diff --git a/TeXmacs/plugins/autosave/progs/init-autosave.scm b/TeXmacs/plugins/autosave/progs/init-autosave.scm
index 22767fd809..599e53ec19 100644
--- a/TeXmacs/plugins/autosave/progs/init-autosave.scm
+++ b/TeXmacs/plugins/autosave/progs/init-autosave.scm
@@ -19,7 +19,7 @@
   (with u
     (pre-serialize lan t)
     (with s
-      (texmacs->code (stree->tree u) "SourceCode")
+      (texmacs->utf8raw (stree->tree u))
       (string-append s "\n<EOF>\n")
     ) ;with
   ) ;with
diff --git a/TeXmacs/plugins/gnuplot/progs/init-gnuplot.scm b/TeXmacs/plugins/gnuplot/progs/init-gnuplot.scm
index 758e5aa75b..dbe0b5278e 100644
--- a/TeXmacs/plugins/gnuplot/progs/init-gnuplot.scm
+++ b/TeXmacs/plugins/gnuplot/progs/init-gnuplot.scm
@@ -20,7 +20,7 @@
     (lazy-format (data gnuplot) gnuplot)
 
     (define (gnuplot-serialize lan t)
-      (let* ((u (pre-serialize lan t)) (s (texmacs->code (stree->tree u) "utf-8")))
+      (let* ((u (pre-serialize lan t)) (s (texmacs->utf8raw (stree->tree u))))
         (string-append s "\n<EOF>\n")
       ) ;let*
     ) ;define
diff --git a/TeXmacs/plugins/goldfish/progs/init-goldfish.scm b/TeXmacs/plugins/goldfish/progs/init-goldfish.scm
index 9b1d83020c..3af2836e0b 100644
--- a/TeXmacs/plugins/goldfish/progs/init-goldfish.scm
+++ b/TeXmacs/plugins/goldfish/progs/init-goldfish.scm
@@ -16,7 +16,7 @@
 (lazy-format (data r7rs) r7rs)
 
 (define (goldfish-serialize lan t)
-  (let* ((u (pre-serialize lan t)) (s (texmacs->code (stree->tree u) "SourceCode")))
+  (let* ((u (pre-serialize lan t)) (s (texmacs->utf8raw (stree->tree u))))
     (string-append s "\n<EOF>\n")
   ) ;let*
 ) ;define
diff --git a/TeXmacs/plugins/julia/progs/init-julia.scm b/TeXmacs/plugins/julia/progs/init-julia.scm
index 8acecfd947..e59d316063 100644
--- a/TeXmacs/plugins/julia/progs/init-julia.scm
+++ b/TeXmacs/plugins/julia/progs/init-julia.scm
@@ -15,7 +15,7 @@
 
 (define (julia-serialize lan t)
   (let* ((u (pre-serialize lan t))
-         (s (texmacs->code (stree->tree u) "SourceCode")))
+         (s (texmacs->utf8raw (stree->tree u))))
     (string-append s "\n<EOF>\n")))
 
 (define (julia-entry)
diff --git a/TeXmacs/plugins/maxima/progs/init-maxima.scm b/TeXmacs/plugins/maxima/progs/init-maxima.scm
index e1be8d90fb..6e1fc79304 100644
--- a/TeXmacs/plugins/maxima/progs/init-maxima.scm
+++ b/TeXmacs/plugins/maxima/progs/init-maxima.scm
@@ -14,7 +14,7 @@
 (use-modules (binary maxima))
 
 (define (maxima-serialize lan t)
-  (with s (string-drop-right (verbatim-serialize lan t) 1)
+  (with s (string-drop-right (utf8raw-serialize lan t) 1)
     (cond ((== s "") "0;\n")
           ((in? (string-ref s (- (string-length s) 1)) '(#\; #\$))
            (string-append s "\n"))
diff --git a/TeXmacs/plugins/python/progs/init-python.scm b/TeXmacs/plugins/python/progs/init-python.scm
index 21db55ef0b..6aa336ade3 100644
--- a/TeXmacs/plugins/python/progs/init-python.scm
+++ b/TeXmacs/plugins/python/progs/init-python.scm
@@ -34,7 +34,7 @@
 
