasks/detail?${params.toString()}`, { scroll: false });
+      }}
+      className="flex flex-1 flex-col gap-0"
+    >
       {/* Top fixed area */}
       <header className="sticky top-0 z-10 flex flex-col gap-2 border-b bg-background/95 px-4 py-2.5 backdrop-blur lg:px-6">
         <div className="flex items-center gap-2">
@@ -404,6 +414,11 @@ function TaskDetailInner() {
           </Button>
         </div>
         <p className="truncate text-xs text-muted-foreground">{task.goal}</p>
+        {task.remark ? (
+          <p className="truncate text-xs text-muted-foreground" title={task.remark}>
+            备注：{task.remark}
+          </p>
+        ) : null}
         {/* Tabs */}
         <div className="no-scrollbar min-w-0 overflow-x-auto">
           <TabsList variant="default" className="min-w-max">

web/src/app/(main)/function/tasks/page.tsx

@@ -137,6 +137,7 @@ import type {
   TaskCategory,
   TaskStatus,
 } from "@/lib/types";
+import { MAX_TASK_REMARK_RUNES, MAX_TASK_WORKER_COUNT, parseWorkerCountInput } from "@/lib/types";
 import { cn } from "@/lib/utils";
 
 // fmtBytes renders a human file size for the upload manifest.
@@ -3104,6 +3105,7 @@ function CreateTaskSheet({
 }) {
   const [open, setOpen] = React.useState(false);
   const [name, setName] = React.useState("");
+  const [remark, setRemark] = React.useState("");
   const [categoryID, setCategoryID] = React.useState<number | undefined>(undefined);
   const [description, setDescription] = React.useState("");
   const [goal, setGoal] = React.useState("");
@@ -3118,6 +3120,7 @@ function CreateTaskSheet({
   const [heartbeatMin, setHeartbeatMin] = React.useState("10"); // planner 心跳(分钟);默认10,下限10(与后端一致)
   const [seedFirstIntent, setSeedFirstIntent] = React.useState(false); // 创建时下发种子意图,worker 免等首轮 planner 直接开跑;默认关闭,走标准先规划再执行
   const [coverageEnabled, setCoverageEnabled] = React.useState(true); // 资产覆盖度功能;默认开。关闭=不计算/展示覆盖度、不累积范围、隐藏范围类工具(company 关联不受影响)
+  const [workerCount, setWorkerCount] = React.useState(""); // 本任务 Worker 数;空/0=用全局默认
   const [interceptRules, setInterceptRules] = React.useState<AssetInterceptRuleInput[]>([]); // 任务级资产拦截规则(仅本任务生效,不进全局表)
   const [scheduleEnabled, setScheduleEnabled] = React.useState(false);
   const [schedule, setSchedule] = React.useState<ScheduleInput>({
@@ -3181,12 +3184,18 @@ function CreateTaskSheet({
       toast.error(`最多关联 ${MAX_SOURCE_TASKS} 个来源任务`);
       return;
     }
+    const workerCountNum = parseWorkerCountInput(workerCount);
+    if (workerCountNum == null) {
+      toast.error(`Worker 运行数量必须是 0-${MAX_TASK_WORKER_COUNT} 的整数`);
+      return;
+    }
     setCreating(true);
     try {
       const timeoutSec = Math.max(0, Math.floor(Number(timeoutMin) || 0)) * 60;
       const heartbeatSec = Math.max(10, Math.floor(Number(heartbeatMin) || 10)) * 60; // 下限 10min，与后端归一一致
       await api.createTask({
         name: name.trim(),
+        remark: remark.trim(),
         categoryId: categoryID,
         description: description.trim(),
         goal: goal.trim(),
@@ -3197,6 +3206,7 @@ function CreateTaskSheet({
         seedFirstIntent,
         planHeartbeatSeconds: heartbeatSec,
         coverageEnabled,
+        workerCount: workerCountNum,
         schedule: scheduleEnabled ? { ...schedule, task_ids: [] } : undefined,
         interceptRules: interceptRules
           .map((r) => ({ ...r, pattern: r.pattern.trim() }))
@@ -3204,6 +3214,7 @@ function CreateTaskSheet({
       });
       toast.success("任务已创建");
       setName("");
+      setRemark("");
       setCategoryID(undefined);
       setDescription("");
       setGoal("");
@@ -3215,6 +3226,7 @@ function CreateTaskSheet({
       setHeartbeatMin("10");
       setSeedFirstIntent(false);
       setCoverageEnabled(true);
+      setWorkerCount("");
       setInterceptRules([]);
       setScheduleEnabled(false);
       setSchedule({
@@ -3263,15 +3275,19 @@ function CreateTaskSheet({
             <TaskTemplateControls
               description={description}
               goal={goal}
+              remark={remark}
               categoryID={categoryID}
               interceptRules={interceptRules}
+              workerCount={workerCount}
               selectedTemplateID={selectedTemplateID}
               onSelectedTemplateIDChange={setSelectedTemplateID}
               onApply={(template) => {
                 setDescription(template.description);
                 setGoal(template.goal);
+                setRemark(template.remark ?? "");
                 setCategoryID(template.category_id ?? undefined);
                 setInterceptRules(template.intercept_rules ?? []);
+                setWorkerCount(template.worker_count ? String(template.worker_count) : "");
                 setUploadCount(0);
               }}
               portalContainer={sheetContentRef}
@@ -3285,6 +3301,17 @@ function CreateTaskSheet({
                 onChange={(e) => setName(e.target.value)}
               />
             </div>
+            <div className="grid gap-2">
+              <Label htmlFor="remark">备注（可选）</Label>
+              <Textarea
+                id="remark"
+                className="min-h-20"
+                maxLength={MAX_TASK_REMARK_RUNES}
+                placeholder="补充说明、环境信息等，便于在任务列表识别"
+                value={remark}
+                onChange={(e) => setRemark(e.target.value)}
+              />
+            </div>
             <Field>
               <FieldLabel htmlFor="task-category">任务分类</FieldLabel>
               <CategoryPicker
@@ -3507,7 +3534,7 @@ function CreateTaskSheet({
               <CollapsibleTrigger className="group flex w-full items-center gap-2 border-t pt-4 text-sm font-medium">
                 <ChevronRightIcon className="text-muted-foreground size-4 transition-transform group-data-[state=open]:rotate-90" />
                 高级设置
-                <span className="text-muted-foreground ml-auto text-xs font-normal">超时 · 心跳 · 首个意图</span>
+                <span className="text-muted-foreground ml-auto text-xs font-normal">超时 · 心跳 · Worker · 首个意图</span>
               </CollapsibleTrigger>
               <CollapsibleContent className="grid gap-5 pt-5">
                 <div className="grid gap-2">
@@ -3541,6 +3568,23 @@ function CreateTaskSheet({
                     worker）。下限 10 分钟。
                   </p>
                 </div>
+                <div className="grid gap-2">
+                  <Label htmlFor="worker-count">Worker 运行数量（可选）</Label>
+                  <Input
+                    id="worker-count"
+                    type="number"
+                    min={0}
+                    max={MAX_TASK_WORKER_COUNT}
+                    className="w-40"
+                    placeholder="留空 = 全局默认"
+                    value={workerCount}
+                    onChange={(e) => setWorkerCount(e.target.value)}
+                  />
+                  <p className="text-muted-foreground text-xs">
+                    单独设置本任务并发运行的工作 Agent 数量；留空或 0 = 用全局默认（系统设置里的工作并发，默认
+                    3）。上限 50。
+                  </p>
+                </div>
                 <div className="grid gap-2">
                   <label htmlFor="seed-first-intent" className="flex items-center gap-2 text-sm">
                     <Checkbox

web/src/app/(main)/system/settings/page.tsx

@@ -14,7 +14,7 @@ import { Select, SelectContent, SelectItem, SelectTrigger, SelectValue } from "@
 import { Switch } from "@/components/ui/switch";
 import { api } from "@/lib/api";
 import { CHAT_SEND_MODE_OPTIONS, type ChatSendMode, setChatSendMode, useChatSendMode } from "@/lib/chat-send-mode";
-import type { Settings } from "@/lib/types";
+import type { LLMProfile, Settings, WorkerPool } from "@/lib/types";
 
 import { BasicAuthCard } from "./_components/basic-auth-card";
 import { UpdateCard } from "./_components/update-card";
@@ -40,11 +40,14 @@ export default function SystemSettingsPage() {
   const [savingGlobalProxy, setSavingGlobalProxy] = React.useState(false);
   const [testing, setTesting] = React.useState(false);
   const [loaded, setLoaded] = React.useState(false);
+  const [settingsLoaded, setSettingsLoaded] = React.useState(false);
   const [saving, setSaving] = React.useState(false);
   const [savingKey, setSavingKey] = React.useState(false);
   const [pyInterp, setPyInterp] = React.useState("");
   const [workers, setWorkers] = React.useState("3");
   const [savingWorkers, setSavingWorkers] = React.useState(false);
+  const [workerPools, setWorkerPools] = React.useState<WorkerPool[]>([]);
+  const [llmProfiles, setLlmProfiles] = React.useState<LLMProfile[]>([]);
   // 操作约束注入范围(默认都开)。
   const [injectPlanner, setInjectPlanner] = React.useState(true);
   const [injectWorker, setInjectWorker] = React.useState(true);
@@ -67,20 +70,58 @@ export default function SystemSettingsPage() {
     setGlobalProxyInput(s.global_proxy ?? "");
     setPyInterp(s.python_interpreter ?? "");
     setWorkers(String(s.workers ?? 3));
+    setWorkerPools(s.worker_pools ?? []);
     setInjectPlanner(s.constraints_inject_planner !== false);
     setInjectWorker(s.constraints_inject_worker !== false);
     setNoaCompaction(!!s.noa_compaction);
   }, []);
 
