新增: 消息级溯源 P2 切 ai_messages + AI Chat 跑题改进 P0-P2 + 标题诊断 + 苛刻测

消息级溯源 P2(一次性切读 b + 备份保留):
- 批次A 读路径切 ai_messages(state.rs AppState.ai_messages + 映射 ChatMessage↔AiMessageRecord +
  switch/export load 切 + fallback 兜底)
- 批次B 写路径切 ai_messages(replace_conversation 单事务全量重写 + save/clear 切 + 旧 messages 备份)
- 摘要/压缩不改 updated_at(save_conversation touch_updated_at,compress false 防时间分组跳变)

AI Chat 跑题改进 P0-P2(5 改进,治跑题 5 根因,三原则优雅/可靠/易迭代):
- P0 系统提示聚焦(## 聚焦准则独立段)+ 意图接入 loop(filter_tool_defs 收敛工具 29→5-10,三重 fallback)
- P1 压缩增强(compress prompt 主题锚点 + 失败关键词兜底 extract_keyword_summary)+
  工具结果压缩(should_summarize + extract_key_info view-only 不改持久化)
- P2 主题检测(TrackedMessage.topic + 双高置信保守标记 + tokenize 2-gram 修复中文锚点)
- title LLM complete 失败诊断日志(助定位返空根因)
- 跑题 P0 审查🟡修复(Debug 加 Data 域 + threshold 注释)

苛刻测 38 条(边界/对抗:全漂移/全停用词/全错误行/2KB 边界/连续主题切换/中文 2-gram/emoji)
df-ai 227 passed + workspace EXIT 0