 (define (python-serialize lan t)
   (with u (pre-serialize lan t)
-    (with s (texmacs->code (stree->tree u) "SourceCode")
+    (with s (texmacs->utf8raw (stree->tree u))
       (string-append  s  "\n<EOF>\n"))))
 
 (define (python-utf8-command)
diff --git a/TeXmacs/plugins/quiver/progs/init-quiver.scm b/TeXmacs/plugins/quiver/progs/init-quiver.scm
index ca9618ddb2..3ef2db3b9a 100644
--- a/TeXmacs/plugins/quiver/progs/init-quiver.scm
+++ b/TeXmacs/plugins/quiver/progs/init-quiver.scm
@@ -18,7 +18,7 @@
 
     (define (quiver-serialize lan t)
       (let* ((u (pre-serialize lan t))
-             (s (texmacs->code (stree->tree u) "SourceCode")))
+             (s (texmacs->utf8raw (stree->tree u))))
         (string-append s "\n<EOF>\n")
       )
     )
diff --git a/TeXmacs/plugins/tikz/progs/init-tikz.scm b/TeXmacs/plugins/tikz/progs/init-tikz.scm
index 13909100d7..8f96af1d8f 100644
--- a/TeXmacs/plugins/tikz/progs/init-tikz.scm
+++ b/TeXmacs/plugins/tikz/progs/init-tikz.scm
@@ -20,7 +20,7 @@
 
     (define (tikz-serialize lan t)
       (let* ((u (pre-serialize lan t))
-             (s (texmacs->code (stree->tree u) "SourceCode")))
+             (s (texmacs->utf8raw (stree->tree u))))
         (string-append s "\n<EOF>\n")
       )
     )
diff --git a/TeXmacs/progs/convert/rewrite/init-rewrite.scm b/TeXmacs/progs/convert/rewrite/init-rewrite.scm
index 355e1851c4..1c420ce385 100644
--- a/TeXmacs/progs/convert/rewrite/init-rewrite.scm
+++ b/TeXmacs/progs/convert/rewrite/init-rewrite.scm
@@ -77,6 +77,16 @@
 
 (converter code-snippet texmacs-tree (:function code-snippet->texmacs))
 
+;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
+;; UTF-8 raw (plugin I/O)
+;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
+
+(tm-define (texmacs->utf8raw t)
+  (if (tree? t) (cpp-tree->utf8raw t) (texmacs->utf8raw (tm->tree t)))
+) ;tm-define
+
+(tm-define (utf8raw->texmacs s) (cpp-utf8raw->tree s))
+
 ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
 ;; Verbatim
 ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
diff --git a/TeXmacs/progs/dynamic/program-edit.scm b/TeXmacs/progs/dynamic/program-edit.scm
index f03c8a523b..4496cb1251 100644
--- a/TeXmacs/progs/dynamic/program-edit.scm
+++ b/TeXmacs/progs/dynamic/program-edit.scm
@@ -742,7 +742,8 @@
                 (opts (input-options t))
                 (st (tree->stree (tree-ref t 1)))
                 (pre (plugin-preprocess lan ses st opts))
-                (in (plugin-serialize lan pre))
+                (pre-u8 (tree->stree (herk-tree->utf8-tree (stree->tree pre))))
+                (in (plugin-serialize lan pre-u8))
                 (rew (if (string-ends? in "\n") (string-drop-right in 1) in))
                 (cmd (string-append "(input-done? " (string-quote rew) ")"))
                 (ret (lambda (done?) (kbd-enter-sub t done?)))
diff --git a/TeXmacs/progs/dynamic/session-edit.scm b/TeXmacs/progs/dynamic/session-edit.scm
index 6e1d5f4147..23cb81e0af 100644
--- a/TeXmacs/progs/dynamic/session-edit.scm
+++ b/TeXmacs/progs/dynamic/session-edit.scm
@@ -1011,7 +1011,8 @@
                 (opts (input-options t))
                 (st (tree->stree (tree-ref t 1)))
                 (pre (plugin-preprocess lan ses st opts))
-                (in (plugin-serialize lan pre))
+                (pre-u8 (tree->stree (herk-tree->utf8-tree (stree->tree pre))))
+                (in (plugin-serialize lan pre-u8))
                 (rew (if (string-ends? in "\n") (string-drop-right in 1) in))
                 (cmd (string-append "(input-done? " (string-quote rew) ")"))
                 (ret (lambda (done?) (kbd-enter-sub t done?)))
diff --git a/TeXmacs/progs/init-research.scm b/TeXmacs/progs/init-research.scm
index 4ed3a316ba..72335338ab 100644
--- a/TeXmacs/progs/init-research.scm
+++ b/TeXmacs/progs/init-research.scm
@@ -129,7 +129,7 @@
 (lazy-define (utils library cursor) notify-cursor-moved)
 (lazy-define (utils edit variants) make-inline-tag-list make-wrapped-tag-list)
 (lazy-define (utils cas cas-out) cas->stree)
-(lazy-define (utils plugins plugin-cmd) pre-serialize verbatim-serialize)
+(lazy-define (utils plugins plugin-cmd) pre-serialize utf8raw-serialize)
 (lazy-define (utils test test-convert)
   delayed-quit
   build-manual
@@ -534,7 +534,7 @@
   export-selection-as-graphics
   clipboard-copy-image
 ) ;lazy-define
-(lazy-define (convert rewrite init-rewrite) texmacs->code texmacs->verbatim)
+(lazy-define (convert rewrite init-rewrite) texmacs->code texmacs->verbatim texmacs->utf8raw utf8raw->texmacs)
 (lazy-define (convert html tmhtml) ext-tmhtml-eqnarray*)
 (define-secure-symbols ext-tmhtml-eqnarray*)
 (lazy-define (convert html tmhtml-expand) tmhtml-env-patch)
diff --git a/TeXmacs/progs/utils/plugins/plugin-cmd.scm b/TeXmacs/progs/utils/plugins/plugin-cmd.scm
index e18274e933..ed5f0aa220 100644
--- a/TeXmacs/progs/utils/plugins/plugin-cmd.scm
+++ b/TeXmacs/progs/utils/plugins/plugin-cmd.scm
@@ -28,14 +28,14 @@
   ) ;cond
 ) ;tm-define
 