   const saveWorkers = () => {
+    if (!settingsLoaded) {
+      toast.error("设置加载失败，请刷新后再保存 Worker 配置");
+      return;
+    }
     const n = Number(workers);
+    if (workerPools.length > 0) {
+      const total = workerPools.reduce((sum, pool) => sum + Number(pool.workers), 0);
+      if (workerPools.some((pool) => !pool.profile_id)) {
+        toast.error("请选择每个模型池的 LLM 配置");
+        return;
+      }
+      if (workerPools.some((pool) => !Number.isInteger(Number(pool.workers)) || Number(pool.workers) <= 0)) {
+        toast.error("每个模型池的 Worker 数必须是大于 0 的整数");
+        return;
+      }
+      if (total > 50) {
+        toast.error("Worker 总数不能超过 50");
+        return;
+      }
+      const ids = workerPools.map((pool) => pool.profile_id);
+      if (new Set(ids).size !== ids.length) {
+        toast.error("不能重复选择同一个 LLM 配置");
+        return;
+      }
+      setSavingWorkers(true);
+      api
+        .setSettings({
+          workers: total,
+          worker_pools: workerPools.map((pool) => ({ ...pool, workers: Number(pool.workers) })),
+        })
+        .then((s) => {
+          apply(s);
+          toast.success("已保存多模型 Worker 池（对之后启动的任务生效）");
+        })
+        .catch((e) => toast.error(`保存失败：${(e as Error).message}`))
+        .finally(() => setSavingWorkers(false));
+      return;
+    }
     if (!Number.isInteger(n) || n <= 0) {
       toast.error("并发数必须是大于 0 的整数");
       return;
     }
     setSavingWorkers(true);
     api
-      .setSettings({ workers: n })
+      .setSettings({ workers: n, worker_pools: [] })
       .then((s) => {
         apply(s);
         toast.success("已保存并发工作 agent 数（对之后启动的任务生效）");
@@ -89,6 +130,8 @@ export default function SystemSettingsPage() {
       .finally(() => setSavingWorkers(false));
   };
 
+  const addWorkerPool = () => setWorkerPools((pools) => [...pools, { profile_id: "", workers: 1 }]);
+
   const savePython = () => {
     setSaving(true);
     api
@@ -112,11 +155,24 @@ export default function SystemSettingsPage() {
   React.useEffect(() => {
     api
       .settings()
-      .then(apply)
-      .catch(() => undefined)
+      .then((s) => {
+        apply(s);
+        setSettingsLoaded(true);
+      })
+      .catch(() => {
+        setSettingsLoaded(false);
+        toast.error("设置加载失败，Worker 配置已锁定，避免覆盖服务器上的模型池");
+      })
       .finally(() => setLoaded(true));
   }, [apply]);
 
+  React.useEffect(() => {
+    api
+      .llmProfiles()
+      .then(setLlmProfiles)
+      .catch(() => setLlmProfiles([]));
+  }, []);
+
   const toggleTraffic = (v: boolean) => {
     setTrafficCapture(v); // optimistic
     setSaving(true);
@@ -756,6 +812,81 @@ export default function SystemSettingsPage() {
             </CardDescription>
           </CardHeader>
           <CardContent className="flex flex-col gap-3">
+            <div className="flex flex-col gap-2 rounded-md border p-3">
+              <div className="flex items-center justify-between">
+                <Label className="text-sm">多模型 Worker 池</Label>
+                <Button
+                  variant="outline"
+                  size="sm"
+                  onClick={addWorkerPool}
+                  disabled={!loaded || !settingsLoaded || savingWorkers}
+                >
+                  添加模型池
+                </Button>
+              </div>
+              {workerPools.map((pool, index) => {
+                const selected = llmProfiles.find((profile) => profile.id === pool.profile_id);
+                const placeholder =
+                  selected?.name || (pool.profile_id ? `配置 #${pool.profile_id} 已删除/不可用` : "选择 LLM 配置");
+                return (
+                  // biome-ignore lint/suspicious/noArrayIndexKey: pool rows are edited in place and only deleted.
+                  <div className="flex items-center gap-2" key={`${pool.profile_id}-${index}`}>
+                    <Select
+                      value={pool.profile_id || undefined}
+                      disabled={!settingsLoaded || savingWorkers}
+                      onValueChange={(value) =>
+                        setWorkerPools((current) =>
+                          current.map((item, itemIndex) =>
+                            itemIndex === index ? { ...item, profile_id: value } : item,
+                          ),
+                        )
+                      }
+                    >
+                      <SelectTrigger className="min-w-0 flex-1">
+                        <SelectValue placeholder={placeholder} />
+                      </SelectTrigger>
+                      <SelectContent>
+                        {!selected && pool.profile_id ? (
+                          <SelectItem value={pool.profile_id}>配置 #{pool.profile_id} 已删除/不可用</SelectItem>
+                        ) : null}
+                        {llmProfiles.map((profile) => (
+                          <SelectItem key={profile.id} value={profile.id}>
+                            {profile.name} · {profile.model}
+                          </SelectItem>
+                        ))}
+                      </SelectContent>
+                    </Select>
+                    <Input
+                      type="number"
+                      min={1}
+                      className="w-24 font-mono text-sm"
+                      value={pool.workers}
+                      disabled={!settingsLoaded || savingWorkers}
+                      onChange={(event) =>
+                        setWorkerPools((current) =>
+                          current.map((item, itemIndex) =>
+                            itemIndex === index ? { ...item, workers: Number(event.target.value) } : item,
+                          ),
+                        )
+                      }
+                    />
+                    <Button
+                      variant="ghost"
+                      size="sm"
+                      disabled={!settingsLoaded || savingWorkers}
+                      onClick={() => setWorkerPools((current) => current.filter((_, itemIndex) => itemIndex !== index))}
+                    >
+                      删除
+                    </Button>
+                  </div>
+                );
+              })}
+              {workerPools.length > 0 ? (
+                <div className="text-muted-foreground text-sm">
+                  当前合计：{workerPools.reduce((sum, pool) => sum + (Number(pool.workers) || 0), 0)} 个 Worker 会话
+                </div>
+              ) : null}
+            </div>
             <div className="flex items-center gap-2">
               <Input
                 type="number"
@@ -763,13 +894,19 @@ export default function SystemSettingsPage() {
                 className="w-32 font-mono text-sm"
                 placeholder="3"
                 value={workers}
-                disabled={!loaded || savingWorkers}
+                disabled={!loaded || !settingsLoaded || savingWorkers || workerPools.length > 0}
                 onChange={(e) => setWorkers(e.target.value)}
               />
-              <Button onClick={saveWorkers} disabled={!loaded || savingWorkers}>
+              {workerPools.length > 0 ? (
+                <span className="text-muted-foreground text-xs">已配置模型池时由各池数量决定并发。</span>
+              ) : null}
+              <Button onClick={saveWorkers} disabled={!loaded || !settingsLoaded || savingWorkers}>
                 保存
               </Button>
             </div>
+            {!settingsLoaded && loaded ? (
+              <p className="text-destructive text-xs">设置加载失败，Worker 配置已锁定；刷新成功后才能编辑或保存。</p>
+            ) : null}
           </CardContent>
         </Card>
 

