diff --git a/mateclaw-server/src/main/java/vip/mate/agent/AgentGraphBuilder.java b/mateclaw-server/src/main/java/vip/mate/agent/AgentGraphBuilder.java
index 168c25bd..5d3648a7 100644
--- a/mateclaw-server/src/main/java/vip/mate/agent/AgentGraphBuilder.java
+++ b/mateclaw-server/src/main/java/vip/mate/agent/AgentGraphBuilder.java
@@ -710,7 +710,8 @@ public class AgentGraphBuilder {
             ObservationProcessor observationProcessor = new ObservationProcessor(graphObservationProperties);
             ObservationNode observationNode = new ObservationNode(observationProcessor, streamTracker);
             SummarizingNode summarizingNode = new SummarizingNode(chatModel, streamingHelper, streamTracker);
-            LimitExceededNode limitExceededNode = new LimitExceededNode(chatModel, observationProcessor, streamingHelper, i18nService);
+            LimitExceededNode limitExceededNode = new LimitExceededNode(
+                    chatModel, observationProcessor, streamingHelper, i18nService, progressLedgerService);
             FinalAnswerNode finalAnswerNode = new FinalAnswerNode(generatedFileCache);
 
             KeyStrategyFactory keyStrategyFactory = KeyStrategy.builder()
diff --git a/mateclaw-server/src/main/java/vip/mate/agent/graph/node/LimitExceededNode.java b/mateclaw-server/src/main/java/vip/mate/agent/graph/node/LimitExceededNode.java
index f5da9d95..3346b387 100644
--- a/mateclaw-server/src/main/java/vip/mate/agent/graph/node/LimitExceededNode.java
+++ b/mateclaw-server/src/main/java/vip/mate/agent/graph/node/LimitExceededNode.java
@@ -47,18 +47,33 @@ public class LimitExceededNode implements NodeAction {
     private final NodeStreamingChatHelper streamingHelper;
     /** Optional i18n service; nullable so legacy/tests without Spring context still work. */
     private final I18nService i18n;
+    /**
+     * Optional ledger loader. When set, the conversation's progress snapshot
+     * (done / in-progress / pending) is appended to the LLM's context so the
+     * "graceful wrap-up" answer can be honest about which steps actually
+     * finished and which were still pending when the iteration cap hit.
+     * Null in legacy/test constructors — the wrap behaves as before.
+     */
+    private final vip.mate.agent.progress.ProgressLedgerService progressLedgerService;
 
     public LimitExceededNode(ChatModel chatModel, ObservationProcessor observationProcessor,
                              NodeStreamingChatHelper streamingHelper) {
-        this(chatModel, observationProcessor, streamingHelper, null);
+        this(chatModel, observationProcessor, streamingHelper, null, null);
     }
 
     public LimitExceededNode(ChatModel chatModel, ObservationProcessor observationProcessor,
                              NodeStreamingChatHelper streamingHelper, I18nService i18n) {
+        this(chatModel, observationProcessor, streamingHelper, i18n, null);
+    }
+
+    public LimitExceededNode(ChatModel chatModel, ObservationProcessor observationProcessor,
+                             NodeStreamingChatHelper streamingHelper, I18nService i18n,
+                             vip.mate.agent.progress.ProgressLedgerService progressLedgerService) {
         this.chatModel = chatModel;
         this.observationProcessor = observationProcessor;
         this.streamingHelper = streamingHelper;
         this.i18n = i18n;
+        this.progressLedgerService = progressLedgerService;
     }
 
     /**
@@ -66,7 +81,7 @@ public class LimitExceededNode implements NodeAction {
      */
     @Deprecated
     public LimitExceededNode(ChatModel chatModel, ObservationProcessor observationProcessor) {
-        this(chatModel, observationProcessor, null, null);
+        this(chatModel, observationProcessor, null, null, null);
     }
 
     @Override
@@ -100,6 +115,25 @@ public class LimitExceededNode implements NodeAction {
             contextForLLM = i18n != null ? i18n.msg("agent.limit_exceeded.empty_context") : "(no tool results)";
         }
 
+        // Prepend the conversation's progress ledger snapshot when available
+        // so the wrap-up answer can be honest about partial completion ("4/10
+        // models researched, 6 still pending") rather than vaguely describing
+        // "what I tried". Without this, hitting the iteration cap on a
+        // 10-step task produces a useless catch-all message — observed in
+        // round-4 of the LLM-review smoke test.
+        String ledgerSnapshot = null;
+        if (progressLedgerService != null && conversationId != null && !conversationId.isBlank()) {
+            try {
+                ledgerSnapshot = progressLedgerService.load(conversationId).renderSnapshot();
+            } catch (Exception e) {
+                log.warn("[LimitExceededNode] Failed to load progress ledger for {}: {}",
+                        conversationId, e.getMessage());
+            }
+        }
+        if (ledgerSnapshot != null) {
+            contextForLLM = ledgerSnapshot + "\n\n---\n\n" + contextForLLM;
+        }
+
         // 构建 prompt
         String systemPrompt = SYSTEM_TEMPLATE.replace("{maxIterations}", String.valueOf(maxIterations));
         String userPrompt = USER_TEMPLATE
diff --git a/mateclaw-server/src/main/resources/prompts/graph/limit-exceeded-system.txt b/mateclaw-server/src/main/resources/prompts/graph/limit-exceeded-system.txt
index fd8f7d26..3393eee1 100644
--- a/mateclaw-server/src/main/resources/prompts/graph/limit-exceeded-system.txt
+++ b/mateclaw-server/src/main/resources/prompts/graph/limit-exceeded-system.txt
@@ -5,6 +5,7 @@
 1. 给出简洁、诚实、可执行的回答
 2. 若信息不足以完全回答，明确说明哪些部分还不确定
 2.1. 若所依据的数据被截断或不完整（含"已截断/TRUNCATED/preview/INCOMPLETE"等标记），明确告知用户数据不完整并指出缺口，**绝不编造、补全或重新编号缺失的数据条目**，也不要把明细列表缩减为范围表述
+2.2. 如果上下文里出现"## 当前任务进度"账本，**以账本为权威完成度** —— done 的步骤真的做完了，pending / in_progress 的步骤明确告诉用户"未完成"，并把这些条目列出来作为下一步建议
 3. 如果还有未完成的方向，简要列出建议的下一步
 4. **不要为「未完成」道歉，不要提及「推理步数」「迭代次数」「工具调用上限」这类技术细节** —— 直接给结论
 5. 保持输出简洁，避免重复已知内容