-(define (hacked-texmacs->code x)
-  (with r (texmacs->code x "utf-8") (string-replace r "`" "`"))
+(define (hacked-texmacs->utf8raw x)
+  (with r (texmacs->utf8raw x) (string-replace r "`" "`"))
 ) ;define
 
-(tm-define (verbatim-serialize lan t)
+(tm-define (utf8raw-serialize lan t)
   (with u
     (pre-serialize lan t)
-    (string-append (escape-verbatim (hacked-texmacs->code u)) "\n")
+    (string-append (escape-verbatim (hacked-texmacs->utf8raw u)) "\n")
   ) ;with
 ) ;tm-define
 
@@ -44,7 +44,7 @@
     (pre-serialize lan t)
     (string-append (char->string #\x02)
       "utf8:"
-      (escape-generic (texmacs->code u "utf-8"))
+      (escape-generic (texmacs->utf8raw u))
       (char->string #\x05)
     ) ;string-append
   ) ;with
@@ -53,7 +53,7 @@
 (tm-define (plugin-serialize lan t)
   (with fun
     (ahash-ref plugin-serializer lan)
-    (if fun (fun lan t) (verbatim-serialize lan t))
+    (if fun (fun lan t) (utf8raw-serialize lan t))
   ) ;with
 ) ;tm-define
 