web/src/components/finding-case-list.tsx

@@ -192,15 +192,29 @@ function FindingMergeSuggestionCard({
   onResolve,
 }: {
   suggestion: FindingCaseSuggestion;
-  onResolve: (s: FindingCaseSuggestion, accept: boolean) => Promise<void>;
+  onResolve: (s: FindingCaseSuggestion, accept: boolean, excludeIds?: string[]) => Promise<void>;
 }) {
   const [expanded, setExpanded] = React.useState(false);
   const [busy, setBusy] = React.useState(false);
+  const ids = React.useMemo(
+    () => (Array.isArray(s.finding_ids) && s.finding_ids.length > 0 ? s.finding_ids : [s.left_id, s.right_id]),
+    [s.finding_ids, s.left_id, s.right_id],
+  );
+  const [excluded, setExcluded] = React.useState<string[]>([]);
+  React.useEffect(() => {
+    setExcluded((current) => {
+      const next = current.filter((id) => ids.includes(id));
+      return next.length === current.length ? current : next;
+    });
+  }, [ids]);
+  const validExcluded = excluded.filter((id) => ids.includes(id));
+  const included = ids.filter((id) => !validExcluded.includes(id));
   async function resolve(accept: boolean) {
     if (busy) return;
+    if (accept && included.length < 2) return;
     setBusy(true);
     try {
-      await onResolve(s, accept);
+      await onResolve(s, accept, accept ? validExcluded : []);
     } finally {
       setBusy(false);
     }
@@ -209,17 +223,30 @@ function FindingMergeSuggestionCard({
     <Card size="sm" className="min-w-0">
       <CardHeader>
         <div className="flex flex-wrap items-center gap-2">
-          <Badge variant="outline">疑似重复 · 待确认</Badge>
+          <Badge variant="outline">疑似重复分组 · {ids.length} 条 · 待确认</Badge>
           <CardDescription>任务 #{s.task_id}</CardDescription>
         </div>
         <CardTitle className="break-words">{s.title}</CardTitle>
-        <CardDescription className="flex flex-wrap gap-3">
-          <Link className="underline underline-offset-4" href={`/function/findings/detail/?id=${s.left_id}`}>
-            查看原报告 #{s.left_id}
-          </Link>
-          <Link className="underline underline-offset-4" href={`/function/findings/detail/?id=${s.right_id}`}>
-            查看原报告 #{s.right_id}
-          </Link>
+        <CardDescription className="flex flex-col gap-1">
+          {ids.map((id) => (
+            <span key={id} className="flex items-center gap-2">
+              <Checkbox
+                checked={!excluded.includes(id)}
+                disabled={busy}
+                onCheckedChange={(value) =>
+                  setExcluded((current) =>
+                    value === true
+                      ? current.filter((item) => item !== id)
+                      : [...current.filter((item) => item !== id), id],
+                  )
+                }
+                aria-label={`归并时包含上报 #${id}`}
+              />
+              <Link className="underline underline-offset-4" href={`/function/findings/detail/?id=${id}`}>
+                查看原报告 #{id}
+              </Link>
+            </span>
+          ))}
         </CardDescription>
       </CardHeader>
       <CardContent>
@@ -238,11 +265,12 @@ function FindingMergeSuggestionCard({
         </Collapsible>
       </CardContent>
       <CardFooter className="flex flex-wrap gap-2">
-        <Button variant="outline" size="sm" disabled={busy} onClick={() => resolve(true)}>
-          {busy ? <Spinner /> : null}确认归并
+        <Button variant="outline" size="sm" disabled={busy || included.length < 2} onClick={() => resolve(true)}>
+          {busy ? <Spinner /> : null}
+          {included.length === ids.length ? "整组确认归并" : `确认归并所选（${included.length} 条）`}
         </Button>
         <Button variant="ghost" size="sm" disabled={busy} onClick={() => resolve(false)}>
-          不是同一漏洞
+          整组不是同一漏洞
         </Button>
       </CardFooter>
     </Card>
@@ -271,21 +299,39 @@ export function FindingCaseReviewPanel({ taskId, onChange }: { taskId?: string;
   const [suggestions, setSuggestions] = React.useState<FindingCaseSuggestion[]>([]);
   const [runs, setRuns] = React.useState<FindingCaseReviewRun[]>([]);
   const [error, setError] = React.useState("");
+  const [reload, setReload] = React.useState(0);
+  const requestSeq = React.useRef(0);
+  const taskKey = taskId ?? "";
+  const taskKeyRef = React.useRef(taskKey);
+  taskKeyRef.current = taskKey;
+  const previousTaskKey = React.useRef(taskKey);
   React.useEffect(() => {
+    if (previousTaskKey.current === taskKey) return;
+    previousTaskKey.current = taskKey;
+    setSuggestions([]);
+    setRuns([]);
+    setError("");
+  }, [taskKey]);
+  React.useEffect(() => {
+    const seq = ++requestSeq.current;
+    const requestedTask = taskKey;
+    void reload;
     let active = true;
     let timer: ReturnType<typeof setTimeout>;
     async function load() {
       try {
         const [s, r] = await Promise.all([api.findingCaseSuggestions(taskId), api.findingCaseReviewRuns()]);
-        if (active) {
+        if (active && seq === requestSeq.current && requestedTask === taskKeyRef.current) {
           setSuggestions(s);
           setRuns(r.filter((x) => !taskId || x.task_id === taskId));
           setError("");
         }
       } catch (e) {
-        if (active) setError((e as Error).message);
+        if (active && seq === requestSeq.current && requestedTask === taskKeyRef.current)
+          setError((e as Error).message);
       } finally {
-        if (active) timer = setTimeout(load, 5000);
+        if (active && seq === requestSeq.current && requestedTask === taskKeyRef.current)
+          timer = setTimeout(load, 5000);
       }
     }
     void load();
@@ -293,19 +339,29 @@ export function FindingCaseReviewPanel({ taskId, onChange }: { taskId?: string;
       active = false;
       clearTimeout(timer);
     };
-  }, [taskId]);
-  async function resolve(s: FindingCaseSuggestion, accept: boolean) {
+  }, [reload, taskId, taskKey]);
+  async function resolve(s: FindingCaseSuggestion, accept: boolean, excludeIds?: string[]) {
+    const seq = ++requestSeq.current;
+    const requestedTask = taskKey;
     try {
-      const result = await api.resolveFindingCaseSuggestion(s.id, accept);
-      setSuggestions((v) => v.filter((x) => x.id !== s.id));
-      if (accept && result.case_id !== "0") {
-        toast.success("已归并，请进入文件夹生成统一报告");
-      } else {
-        toast.success("已否决，不会自动再次归并");
+      const result = await api.resolveFindingCaseSuggestion(s.id, accept, excludeIds);
+      if (requestedTask === taskKeyRef.current) {
+        setReload((v) => v + 1);
+        if (seq === requestSeq.current) {
+          setSuggestions((v) => v.filter((x) => x.id !== s.id));
+          if (accept && result.case_id !== "0") {
+            toast.success("已归并，请进入文件夹生成统一报告");
+          } else {
+            toast.success("已否决，不会自动再次归并");
+          }
+          onChange?.();
+        }
       }
-      onChange?.();
     } catch (e) {
-      toast.error((e as Error).message);
+      if (requestedTask === taskKeyRef.current) {
+        setReload((v) => v + 1);
+        if (seq === requestSeq.current) toast.error((e as Error).message);
+      }
     }
   }
   return (