src-tauri/src/commands/ai/prompt.rs

@@ -68,7 +68,12 @@ fn system_prompt_parts(lang: &str) -> (&'static str, &'static str, &'static str)
              - Briefly explain your intent before executing actions\n\
              - Ask for clarification if the user's intent is unclear\n\
              - Prefer using tools to complete actions rather than just describing steps\n\
-             - When a tool call fails, clearly tell the user it failed and why. Never disguise a fallback action as the original intent's success (e.g. don't write to description to fake a directory binding), and never falsely report success\n",
+             - When a tool call fails, clearly tell the user it failed and why. Never disguise a fallback action as the original intent's success (e.g. don't write to description to fake a directory binding), and never falsely report success\n\
+             ## Focus\n\
+             - Always center your response on the core goal of the user's current request; the previous round's topic is only background, not the current task.\n\
+             - When the user switches topics (a clear new intent), follow the latest request; do not drag the old topic into the new answer.\n\
+             - Give the conclusion or action first, then add only necessary explanation; omit tangential information unrelated to the current request.\n\
+             - Do not proactively expand context (files/data) that is irrelevant to the current request.\n",
             "\n## Current Projects\n",
             "\n## Current Tasks\n",
         ),
@@ -86,7 +91,12 @@ fn system_prompt_parts(lang: &str) -> (&'static str, &'static str, &'static str)
              - 执行操作前简要说明你的意图\n\
              - 如果不确定用户意图，先提问\n\
              - 优先使用工具完成操作，而不是只描述步骤\n\
-             - 工具调用失败时必须明确告知用户失败原因，严禁用替代操作冒充原意图成功（如绑定目录失败不得改写描述冒充已绑定），也绝不谎报成功\n",
+             - 工具调用失败时必须明确告知用户失败原因，严禁用替代操作冒充原意图成功（如绑定目录失败不得改写描述冒充已绑定），也绝不谎报成功\n\
+             ## 聚焦准则\n\
+             - 始终围绕用户当前请求的核心目标回答；上一轮的主题只是背景，不是当前任务。\n\
+             - 用户切换话题（明显的新意图）时，以最新请求为准，不要把旧话题带进新回答。\n\
+             - 回答先给结论/动作，再补必要的解释；与当前请求无关的扩展信息省略。\n\
+             - 需要的上下文（文件/数据）若与当前请求无关，不要主动展开。\n",
             "\n## 当前项目\n",
             "\n## 当前任务\n",
         ),
@@ -168,6 +178,9 @@ pub(crate) fn compress_prompt(lang: &str) -> &'static str {
                  - Be concise; prefer bullet points.\n\
                  - Drop small talk and transient pleasantries; keep only technically load-bearing facts.\n\
                  - Preserve file paths, identifiers, and error messages verbatim.\n\
+                 - Must preserve core topic words, entity names, and technical terms the user \
+                   repeatedly mentions; they are anchors for continuing the conversation and losing \
+                   them breaks the context thread.\n\
                  - Do NOT invent facts not present in the conversation.\n\
                  - Output the four sections only, no preamble or extra commentary.",
         _ => "你是对话总结器。请把以下对话压缩为结构化摘要,保留继续推进工作所必需的关键上下文。\
@@ -189,7 +202,84 @@ pub(crate) fn compress_prompt(lang: &str) -> &'static str {
                - 简洁,优先用要点。\n\
                - 去掉寒暄、过渡性客套,只保留技术上有价值的事实。\n\
                - 文件路径、标识符、错误信息等照原样保留。\n\
+               - 必须保留用户反复提及的核心主题词、实体名、技术名词(它们是对话续接的锚点,丢失会致上下文断裂)。\n\
                - 不要编造对话中没有的事实。\n\
                - 只输出上述四段内容,不要前言、解释或额外评论。",
     }
 }
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    // 改进1 系统提示聚焦段:中文版含「聚焦准则」独立段
+    #[test]
+    fn system_prompt_zh_has_focus_section() {
+        let (prefix, _, _) = system_prompt_parts("zh-CN");
+        assert!(
+            prefix.contains("聚焦准则"),
+            "中文 system prompt 应含聚焦准则段,实际: {}",
+            prefix
+        );
+        // 独立段标题,非稀释在行为准则里
+        assert!(prefix.contains("## 聚焦准则"));
+        // 核心条款抽样
+        assert!(prefix.contains("始终围绕用户当前请求的核心目标"));
+        assert!(prefix.contains("也绝不谎报成功"));
+    }
+
+    // 改进1 系统提示聚焦段:英文版含「Focus」独立段
+    #[test]
+    fn system_prompt_en_has_focus_section() {
+        let (prefix, _, _) = system_prompt_parts("en");
+        assert!(
+            prefix.contains("Focus"),
+            "英文 system prompt 应含 Focus 段,实际: {}",
+            prefix
+        );
+        assert!(prefix.contains("## Focus"));
+        assert!(prefix.contains("never falsely report success"));
+        assert!(prefix.contains("core goal of the user's current request"));
+    }
+
+    // 兜底:lang 未匹配回落中文,聚焦段仍存在
+    #[test]
+    fn system_prompt_unknown_lang_falls_back_zh_with_focus() {
+        let (prefix, _, _) = system_prompt_parts("fr");
+        assert!(prefix.contains("聚焦准则"));
+    }
+
+    #[test]
+    fn compress_prompt_unaffected_by_focus_addition() {
+        // 压缩 prompt 是独立函数,聚焦段改动不应波及
+        let zh = compress_prompt("zh-CN");
+        assert!(zh.contains("意图"));
+        let en = compress_prompt("en");
+        assert!(en.contains("Intent"));
+    }
+
+    // 改进3 A: 压缩 prompt 主题保留段(锚点词防上下文断裂)
+    #[test]
+    fn compress_prompt_zh_keeps_topic_anchor_clause() {
+        let zh = compress_prompt("zh-CN");
+        assert!(
+            zh.contains("主题词"),
+            "中文 compress_prompt 应含主题词/锚点保留要求,实际: {}",
+            zh
+        );
+        assert!(zh.contains("锚点"));
+        assert!(zh.contains("上下文断裂"));
+    }
+
+    #[test]
+    fn compress_prompt_en_keeps_topic_anchor_clause() {
+        let en = compress_prompt("en");
+        assert!(
+            en.contains("anchors"),
+            "英文 compress_prompt 应含 anchors 保留要求,实际: {}",
+            en
+        );
+        assert!(en.contains("topic words"));
+        assert!(en.contains("breaks the context"));
+    }
+}