diff --git a/TeXmacs/tests/0828.scm b/TeXmacs/tests/0828.scm
new file mode 100644
index 0000000000..a374f62b2e
--- /dev/null
+++ b/TeXmacs/tests/0828.scm
@@ -0,0 +1,222 @@
+;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
+;;
+;; MODULE      : 0828.scm
+;; DESCRIPTION : UTF-8 raw 插件 I/O 序列化器测试
+;; COPYRIGHT   : (C) 2026 AcceleratorX
+;;
+;; This software falls under the GNU general public license version 3 or later.
+;; It comes WITHOUT ANY WARRANTY WHATSOEVER. For details, see the file LICENSE
+;; in the root directory or <http://www.gnu.org/licenses/gpl-3.0.html>.
+;;
+;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
+
+(import (scheme base)
+        (liii check)
+        (liii string))
+
+(check-set-mode! 'report-failed)
+
+;;; ========== 辅助函数 ==========
+
+;; 由含 UTF-8 字符串标签的 stree 直接构造 UTF-8 tree
+(define (utf8-tree label)
+  (stree->tree `(document ,label)))
+
+;; 加载插件 init 脚本（注册其自定义 serializer）
+(define (load-plugin-init name)
+  (load (string-append (url->system (get-texmacs-path))
+                       "/plugins/" name "/progs/init-" name ".scm")))
+
+;; 用 UTF-8 stree 调 plugin-serialize（plugin-serialize 约定接收 stree）
+(define (plugin-serialize-utf8 lan label)
+  (plugin-serialize lan `(document ,label)))
+
+;;; ========== texmacs->utf8raw / utf8raw->texmacs 往返 ==========
+
+(define (test-utf8raw-roundtrip)
+  (define (roundtrip s)
+    (texmacs->utf8raw (utf8raw->texmacs s)))
+  (check (roundtrip "中文测试") => "中文测试")
+  (check (roundtrip "αβγδ") => "αβγδ")
+  (check (roundtrip "Hello 世界") => "Hello 世界")
+  (check (roundtrip "∑∏∫√") => "∑∏∫√")
+  (check (roundtrip "🎉🎊") => "🎉🎊")
+  (check (roundtrip "line1\nline2") => "line1\nline2")
+  (check (roundtrip "a\n\nb") => "a\n\nb")
+  (check (roundtrip "") => "")
+  (check (roundtrip "  ") => "  "))
+
+;;; ========== 由 UTF-8 tree 序列化 ==========
+
+(define (test-utf8raw-from-tree)
+  (check (texmacs->utf8raw (utf8-tree "中文测试")) => "中文测试")
+  (check (texmacs->utf8raw (utf8-tree "αβγδ")) => "αβγδ")
+  (check (texmacs->utf8raw (utf8-tree "Hello 世界")) => "Hello 世界")
+  (check (texmacs->utf8raw (utf8-tree "∑∏∫√")) => "∑∏∫√")
+  (check (texmacs->utf8raw (stree->tree '(document "line1" "line2"))) => "line1\nline2"))
+
+;;; ========== utf8raw->texmacs 保留原始字节 ==========
+
+(define (test-utf8raw-to-tree)
+  ;; utf8: 块里的 <#XXXX> 字面必须原样保留，不得被解码
+  (define t (utf8raw->texmacs "<#5206><#5B50>"))
+  (check (texmacs->utf8raw t) => "<#5206><#5B50>")
+  ;; CR/LF 归一化
+  (check (texmacs->utf8raw (utf8raw->texmacs "a\r\nb")) => "a\nb")
+  (check (texmacs->utf8raw (utf8raw->texmacs "a\rb")) => "a\nb"))
+
+;;; ========== 默认 serializer 输出 UTF-8 raw ==========
+
+(define (test-utf8raw-serialize-utf8)
+  (define s (utf8raw-serialize "python" (utf8-tree "print('中文')")))
+  (check (string-contains? s "中文") => #t)
+  (check (string-contains? s "<#4E2D>") => #f))
+
+(define (test-generic-serialize-utf8)
+  (define s (generic-serialize "python" (utf8-tree "print('中文')")))
+  (check (string-starts? s (char->string #\x02)) => #t)
+  (check (string-contains? s "utf8:") => #t)
+  (check (string-contains? s "中文") => #t)
+  (check (string-contains? s "<#4E2D>") => #f)
+  (check (string-ends? s (char->string #\x05)) => #t))
+
+;;; ========== 各插件自定义 serializer 输出 UTF-8 ==========
+
+(define (test-python-serialize-utf8)
+  (load-plugin-init "python")
+  (with s (plugin-serialize-utf8 "python" "print('中文')")
+    (check (string-contains? s "中文") => #t)
+    (check (string-contains? s "<#4E2D>") => #f)
+    (when (string-contains? s "<EOF>")