web/src/components/finding-repeater-dialog.tsx

@@ -0,0 +1,336 @@
+"use client";
+
+import * as React from "react";
+
+import { SendIcon } from "lucide-react";
+import { toast } from "sonner";
+
+import { Alert, AlertDescription } from "@/components/ui/alert";
+import { Badge } from "@/components/ui/badge";
+import { Button } from "@/components/ui/button";
+import { Checkbox } from "@/components/ui/checkbox";
+import { Dialog, DialogContent, DialogDescription, DialogHeader, DialogTitle } from "@/components/ui/dialog";
+import { Input } from "@/components/ui/input";
+import { Label } from "@/components/ui/label";
+import { Select, SelectContent, SelectGroup, SelectItem, SelectTrigger, SelectValue } from "@/components/ui/select";
+import { Textarea } from "@/components/ui/textarea";
+import { api } from "@/lib/api";
+import type { RepeaterSendResponse } from "@/lib/types";
+
+const METHODS = ["GET", "POST", "PUT", "PATCH", "DELETE", "HEAD", "OPTIONS"] as const;
+
+export type RepeaterInitial = {
+  method?: string;
+  url?: string;
+  headersText?: string;
+  body?: string;
+};
+
+type HistoryEntry = {
+  id: number;
+  method: string;
+  url: string;
+  headersText: string;
+  body: string;
+  useProxy: boolean;
+  insecure: boolean;
+  response: RepeaterSendResponse;
+};
+
+export function parseHeadersText(text: string): Record<string, string> {
+  const result: Record<string, string> = {};
+  for (const line of text.replace(/\r\n?/g, "\n").split("\n")) {
+    const separator = line.indexOf(":");
+    if (separator <= 0) continue;
+    const name = line.slice(0, separator).trim();
+    const value = line.slice(separator + 1).trim();
+    if (!name) continue;
+    // The API accepts one value per name. Cookies use semicolon separators;
+    // other duplicate headers preserve the common comma-separated semantics.
+    const existingName = Object.keys(result).find((item) => item.toLowerCase() === name.toLowerCase());
+    if (!existingName) {
+      result[name] = value;
+    } else {
+      const separator = existingName.toLowerCase() === "cookie" ? "; " : ", ";
+      result[existingName] = `${result[existingName]}${separator}${value}`;
+    }
+  }
+  return result;
+}
+
+export function formatHeaders(headers: Record<string, string[]> | Record<string, string>): string {
+  return Object.entries(headers)
+    .flatMap(([name, value]) =>
+      Array.isArray(value) ? value.map((item) => `${name}: ${item}`) : [`${name}: ${value}`],
+    )
+    .join("\n");
+}
+
+export function buildRawResponse(response: RepeaterSendResponse): string {
+  if (!response.status) return "";
+  const statusLine = `${response.proto || "HTTP/1.1"} ${response.status_text || response.status}`.trim();
+  const headers = formatHeaders(response.headers);
+  return `${statusLine}\n${headers}\n\n${response.body}`;
+}
+
+export function headersFromReqHead(reqHead?: string): string {
+  if (!reqHead) return "";
+  const lines = reqHead.replace(/\r\n?/g, "\n").split("\n");
+  // Evidence heads normally start with `METHOD path HTTP/x.y`, but tolerate
+  // leading blank lines and CR-only captures from older traffic records.
+  let requestLine = -1;
+  for (let index = 0; index < lines.length; index += 1) {
+    const candidate = lines[index].trim();
+    if (!candidate) continue;
+    if (/^(?:[A-Z]+)\s+\S+\s+HTTP\/\d(?:\.\d)?$/i.test(candidate)) requestLine = index;
+    break;
+  }
+  const start = requestLine >= 0 ? requestLine + 1 : 0;
+  return lines.slice(start).join("\n").replace(/\n+$/, "");
+}
+
+export function FindingRepeaterDialog({
+  open,
+  onOpenChange,
+  initial,
+  findingLabel,
+}: {
+  open: boolean;
+  onOpenChange: (open: boolean) => void;
+  initial?: RepeaterInitial;
+  findingLabel?: string;
+}) {
+  const [method, setMethod] = React.useState("GET");
+  const [url, setURL] = React.useState("");
+  const [headersText, setHeadersText] = React.useState("");
+  const [body, setBody] = React.useState("");
+  const [useProxy, setUseProxy] = React.useState(true);
+  const [insecure, setInsecure] = React.useState(false);
+  const [busy, setBusy] = React.useState(false);
+  const [response, setResponse] = React.useState<RepeaterSendResponse | null>(null);
+  const [history, setHistory] = React.useState<HistoryEntry[]>([]);
+  const sequence = React.useRef(0);
+  const generation = React.useRef(0);
+  const sending = React.useRef(false);
+
+  React.useEffect(() => {
+    sending.current = false;
+    setBusy(false);
+    if (!open) return;
+    const nextMethod = (initial?.method ?? "GET").toUpperCase();
+    setMethod(METHODS.includes(nextMethod as (typeof METHODS)[number]) ? nextMethod : "GET");
+    setURL(initial?.url ?? "");
+    setHeadersText(initial?.headersText ?? "");
+    setBody(initial?.body ?? "");
+    setUseProxy(true);
+    setInsecure(false);
+    setResponse(null);
+    return () => {
+      // Closing, reseeding, or unmounting makes any pending result obsolete.
+      generation.current += 1;
+      sending.current = false;
+    };
+  }, [open, initial?.method, initial?.url, initial?.headersText, initial?.body]);
+
+  async function send() {
+    // State alone cannot stop a second click before React disables the button.
+    if (sending.current || !open) return;
+    const target = url.trim();
+    if (!target) {
+      toast.error("请填写 URL");
+      return;
+    }
+    const requestGeneration = ++generation.current;
+    sending.current = true;
+    setBusy(true);
+    setResponse(null);
+    try {
+      const result = await api.repeaterSend({
+        method,
+        url: target,
+        headers: parseHeadersText(headersText),
+        body,
+        use_recording_proxy: useProxy,
+        insecure_skip_verify: insecure,
+      });
+      if (requestGeneration !== generation.current) return;
+      setResponse(result);
+      const id = ++sequence.current;
+      setHistory((previous) =>
+        [
+          { id, method, url: target, headersText, body, useProxy, insecure, response: result },
+          ...previous,
+        ].slice(0, 20),
+      );
+    } catch (error) {
+      if (requestGeneration === generation.current) toast.error((error as Error).message);
+    } finally {
+      if (requestGeneration === generation.current) {
+        sending.current = false;
+        setBusy(false);
+      }
+    }
+  }
+
+  return (
+    <Dialog
+      open={open}
+      onOpenChange={(nextOpen) => {
+        if (!nextOpen) {
+          // Invalidate immediately, before the parent's close render/effect.
+          generation.current += 1;
+          sending.current = false;
+          setBusy(false);
+        }
+        onOpenChange(nextOpen);
+      }}
+    >
+      <DialogContent className="flex max-h-[92vh] w-[min(1150px,96vw)] max-w-none flex-col gap-3 overflow-hidden sm:max-w-none">
+        <DialogHeader>
+          <DialogTitle>Repeater · 请求重放</DialogTitle>
+          <DialogDescription>
+            手工编辑并发送 HTTP 请求，便于验证漏洞与撰写报告{findingLabel ? ` · ${findingLabel}` : ""}
+          </DialogDescription>
+        </DialogHeader>
+        <div className="flex flex-wrap items-end gap-2">
+          <div className="w-28">
+            <Label className="mb-1 block text-xs text-muted-foreground">方法</Label>
+            <Select value={method} onValueChange={setMethod} disabled={busy}>
+              <SelectTrigger>
+                <SelectValue />
+              </SelectTrigger>
+              <SelectContent>
+                <SelectGroup>
+                  {METHODS.map((item) => (
+                    <SelectItem key={item} value={item}>
+                      {item}
+                    </SelectItem>
+                  ))}
+                </SelectGroup>
+              </SelectContent>
+            </Select>
+          </div>
+          <div className="min-w-[240px] flex-1">
+            <Label className="mb-1 block text-xs text-muted-foreground">URL</Label>
+            <Input
+              disabled={busy}
+              value={url}
+              onChange={(event) => setURL(event.target.value)}
+              placeholder="https://target.example/path"
+              spellCheck={false}
+            />
+          </div>
+          <Button disabled={busy} onClick={() => void send()}>
+            <SendIcon data-icon="inline-start" />
+            发送
+          </Button>
+        </div>
+        <div className="flex flex-wrap items-center gap-4 text-sm">
+          <Label className="flex items-center gap-2" htmlFor="repeater-use-proxy">
+            <Checkbox
+              id="repeater-use-proxy"
+              disabled={busy}
+              checked={useProxy}
+              onCheckedChange={(value) => setUseProxy(value === true)}
+            />
+            经流量代理
+          </Label>
+          <Label className="flex items-center gap-2" htmlFor="repeater-insecure">
+            <Checkbox
+              id="repeater-insecure"
+              disabled={busy}
+              checked={insecure}
+              onCheckedChange={(value) => setInsecure(value === true)}
+            />
+            跳过 TLS 校验
+          </Label>
+          {response ? (
+            <div className="flex items-center gap-2 text-xs text-muted-foreground">
+              {response.status ? <Badge variant="secondary">{response.status_text || response.status}</Badge> : null}
+              <span>{response.elapsed_ms} ms</span>
+              <span>{response.body_bytes.toLocaleString()} 字节</span>
+              {response.body_truncated ? <Badge variant="outline">正文已截断</Badge> : null}
+            </div>
+          ) : null}
+        </div>
+        <div className="grid min-h-0 flex-1 gap-3 overflow-hidden lg:grid-cols-[180px_1fr_1fr]">
+          <div className="flex min-h-0 flex-col gap-2 overflow-hidden rounded-md border p-2">
+            <div className="text-xs font-medium text-muted-foreground">会话历史</div>
+            <div className="min-h-0 flex-1 space-y-1 overflow-y-auto">
+              {history.length === 0 ? (
+                <p className="text-xs text-muted-foreground">发送后会出现在这里</p>
+              ) : (
+                history.map((entry) => (
+                  <button
+                    key={entry.id}
+                    type="button"
+                    disabled={busy}
+                    className="block w-full rounded-md px-2 py-1.5 text-left hover:bg-muted"
+                    onClick={() => {
+                      setMethod(entry.method);
+                      setURL(entry.url);
+                      setHeadersText(entry.headersText);
+                      setBody(entry.body);
+                      setUseProxy(entry.useProxy);
+                      setInsecure(entry.insecure);
+                      setResponse(entry.response);
+                    }}
+                  >
+                    <div className="truncate font-mono text-xs">
+                      {entry.method} {entry.response.status || "ERR"}
+                    </div>
+                    <div className="truncate text-[11px] text-muted-foreground">{entry.url}</div>
+                  </button>
+                ))
+              )}
+            </div>
+          </div>
+          <div className="flex min-h-0 flex-col gap-2 overflow-hidden">
+            <div className="text-xs font-medium">请求</div>
+            <div className="flex min-h-0 flex-1 flex-col gap-2">
+              <div className="flex min-h-0 flex-1 flex-col gap-1">
+                <Label className="text-xs text-muted-foreground">请求头</Label>
+                <Textarea
+                  className="min-h-[120px] flex-1 font-mono text-xs"
+                  disabled={busy}
+                  value={headersText}
+                  onChange={(event) => setHeadersText(event.target.value)}
+                  placeholder="Header: value"
+                  spellCheck={false}
+                />
+              </div>
+              <div className="flex min-h-0 flex-1 flex-col gap-1">
+                <Label className="text-xs text-muted-foreground">请求体</Label>
+                <Textarea
+                  className="min-h-[120px] flex-1 font-mono text-xs"
+                  disabled={busy}
+                  value={body}
+                  onChange={(event) => setBody(event.target.value)}
+                  placeholder="请求正文（可选）"
+                  spellCheck={false}
+                />
+              </div>
+            </div>
+          </div>
+          <div className="flex min-h-0 flex-col gap-2 overflow-hidden">
+            <div className="text-xs font-medium">响应</div>
+            {response?.error ? (
+              <Alert variant={response.status ? "default" : "destructive"}>
+                <AlertDescription className="break-all font-mono text-xs">{response.error}</AlertDescription>
+              </Alert>
+            ) : null}
+            <div className="min-h-0 flex-1 overflow-auto rounded-md border">
+              {response ? (
+                <pre className="whitespace-pre-wrap break-words p-3 font-mono text-xs">
+                  {buildRawResponse(response)}
+                </pre>
+              ) : (
+                <p className="p-3 text-xs text-muted-foreground">发送后显示原始响应</p>
+              )}
+            </div>
+          </div>
+        </div>
+      </DialogContent>
+    </Dialog>
+  );
+}

web/src/components/finding-traffic-panel.tsx

@@ -5,6 +5,7 @@ import * as React from "react";
 import { ArrowDownIcon, ArrowUpIcon, PlusIcon } from "lucide-react";
 import { toast } from "sonner";
 
+import { FindingRepeaterDialog, headersFromReqHead, type RepeaterInitial } from "@/components/finding-repeater-dialog";
 import { TrafficEvidenceViewer } from "@/components/traffic-evidence-viewer";
 import { TrafficPickerDialog } from "@/components/traffic-picker-dialog";
 import { Alert, AlertDescription } from "@/components/ui/alert";
@@ -49,6 +50,9 @@ export function FindingTrafficPanel({
   const [error, setError] = React.useState("");
   const [adding, setAdding] = React.useState(false);
   const [preview, setPreview] = React.useState<string | null>(null);
+  const [repeaterOpen, setRepeaterOpen] = React.useState(false);
+  const [repeaterInitial, setRepeaterInitial] = React.useState<RepeaterInitial>();
+  const [repeaterBusy, setRepeaterBusy] = React.useState(false);
   const [editing, setEditing] = React.useState<FindingTrafficBinding | null>(null);
   const [role, setRole] = React.useState<TrafficEvidenceRole>("supporting");
   const [note, setNote] = React.useState("");
@@ -94,20 +98,52 @@ export function FindingTrafficPanel({
     void mutate(() => api.orderFindingTraffic(findingId, ids, data.version, contextTask));
   }
 
+  async function seedRepeater(bindingId: string) {
+    setRepeaterBusy(true);
+    try {
+      const detail = await api.findingTrafficDetail(findingId, bindingId, contextTask);
+      if (detail.request.binary) toast.info("请求体为二进制，已留空；完整正文请先下载");
+      else if (detail.request.truncated) toast.info("请求体为预览片段，完整正文请先下载");
+      setRepeaterInitial({
+        method: detail.binding.snapshot.method || "GET",
+        url: detail.binding.snapshot.url || "",
+        headersText: headersFromReqHead(detail.binding.snapshot.req_head),
+        body: detail.request.binary ? "" : detail.request.content || "",
+      });
+      setRepeaterOpen(true);
+    } catch (e) {
+      toast.error((e as Error).message);
+    } finally {
+      setRepeaterBusy(false);
+    }
+  }
+
   return (
     <>
       <Card>
         <CardHeader>
           <div className="flex flex-wrap items-center justify-between gap-2">
             <CardTitle>关联流量 {data ? `(${data.bindings.length})` : ""}</CardTitle>
-            {!readOnly ? (
-              <Button variant="outline" size="sm" disabled={!data || busy} onClick={() => setAdding(true)}>
-                <PlusIcon data-icon="inline-start" />
-                绑定流量
+            <div className="flex flex-wrap gap-2">
+              <Button
+                variant="outline"
+                size="sm"
+                onClick={() => {
+                  setRepeaterInitial({ method: "GET", url: "", headersText: "", body: "" });
+                  setRepeaterOpen(true);
+                }}
+              >
+                打开 Repeater
               </Button>
-            ) : (
-              <Badge variant="outline">继承证据 · 只读</Badge>
-            )}
+              {!readOnly ? (
+                <Button variant="outline" size="sm" disabled={!data || busy} onClick={() => setAdding(true)}>
+                  <PlusIcon data-icon="inline-start" />
+                  绑定流量
+                </Button>
+              ) : (
+                <Badge variant="outline">继承证据 · 只读</Badge>
+              )}
+            </div>
           </div>
           <CardDescription>按复现顺序组织请求与响应。清理原始流量后，已绑定证据仍然保留。</CardDescription>
         </CardHeader>
@@ -159,6 +195,9 @@ export function FindingTrafficPanel({
                     <Button variant="outline" size="sm" onClick={() => setPreview(b.id)}>
                       查看报文
                     </Button>
+                    <Button variant="outline" size="sm" disabled={repeaterBusy} onClick={() => void seedRepeater(b.id)}>
+                      送入 Repeater
+                    </Button>
                     {!readOnly ? (
                       <>
                         <Button
@@ -228,6 +267,12 @@ export function FindingTrafficPanel({
         contextTask={contextTask}
         onClose={() => setPreview(null)}
       />
+      <FindingRepeaterDialog
+        open={repeaterOpen}
+        onOpenChange={setRepeaterOpen}
+        initial={repeaterInitial}
+        findingLabel={`发现 #${findingId}`}
+      />
       <Dialog
         open={editing !== null}
         onOpenChange={(open) => {

web/src/components/intelligence-panel.tsx

@@ -193,8 +193,8 @@ export function IntelligencePanel({
           </Tabs>
           <div className="flex flex-wrap items-center gap-2">
             <Input
-              aria-label="搜索情报标题或上报者"
-              placeholder="搜索标题或上报者"
+              aria-label="搜索情报标题、内容或上报者"
+              placeholder="搜索标题、内容或上报者"
               value={search}
               maxLength={200}
               onChange={(e) => setSearch(e.target.value)}

web/src/components/remark-editor.tsx

@@ -8,6 +8,7 @@ import { toast } from "sonner";
 import { Button } from "@/components/ui/button";
 import { Input } from "@/components/ui/input";
 import { Textarea } from "@/components/ui/textarea";
+import { MAX_TASK_REMARK_RUNES } from "@/lib/types";
 import { cn } from "@/lib/utils";
 
 type EditorProps = {
@@ -17,7 +18,7 @@ type EditorProps = {
 };
 
 export function RemarkEditor(props: EditorProps) {
-  return <InlineTextEditor {...props} multiline fieldName="备注" maxCharacters={2000} emptyLabel="添加备注" />;
+  return <InlineTextEditor {...props} multiline fieldName="备注" maxCharacters={MAX_TASK_REMARK_RUNES} emptyLabel="添加备注" />;
 }
 
 export function ScheduleNameEditor(props: EditorProps) {

web/src/components/task-template-controls.tsx

@@ -33,19 +33,28 @@ import { Sheet, SheetContent, SheetDescription, SheetFooter, SheetHeader, SheetT
 import { Spinner } from "@/components/ui/spinner";
 import { Textarea } from "@/components/ui/textarea";
 import { api } from "@/lib/api";
-import type { AssetInterceptRuleInput, TaskCategory, TaskTemplate } from "@/lib/types";
+import {
+  MAX_TASK_REMARK_RUNES,
+  MAX_TASK_WORKER_COUNT,
+  parseWorkerCountInput,
+  type AssetInterceptRuleInput,
+  type TaskCategory,
+  type TaskTemplate,
+} from "@/lib/types";
 import { cn } from "@/lib/utils";
 
 interface TemplateDraft {
   name: string;
   description: string;
   goal: string;
+  remark: string;
   categoryID: number | null;
   interceptRules: AssetInterceptRuleInput[];
+  workerCount: string; // 输入框用字符串;空/0=用全局默认
 }
 
 // TemplateSeed 是「另存为模板」时从创建表单带入的初值。
-type TemplateSeed = Pick<TemplateDraft, "description" | "goal" | "categoryID" | "interceptRules">;
+type TemplateSeed = Pick<TemplateDraft, "description" | "goal" | "remark" | "categoryID" | "interceptRules" | "workerCount">;
 
 interface TaskTemplateManagerProps {
   open: boolean;
@@ -61,8 +70,10 @@ const emptyDraft = (): TemplateDraft => ({
   name: "",
   description: "",
   goal: "",
+  remark: "",
   categoryID: null,
   interceptRules: [],
+  workerCount: "",
 });
 
 function templateDraft(template: TaskTemplate): TemplateDraft {
@@ -70,8 +81,10 @@ function templateDraft(template: TaskTemplate): TemplateDraft {
     name: template.name,
     description: template.description,
     goal: template.goal,
+    remark: template.remark ?? "",
     categoryID: template.category_id ?? null,
     interceptRules: template.intercept_rules ?? [],
+    workerCount: template.worker_count ? String(template.worker_count) : "",
   };
 }
 
@@ -108,8 +121,10 @@ function TaskTemplateManager({
           name: "",
           description: seed.description,
           goal: seed.goal,
+          remark: seed.remark,
           categoryID: seed.categoryID,
           interceptRules: seed.interceptRules,
+          workerCount: seed.workerCount,
         });
       } else if (templates[0]) {
         setSelectedID(templates[0].id);
@@ -132,7 +147,7 @@ function TaskTemplateManager({
     setDraft(emptyDraft());
   };
 
-  const updateDraft = (field: "name" | "description" | "goal", value: string) => {
+  const updateDraft = (field: "name" | "description" | "goal" | "remark" | "workerCount", value: string) => {
     setDraft((current) => ({ ...current, [field]: value }));
   };
 
@@ -141,11 +156,18 @@ function TaskTemplateManager({
   };
 
   async function save() {
+    const workerCount = parseWorkerCountInput(draft.workerCount);
+    if (workerCount == null) {
+      toast.error(`Worker 运行数量必须是 0-${MAX_TASK_WORKER_COUNT} 的整数`);
+      return;
+    }
     const input = {
       name: draft.name.trim(),
       description: draft.description.trim(),
       goal: draft.goal.trim(),
+      remark: draft.remark.trim(),
       category_id: draft.categoryID,
+      worker_count: workerCount,
       intercept_rules: draft.interceptRules
         .map((r) => ({ ...r, pattern: r.pattern.trim() }))
         .filter((r) => r.pattern !== ""),
@@ -207,7 +229,7 @@ function TaskTemplateManager({
         <SheetContent className="grid h-full w-full! max-w-none! grid-rows-[auto_minmax(0,1fr)_auto] gap-0 overflow-hidden p-0 sm:w-[48rem]! sm:max-w-[48rem]!">
           <SheetHeader className="border-b px-6 py-5">
             <SheetTitle>任务模板管理</SheetTitle>
-            <SheetDescription>模板保存描述、目标、分类与任务级拦截/允许规则；修改不会影响已经创建的任务。</SheetDescription>
+            <SheetDescription>模板保存描述、目标、备注、Worker 数量、分类与任务级拦截/允许规则；修改不会影响已经创建的任务。</SheetDescription>
           </SheetHeader>
           <div className="grid min-h-0 overflow-y-auto lg:grid-cols-[15rem_minmax(0,1fr)] lg:overflow-hidden">
             <div className="flex min-h-0 flex-col border-b p-3 lg:border-r lg:border-b-0">
@@ -232,6 +254,9 @@ function TaskTemplateManager({
                     >
                       <span className="block truncate font-medium text-sm">{template.name}</span>
                       <span className="block truncate text-muted-foreground text-xs">{template.description}</span>
+                      {template.remark ? (
+                        <span className="block truncate text-muted-foreground/80 text-xs">备注：{template.remark}</span>
+                      ) : null}
                     </button>
                   ))}
                 </div>
@@ -269,6 +294,31 @@ function TaskTemplateManager({
                     onChange={(event) => updateDraft("goal", event.target.value)}
                   />
                 </Field>
+                <Field>
+                  <FieldLabel htmlFor="task-template-remark">备注（可选）</FieldLabel>
+                  <Textarea
+                    id="task-template-remark"
+                    className="min-h-20"
+                    maxLength={MAX_TASK_REMARK_RUNES}
+                    value={draft.remark}
+                    placeholder="预设备注，应用模板时写入新任务"
+                    onChange={(event) => updateDraft("remark", event.target.value)}
+                  />
+                </Field>
+                <Field>
+                  <FieldLabel htmlFor="task-template-worker-count">Worker 运行数量（可选）</FieldLabel>
+                  <Input
+                    id="task-template-worker-count"
+                    type="number"
+                    min={0}
+                    max={MAX_TASK_WORKER_COUNT}
+                    className="w-40"
+                    placeholder="留空 = 全局默认"
+                    value={draft.workerCount}
+                    onChange={(event) => updateDraft("workerCount", event.target.value)}
+                  />
+                  <FieldDescription>应用模板时预填本任务 Worker 数；留空或 0 = 用全局默认。</FieldDescription>
+                </Field>
                 <Field>
                   <FieldLabel htmlFor="task-template-category">任务分类</FieldLabel>
                   <NativeSelect
@@ -347,8 +397,10 @@ function TaskTemplateManager({
 interface TaskTemplateControlsProps {
   description: string;
   goal: string;
+  remark?: string;
   categoryID?: number;
   interceptRules?: AssetInterceptRuleInput[];
+  workerCount?: string;
   selectedTemplateID: number | null;
   onSelectedTemplateIDChange: (id: number | null) => void;
   onApply: (template: TaskTemplate) => void;
@@ -358,8 +410,10 @@ interface TaskTemplateControlsProps {
 export function TaskTemplateControls({
   description,
   goal,
+  remark = "",
   categoryID,
   interceptRules,
+  workerCount = "",
   selectedTemplateID,
   onSelectedTemplateIDChange,
   onApply,
@@ -410,8 +464,18 @@ export function TaskTemplateControls({
       return;
     }
     setTemplateInputValue(template.name);
-    const hasContent = description.trim() !== "" || goal.trim() !== "";
-    const changesContent = description !== template.description || goal !== template.goal;
+    const currentWorkerCount = parseWorkerCountInput(workerCount);
+    const templateWorkerCount = template.worker_count ?? 0;
+    const hasContent =
+      description.trim() !== "" ||
+      goal.trim() !== "" ||
+      remark.trim() !== "" ||
+      (currentWorkerCount == null ? workerCount.trim() !== "" : currentWorkerCount !== 0);
+    const changesContent =
+      description !== template.description ||
+      goal !== template.goal ||
+      remark !== (template.remark ?? "") ||
+      (currentWorkerCount == null ? workerCount.trim() !== "" : currentWorkerCount !== templateWorkerCount);
     if (hasContent && changesContent) {
       setPendingTemplate(template);
       return;
@@ -455,8 +519,10 @@ export function TaskTemplateControls({
                 openManager({
                   description,
                   goal,
+                  remark,
                   categoryID: categoryID ?? null,
                   interceptRules: interceptRules ?? [],
+                  workerCount,
                 })
               }
             >
@@ -498,7 +564,9 @@ export function TaskTemplateControls({
             </ComboboxList>
           </ComboboxContent>
         </Combobox>
-        <FieldDescription>选择后会复制模板的描述、目标、分类与任务级规则，不与模板保持关联。</FieldDescription>
+        <FieldDescription>
+          选择后会复制模板的描述、目标、备注、Worker 数量、分类与任务级规则，不与模板保持关联。
+        </FieldDescription>
       </Field>
 
       <AlertDialog
@@ -512,7 +580,7 @@ export function TaskTemplateControls({
         <AlertDialogContent>
           <AlertDialogHeader>
             <AlertDialogTitle>使用模板「{pendingTemplate?.name}」？</AlertDialogTitle>
-            <AlertDialogDescription>当前已填写的描述和目标将被模板内容覆盖。</AlertDialogDescription>
+            <AlertDialogDescription>当前已填写的描述、目标、备注和 Worker 数量将被模板内容覆盖。</AlertDialogDescription>
           </AlertDialogHeader>
           <AlertDialogFooter>
             <AlertDialogCancel>取消</AlertDialogCancel>

web/src/components/traffic-evidence-viewer.tsx

@@ -4,6 +4,7 @@ import * as React from "react";
 
 import { toast } from "sonner";
 
+import { FindingRepeaterDialog, headersFromReqHead, type RepeaterInitial } from "@/components/finding-repeater-dialog";
 import { HttpCodeBlock } from "@/components/http-code-block";
 import { Alert, AlertDescription } from "@/components/ui/alert";
 import { Button } from "@/components/ui/button";
@@ -27,6 +28,8 @@ export function TrafficEvidenceViewer({
   const [detail, setDetail] = React.useState<FindingTrafficDetail | null>(null);
   const [error, setError] = React.useState("");
   const [busy, setBusy] = React.useState(false);
+  const [repeaterOpen, setRepeaterOpen] = React.useState(false);
+  const [repeaterInitial, setRepeaterInitial] = React.useState<RepeaterInitial>();
   React.useEffect(() => {
     if (!bindingId) return;
     let active = true;
@@ -62,64 +65,94 @@ export function TrafficEvidenceViewer({
     }
   }
 
+  function openRepeater() {
+    if (!detail) return;
+    if (detail.request.binary) toast.info("请求体为二进制，已留空；完整正文请先下载");
+    else if (detail.request.truncated) toast.info("请求体为预览片段，完整正文请先下载");
+    setRepeaterInitial({
+      method: detail.binding.snapshot.method || "GET",
+      url: detail.binding.snapshot.url || "",
+      headersText: headersFromReqHead(detail.binding.snapshot.req_head),
+      body: detail.request.binary ? "" : detail.request.content || "",
+    });
+    setRepeaterOpen(true);
+  }
+
   return (
-    <Dialog
-      open={bindingId !== null}
-      onOpenChange={(open) => {
-        if (!open) onClose();
-      }}
-    >
-      <DialogContent className="max-h-[90vh] overflow-y-auto sm:max-w-4xl">
-        <DialogHeader>
-          <DialogTitle>流量证据 #{bindingId}</DialogTitle>
-          <DialogDescription className="break-all">
-            {detail?.binding.snapshot.url ?? "查看绑定时保存的请求与响应"}
-          </DialogDescription>
-        </DialogHeader>
-        {error ? (
-          <Alert variant="destructive">
-            <AlertDescription>{error}</AlertDescription>
-          </Alert>
-        ) : detail ? (
-          <Tabs defaultValue="request">
-            <TabsList>
-              <TabsTrigger value="request">请求 Request</TabsTrigger>
-              <TabsTrigger value="response">响应 Response</TabsTrigger>
-            </TabsList>
-            {(["request", "response"] as const).map((side) => (
-              <TabsContent key={side} value={side}>
-                <div className="flex flex-wrap items-center justify-between gap-2">
-                  <span className="text-xs text-muted-foreground">
-                    正文 {detail[side].total.toLocaleString()} 字节{detail[side].truncated ? " · 当前为预览" : ""}
-                  </span>
-                  <Button
-                    variant="outline"
-                    size="sm"
-                    onClick={() =>
-                      void api
-                        .downloadFindingTrafficBody(findingId, detail.binding.id, side, contextTask)
-                        .catch((e: Error) => toast.error(e.message))
-                    }
-                  >
-                    下载完整{side === "request" ? "请求" : "响应"}正文
-                  </Button>
-                </div>
-                <HttpCodeBlock
-                  raw={`${(side === "request" ? detail.binding.snapshot.req_head : detail.binding.snapshot.resp_head)?.trimEnd() ?? ""}\n\n${detail[side].content}`}
-                />
-                {detail[side].truncated && !detail[side].binary ? (
-                  <Button variant="outline" size="sm" disabled={busy} onClick={() => void more(side)}>
-                    加载更多正文
-                  </Button>
-                ) : null}
-              </TabsContent>
-            ))}
-          </Tabs>
-        ) : (
-          <Skeleton className="h-56 w-full" />
-        )}
-      </DialogContent>
-    </Dialog>
+    <>
+      <Dialog
+        open={bindingId !== null}
+        onOpenChange={(open) => {
+          if (!open) onClose();
+        }}
+      >
+        <DialogContent className="max-h-[90vh] overflow-y-auto sm:max-w-4xl">
+          <DialogHeader>
+            <div className="flex flex-wrap items-start justify-between gap-2 pe-8">
+              <div>
+                <DialogTitle>流量证据 #{bindingId}</DialogTitle>
+                <DialogDescription className="break-all">
+                  {detail?.binding.snapshot.url ?? "查看绑定时保存的请求与响应"}
+                </DialogDescription>
+              </div>
+              {detail ? (
+                <Button variant="outline" size="sm" onClick={openRepeater}>
+                  送入 Repeater
+                </Button>
+              ) : null}
+            </div>
+          </DialogHeader>
+          {error ? (
+            <Alert variant="destructive">
+              <AlertDescription>{error}</AlertDescription>
+            </Alert>
+          ) : detail ? (
+            <Tabs defaultValue="request">
+              <TabsList>
+                <TabsTrigger value="request">请求 Request</TabsTrigger>
+                <TabsTrigger value="response">响应 Response</TabsTrigger>
+              </TabsList>
+              {(["request", "response"] as const).map((side) => (
+                <TabsContent key={side} value={side}>
+                  <div className="flex flex-wrap items-center justify-between gap-2">
+                    <span className="text-xs text-muted-foreground">
+                      正文 {detail[side].total.toLocaleString()} 字节{detail[side].truncated ? " · 当前为预览" : ""}
+                    </span>
+                    <Button
+                      variant="outline"
+                      size="sm"
+                      onClick={() =>
+                        void api
+                          .downloadFindingTrafficBody(findingId, detail.binding.id, side, contextTask)
+                          .catch((e: Error) => toast.error(e.message))
+                      }
+                    >
+                      下载完整{side === "request" ? "请求" : "响应"}正文
+                    </Button>
+                  </div>
+                  <HttpCodeBlock
+                    raw={`${(side === "request" ? detail.binding.snapshot.req_head : detail.binding.snapshot.resp_head)?.trimEnd() ?? ""}\n\n${detail[side].content}`}
+                  />
+                  {detail[side].truncated && !detail[side].binary ? (
+                    <Button variant="outline" size="sm" disabled={busy} onClick={() => void more(side)}>
+                      加载更多正文
+                    </Button>
+                  ) : null}
+                </TabsContent>
+              ))}
+            </Tabs>
+          ) : (
+            <Skeleton className="h-56 w-full" />
+          )}
+        </DialogContent>
+      </Dialog>
+      <FindingRepeaterDialog
+        open={repeaterOpen}
+        onOpenChange={setRepeaterOpen}
+        initial={repeaterInitial}
+        findingLabel={bindingId ? `证据 #${bindingId}` : undefined}
+      />
+    </>
   );
 }
 

web/src/lib/api.ts

@@ -51,6 +51,8 @@ import type {
   FindingsPage,
   FindingTraffic,
   FindingTrafficDetail,
+  RepeaterSendRequest,
+  RepeaterSendResponse,
   IntelligencePage,
   IntentAsset,
   InterceptApprovalFilter,
@@ -300,6 +302,7 @@ export const api = {
   task: (id: string) => get<Task>(`/tasks/${encodeURIComponent(id)}`),
   createTask: (input: {
     name?: string;
+    remark?: string;
     categoryId?: number;
     description: string;
     goal: string;
@@ -310,11 +313,13 @@ export const api = {
     seedFirstIntent?: boolean;
     planHeartbeatSeconds?: number;
     coverageEnabled?: boolean;
+    workerCount?: number;
     interceptRules?: AssetInterceptRuleInput[];
     schedule?: ScheduleInput;
   }) =>
     post<Task>("/tasks", {
       name: input.name ?? "",
+      remark: input.remark ?? "",
       category_id: input.categoryId ?? null,
       description: input.description,
       goal: input.goal,
@@ -325,6 +330,7 @@ export const api = {
       seed_first_intent: input.seedFirstIntent ?? false,
       plan_heartbeat_seconds: input.planHeartbeatSeconds ?? 0, // 0 = 后端归一到默认 600(10min)
       coverage_enabled: input.coverageEnabled ?? true, // 默认开;false=关闭资产覆盖度功能
+      worker_count: input.workerCount ?? 0, // 0 = 用全局默认 workers
       intercept_rules: input.interceptRules ?? [], // 任务级资产拦截规则
       schedule: input.schedule ? schedulePayload(input.schedule) : undefined,
     }),
@@ -357,11 +363,11 @@ export const api = {
     }),
   taskTemplates: () => get<{ templates: TaskTemplate[] }>("/task-templates").then((r) => arr(r.templates)),
   createTaskTemplate: (
-    input: Pick<TaskTemplate, "name" | "description" | "goal" | "category_id" | "intercept_rules">,
+    input: Pick<TaskTemplate, "name" | "description" | "goal" | "remark" | "category_id" | "intercept_rules" | "worker_count">,
   ) => post<TaskTemplate>("/task-templates", input),
   updateTaskTemplate: (
     id: number,
-    input: Partial<Pick<TaskTemplate, "name" | "description" | "goal" | "category_id" | "intercept_rules">>,
+    input: Partial<Pick<TaskTemplate, "name" | "description" | "goal" | "remark" | "category_id" | "intercept_rules" | "worker_count">>,
   ) => patch<TaskTemplate>(`/task-templates/${id}`, input),
   deleteTaskTemplate: (id: number) => del<{ deleted: number }>(`/task-templates/${id}`),
   updateTaskLLMProfiles: (id: string, llmProfileIds: number[], activeLLMProfileId?: number) =>
@@ -645,8 +651,11 @@ export const api = {
     get<FindingCaseSuggestion[]>(
       `/exploration/finding-case-suggestions${task ? `?task_id=${encodeURIComponent(task)}` : ""}`,
     ),
-  resolveFindingCaseSuggestion: (id: string, accept: boolean) =>
-    post<{ case_id: string }>(`/exploration/finding-case-suggestions/${id}`, { accept }),
+  resolveFindingCaseSuggestion: (id: string, accept: boolean, excludeIds?: string[]) =>
+    post<{ case_id: string }>(`/exploration/finding-case-suggestions/${id}`, {
+      accept,
+      ...(excludeIds && excludeIds.length > 0 ? { exclude_ids: excludeIds } : {}),
+    }),
   reviewFindingCases: (ids: string[]) =>
     post<{ runs: { conversation_id: number; task_id: string }[] }>("/exploration/finding-case-review", {
       finding_ids: ids,
@@ -733,6 +742,7 @@ export const api = {
     get<FindingTrafficDetail>(
       `/exploration/findings/${id}/traffic/${bindingId}${contextTask ? `?context_task=${encodeURIComponent(contextTask)}` : ""}`,
     ),
+  repeaterSend: (req: RepeaterSendRequest) => post<RepeaterSendResponse>("/repeater/send", req),
   findingTrafficBody: (
     id: string,
     bindingId: string,

web/src/lib/mock/data.ts

@@ -3338,6 +3338,7 @@ export const settings: Settings = {
   global_proxy: "",
   python_interpreter: "/usr/bin/python3",
   workers: 3,
+  worker_pools: [],
   llm_pool_enabled: true,
   llm_pool_bind_fallback: false,
   noa_compaction: false,

web/src/lib/mock/handler.ts

@@ -33,6 +33,7 @@ import type {
   TrafficEvidenceRole,
   TrafficEvidenceSnapshot,
 } from "../types";
+import { MAX_TASK_REMARK_RUNES, MAX_TASK_WORKER_COUNT } from "../types";
 import * as D from "./data";
 
 const delay = (ms = 120) => new Promise((r) => setTimeout(r, ms));
@@ -998,6 +999,40 @@ function parseBody(body?: BodyInit | null): Record<string, unknown> {
   }
 }
 
+function mockRemark(value: unknown): string {
+  if (value === undefined || value === null) return "";
+  if (typeof value !== "string") throw new Error("备注必须是字符串");
+  const remark = value.trim();
+  if (Array.from(remark).length > MAX_TASK_REMARK_RUNES) {
+    throw new Error(`备注最多 ${MAX_TASK_REMARK_RUNES} 个字符`);
+  }
+  return remark;
+}
+
+function mockWorkerCount(value: unknown): number {
+  if (value === undefined || value === null) return 0;
+  if (typeof value !== "number" || !Number.isInteger(value)) {
+    throw new Error("Worker 运行数量必须是整数");
+  }
+  if (value < 0) throw new Error("Worker 运行数量不能为负数");
+  if (value > MAX_TASK_WORKER_COUNT) throw new Error(`Worker 运行数量最多 ${MAX_TASK_WORKER_COUNT}`);
+  return value;
+}
+
+function mockStringField(value: unknown, label: string): string {
+  if (value === undefined || value === null) return "";
+  if (typeof value !== "string") throw new Error(`${label} 必须是字符串`);
+  return value;
+}
+
+function mockIntegerField(value: unknown, label: string): number | null {
+  if (value === undefined || value === null) return null;
+  if (typeof value !== "number" || !Number.isInteger(value)) {
+    throw new Error(`${label} 必须是整数`);
+  }
+  return value;
+}
+
 export async function mockHandle<T>(method: string, rawPath: string, body?: BodyInit | null): Promise<T> {
   advanceMockRetests();
   await delay();
@@ -1012,6 +1047,24 @@ export async function mockHandle<T>(method: string, rawPath: string, body?: Body
 function route(m: string, path: string, seg: string[], q: URLSearchParams, b: Record<string, unknown>): unknown {
   const task = q.get("task") ?? undefined;
 
+  // The demo client still expects the complete repeater response shape. No
+  // network request is made in mock mode, but returning the shape keeps the
+  // dialog's status/body counters safe for every method.
+  if (path === "/repeater/send" && m === "POST") {
+    const responseBody = "（演示模式）未发送真实请求。";
+    return {
+      status: 200,
+      status_text: "200 OK",
+      proto: "HTTP/1.1",
+      headers: { "Content-Type": ["text/plain; charset=utf-8"] },
+      body: responseBody,
+      body_truncated: false,
+      body_bytes: new TextEncoder().encode(responseBody).length,
+      elapsed_ms: 0,
+      url: String(b.url ?? ""),
+    };
+  }
+
   if (path === "/chat/mentions" && m === "GET") {
     const kind = q.get("kind") ?? "";
     const query = (q.get("q") ?? "").trim().toLowerCase();
@@ -1173,19 +1226,24 @@ function route(m: string, path: string, seg: string[], q: URLSearchParams, b: Re
     const profileIDs = [...((b.llm_profile_ids as number[] | undefined) ?? [])];
     const sourceTaskIDs = [...((b.source_task_ids as string[] | undefined) ?? [])];
     const companyIDs = [...new Set((b.company_ids as number[] | undefined) ?? [])];
+    const name = mockStringField(b.name, "任务名称");
+    const description = mockStringField(b.description, "任务描述");
+    const goal = mockStringField(b.goal, "任务目标");
     if (companyIDs.some((companyID) => !mockCompanies.some((company) => company.id === companyID))) {
       throw new Error("关联企业不存在或无效");
     }
-    const categoryID = typeof b.category_id === "number" ? b.category_id : undefined;
+    const categoryID = mockIntegerField(b.category_id, "category_id") ?? undefined;
     const category = categoryID === undefined ? undefined : mockTaskCategories.find((item) => item.id === categoryID);
     if (categoryID !== undefined && !category) throw new Error("任务分类不存在");
     const created: Task = {
       id,
-      name: String(b.name ?? ""),
+      name,
+      remark: mockRemark(b.remark),
       category_id: category?.id,
       category_name: category?.name,
-      description: String(b.description ?? "新任务"),
-      goal: String(b.goal ?? ""),
+      description: description || "新任务",
+      goal,
+      worker_count: mockWorkerCount(b.worker_count),
       status: "created",
       created_at: now.toISOString(),
       created_unix: Math.floor(now.getTime() / 1000),
@@ -1300,9 +1358,9 @@ function route(m: string, path: string, seg: string[], q: URLSearchParams, b: Re
   if (path === "/task-templates" && m === "GET") return { templates: mockTaskTemplates };
   if (path === "/task-templates" && m === "POST") {
     const now = new Date().toISOString();
-    const name = normalizedTemplateName(b.name);
-    const description = String(b.description ?? "").trim();
-    const goal = String(b.goal ?? "").trim();
+    const name = normalizedTemplateName(mockStringField(b.name, "模板名称"));
+    const description = mockStringField(b.description, "模板描述").trim();
+    const goal = mockStringField(b.goal, "模板目标").trim();
     if (!name || !description || !goal) throw new Error("请填写模板名称、描述和目标");
     if (
       mockTaskTemplates.some((template) => normalizedTemplateName(template.name).toLowerCase() === name.toLowerCase())
@@ -1310,11 +1368,19 @@ function route(m: string, path: string, seg: string[], q: URLSearchParams, b: Re
       throw new Error("模板名称已存在");
     }
     const nextID = mockTaskTemplates.reduce((max, template) => Math.max(max, template.id), 0) + 1;
+    const categoryID = mockIntegerField(b.category_id, "category_id");
+    if (b.intercept_rules !== undefined && b.intercept_rules !== null && !Array.isArray(b.intercept_rules)) {
+      throw new Error("intercept_rules 必须是数组");
+    }
     const created: TaskTemplate = {
       id: nextID,
       name,
       description,
       goal,
+      remark: mockRemark(b.remark),
+      category_id: categoryID,
+      intercept_rules: Array.isArray(b.intercept_rules) ? b.intercept_rules : [],
+      worker_count: mockWorkerCount(b.worker_count),
       created_at: now,
       updated_at: now,
     };
@@ -1324,9 +1390,17 @@ function route(m: string, path: string, seg: string[], q: URLSearchParams, b: Re
   if (seg[0] === "task-templates" && seg.length === 2 && m === "PATCH") {
     const template = mockTaskTemplates.find((item) => item.id === Number(seg[1]));
     if (!template) return {};
-    const name = typeof b.name === "string" ? normalizedTemplateName(b.name) : template.name;
-    const description = typeof b.description === "string" ? b.description.trim() : template.description;
-    const goal = typeof b.goal === "string" ? b.goal.trim() : template.goal;
+    const has = (key: string) => Object.prototype.hasOwnProperty.call(b, key);
+    const nameValue = has("name") ? mockStringField(b.name, "模板名称") : undefined;
+    const descriptionValue = has("description") ? mockStringField(b.description, "模板描述") : undefined;
+    const goalValue = has("goal") ? mockStringField(b.goal, "模板目标") : undefined;
+    const categoryID = has("category_id") ? mockIntegerField(b.category_id, "category_id") : template.category_id;
+    if (has("intercept_rules") && b.intercept_rules !== null && !Array.isArray(b.intercept_rules)) {
+      throw new Error("intercept_rules 必须是数组");
+    }
+    const name = nameValue === undefined || b.name === null ? template.name : normalizedTemplateName(nameValue);
+    const description = descriptionValue === undefined || b.description === null ? template.description : descriptionValue.trim();
+    const goal = goalValue === undefined || b.goal === null ? template.goal : goalValue.trim();
     if (!name || !description || !goal) throw new Error("请填写模板名称、描述和目标");
     if (
       mockTaskTemplates.some(
@@ -1335,9 +1409,25 @@ function route(m: string, path: string, seg: string[], q: URLSearchParams, b: Re
     ) {
       throw new Error("模板名称已存在");
     }
+    const remark = has("remark") ? mockRemark(b.remark) : template.remark;
+    const workerCount = has("worker_count") ? mockWorkerCount(b.worker_count) : template.worker_count;
+    const hasName = has("name") && b.name !== null;
+    const hasDescription = has("description") && b.description !== null;
+    const hasGoal = has("goal") && b.goal !== null;
+    const hasCategory = has("category_id");
+    const hasRules = has("intercept_rules");
+    const hasRemark = has("remark");
+    const hasWorkerCount = has("worker_count");
+    if (!hasName && !hasDescription && !hasGoal && !hasCategory && !hasRules && !hasRemark && !hasWorkerCount) {
+      throw new Error("至少需要提供 name、description、goal、remark、category_id、intercept_rules 或 worker_count");
+    }
     template.name = name;
     template.description = description;
     template.goal = goal;
+    template.remark = remark;
+    if (hasCategory) template.category_id = categoryID;
+    if (hasRules) template.intercept_rules = Array.isArray(b.intercept_rules) ? b.intercept_rules : [];
+    template.worker_count = workerCount;
     template.updated_at = new Date().toISOString();
     mockTaskTemplates.sort((a, b) => b.updated_at.localeCompare(a.updated_at) || b.id - a.id);
     return template;
@@ -1356,12 +1446,34 @@ function route(m: string, path: string, seg: string[], q: URLSearchParams, b: Re
   if (seg[0] === "tasks" && seg.length === 2 && m === "PATCH") {
     const task = mockTasks.find((item) => item.id === seg[1]);
     if (!task) throw new Error("任务不存在");
-    if (typeof b.name === "string") task.name = b.name.trim();
-    if (typeof b.remark === "string") task.remark = b.remark.trim();
-    if (typeof b.pinned === "boolean") {
-      task.pinned = b.pinned;
-      task.pinned_at = b.pinned ? (task.pinned_at ?? new Date().toISOString()) : null;
+    const has = (key: string) => Object.prototype.hasOwnProperty.call(b, key);
+    if (has("name") && b.name !== null && typeof b.name !== "string") throw new Error("任务名称必须是字符串");
+    if (has("remark") && b.remark !== null && typeof b.remark !== "string") throw new Error("备注必须是字符串");
+    if (has("pinned") && b.pinned !== null && typeof b.pinned !== "boolean") throw new Error("pinned 必须是布尔值");
+    const namePresent = has("name") && b.name !== null;
+    const remarkPresent = has("remark") && b.remark !== null;
+    const pinnedPresent = has("pinned") && b.pinned !== null;
+    if (!namePresent && !remarkPresent && !pinnedPresent) throw new Error("至少需要提供 name、pinned 或 remark");
+    let nextName = task.name;
+    let nextRemark = task.remark;
+    let nextPinned = task.pinned;
+    let nextPinnedAt = task.pinned_at;
+    if (namePresent) {
+      const name = (b.name as string).trim();
+      if (!name) throw new Error("任务名称不能为空");
+      if (Array.from(name).length > 200) throw new Error("任务名称最多 200 个字符");
+      nextName = name;
     }
+    if (remarkPresent) nextRemark = mockRemark(b.remark);
+    if (pinnedPresent) {
+      const pinned = b.pinned as boolean;
+      nextPinned = pinned;
+      nextPinnedAt = pinned ? (task.pinned_at ?? new Date().toISOString()) : null;
+    }
+    task.name = nextName;
+    task.remark = nextRemark;
+    task.pinned = nextPinned;
+    task.pinned_at = nextPinnedAt;
     sortMockTasks();
     return structuredClone(task);
   }
@@ -2249,7 +2361,15 @@ function route(m: string, path: string, seg: string[], q: URLSearchParams, b: Re
   if (path === "/traffic") return D.traffic;
   if (path === "/traffic/exchange") return D.trafficDetail;
   if (path === "/settings" && m === "GET") return D.settings;
-  if (path === "/settings" && m === "PUT") return { ...D.settings, ...b };
+  if (path === "/settings" && m === "PUT") {
+    const next = { ...D.settings, ...b };
+    if (!Array.isArray(next.worker_pools)) next.worker_pools = D.settings.worker_pools;
+    if (next.worker_pools.length > 0) {
+      next.workers = next.worker_pools.reduce((sum, pool) => sum + Number(pool.workers), 0);
+    }
+    Object.assign(D.settings, next);
+    return next;
+  }
   if (path === "/settings/basic-auth" && m === "GET") return { enabled: false, username: "", password_set: false };
   if (path === "/settings/basic-auth" && m === "PUT")
     throw new Error("演示模式不支持修改 HTTP Basic Auth，请在实际部署中配置");
@@ -2467,7 +2587,8 @@ function route(m: string, path: string, seg: string[], q: URLSearchParams, b: Re
     const from = q.get("from") || "";
     const to = q.get("to") || "";
     if (status && !["pending", "allowed", "denied", "timeout"].includes(status)) throw new Error("无效审批状态");
-    if (decisionSource && !["model", "rule", "manual", "unknown"].includes(decisionSource)) throw new Error("无效判定来源");
+    if (decisionSource && !["model", "rule", "manual", "unknown"].includes(decisionSource))
+      throw new Error("无效判定来源");
     if ([...search].length > 200) throw new Error("搜索关键词不能超过 200 个字符");
     if (search.startsWith("#") && !/^#[1-9]\d*$/.test(search)) throw new Error("审批编号必须为 # 后接正整数");
     for (const value of [from, to]) {

web/src/lib/types.ts

@@ -4,10 +4,27 @@
 export type TaskStatus = "created" | "queued" | "running" | "paused" | "done" | "failed" | "timeout";
 export type EngineMode = "exploring" | "paused" | "stalled" | "idle";
 
+// Keep these limits aligned with db.MaxTaskRemarkRunes and db.MaxTaskWorkerCount.
+export const MAX_TASK_REMARK_RUNES = 2000;
+export const MAX_TASK_WORKER_COUNT = 50;
+
+// Parse the optional worker-count input without silently changing invalid values.
+// An empty field means 0 (use the global default); every other value must be an
+// integer in the backend's accepted range.
+export function parseWorkerCountInput(value: string): number | null {
+  const trimmed = value.trim();
+  if (trimmed === "") return 0;
+  const parsed = Number(trimmed);
+  if (!Number.isFinite(parsed) || !Number.isInteger(parsed)) return null;
+  if (parsed < 0 || parsed > MAX_TASK_WORKER_COUNT) return null;
+  return parsed;
+}
+
 export interface Task {
   distinct_findings?: FindingDistinctStats;
   id: string;
   remark?: string;
+  worker_count?: number; // 本任务 Worker 数;0/缺省=用全局默认
   name?: string; // 可选任务名称;空/缺省=未命名,展示时回退到描述
   category_id?: number;
   category_name?: string;
@@ -56,8 +73,10 @@ export interface TaskTemplate {
   name: string;
   description: string;
   goal: string;
+  remark?: string; // 预设备注
   category_id?: number | null; // 预设分类；null/缺省=无
   intercept_rules?: AssetInterceptRuleInput[]; // 预设的任务级拦截/允许规则
+  worker_count?: number; // 预设 Worker 数;0/缺省=用全局默认
   created_at: string;
   updated_at: string;
 }
@@ -968,6 +987,7 @@ export interface Settings {
   global_proxy?: string;
   python_interpreter?: string; // 自定义脚本工具的 python 解释器路径(空=运行时检测)
   workers?: number; // 并发工作 agent 数(默认3)；对之后启动的任务生效
+  worker_pools: WorkerPool[];
   // 任务并发上限:同时「运行中」的任务数上限。关闭=不限;开启后新建任务超限则排队,有空位自动启动。
   task_concurrency_enabled?: boolean; // 默认 false
   task_concurrency_limit?: number; // 开启后默认 5
@@ -989,6 +1009,11 @@ export interface Settings {
   notify_digest_interval_min?: number; // 汇总模式周期（分钟），默认 30
 }
 
+export interface WorkerPool {
+  profile_id: string;
+  workers: number;
+}
+
 // ---- 漏洞 IM 推送 ----
 
 // NotificationFilter 是渠道的过滤条件，字段全部可选，缺省即不过滤。
@@ -1618,6 +1643,29 @@ export interface FindingTrafficDetail {
   response: EvidenceBodyPreview;
 }
 
+export interface RepeaterSendRequest {
+  method: string;
+  url: string;
+  headers?: Record<string, string>;
+  body?: string;
+  timeout_ms?: number;
+  insecure_skip_verify?: boolean;
+  use_recording_proxy?: boolean;
+}
+
+export interface RepeaterSendResponse {
+  status: number;
+  status_text: string;
+  proto: string;
+  headers: Record<string, string[]>;
+  body: string;
+  body_truncated: boolean;
+  body_bytes: number;
+  elapsed_ms: number;
+  url: string;
+  error?: string;
+}
+
 export interface UpdateSettings {
   proxy: string;
 }
@@ -1734,6 +1782,7 @@ export interface FindingCaseSuggestion {
   task_id: string;
   left_id: string;
   right_id: string;
+  finding_ids: string[];
   title: string;
   reason: string;
   state: string;