+      (check (string-ends? s "\n<EOF>\n") => #t))))
+
+(define (test-julia-serialize-utf8)
+  (load-plugin-init "julia")
+  (with s (plugin-serialize-utf8 "julia" "println(\"中文\")")
+    (check (string-contains? s "中文") => #t)
+    (check (string-contains? s "<#4E2D>") => #f)
+    (when (string-contains? s "<EOF>")
+      (check (string-ends? s "\n<EOF>\n") => #t))))
+
+(define (test-goldfish-serialize-utf8)
+  (load-plugin-init "goldfish")
+  (with s (plugin-serialize-utf8 "goldfish" "(display \"中文\")")
+    (check (string-contains? s "中文") => #t)
+    (check (string-contains? s "<#4E2D>") => #f)
+    (when (string-contains? s "<EOF>")
+      (check (string-ends? s "\n<EOF>\n") => #t))))
+
+(define (test-gnuplot-serialize-utf8)
+  (load-plugin-init "gnuplot")
+  (with s (plugin-serialize-utf8 "gnuplot" "set title \"中文\"")
+    (check (string-contains? s "中文") => #t)
+    (check (string-contains? s "<#4E2D>") => #f)
+    (when (string-contains? s "<EOF>")
+      (check (string-ends? s "\n<EOF>\n") => #t))))
+
+(define (test-autosave-serialize-utf8)
+  (load-plugin-init "autosave")
+  (with s (plugin-serialize-utf8 "autosave" "(display \"中文\")")
+    (check (string-contains? s "中文") => #t)
+    (check (string-contains? s "<#4E2D>") => #f)
+    (when (string-contains? s "<EOF>")
+      (check (string-ends? s "\n<EOF>\n") => #t))))
+
+(define (test-tikz-serialize-utf8)
+  (load-plugin-init "tikz")
+  (with s (plugin-serialize-utf8 "tikz" "\\node {中文};")
+    (check (string-contains? s "中文") => #t)
+    (check (string-contains? s "<#4E2D>") => #f)
+    (when (string-contains? s "<EOF>")
+      (check (string-ends? s "\n<EOF>\n") => #t))))
+
+(define (test-quiver-serialize-utf8)
+  (load-plugin-init "quiver")
+  (with s (plugin-serialize-utf8 "quiver" "\\node {中文};")
+    (check (string-contains? s "中文") => #t)
+    (check (string-contains? s "<#4E2D>") => #f)
+    (when (string-contains? s "<EOF>")
+      (check (string-ends? s "\n<EOF>\n") => #t))))
+
+;; Maxima 经 (plugin-configure :serializer ...) 注册，依赖二进制存在；测试环境无二进制，
+;; plugin-serialize 会落回默认 utf8raw-serialize 而测不到 maxima-serialize，故直接调用。
+(define (test-maxima-serialize-utf8)
+  (load-plugin-init "maxima")
+  (with s (maxima-serialize "maxima" '(document "x:中文"))
+    (check (string-contains? s "中文") => #t)
+    (check (string-contains? s "<#4E2D>") => #f)
+    (check (string-ends? s ";\n") => #t))
+  (check (maxima-serialize "maxima" '(document "")) => "0;\n")
+  (with s (maxima-serialize "maxima" '(document "中文$"))
+    (check (string-ends? s "中文$\n") => #t)))
+
+;;; ========== 输入端：utf8raw->texmacs 后转 Herk ==========
+
+;; 模拟 input.cpp：把 utf8: 负载解析成 UTF-8 tree，再转成内部 Herk tree。
+(define (test-input-utf8-to-herk)
+  (define herk-str (texmacs->stm (utf8-tree->herk-tree (utf8raw->texmacs "print('中文')"))))
+  (check (string-contains? herk-str "<#4E2D>") => #t)
+  (check (string-contains? herk-str "<#6587>") => #t)
+  (check (string-contains? herk-str "中文") => #f))
+
+;; emoji（BMP 外）在 utf8_to_herk 映射表里没有对应项，tree_utf8_to_herk 须以 <#XXXX>
+;; 保留而非丢弃——这是 0828 相对 0827（走 utf8_to_cork 会静默丢失此类字符）的关键修正。
+(define (test-input-emoji-to-herk)
+  (define herk-str (texmacs->stm (utf8-tree->herk-tree (utf8raw->texmacs "x:🎉"))))
+  (check (string-contains? herk-str "<#1F389>") => #t)
+  (check (string-contains? herk-str "🎉") => #f))
+
+;; 普通文本与 <#XXXX> 字面混合：<#XXXX> 原样存活（不被当作 UTF-8 字节再解码），CJK 正常往返。
+(define (test-utf8raw-mixed-hex-literal)
+  (define t (utf8raw->texmacs "代码<#5206>"))
+  (check (texmacs->utf8raw t) => "代码<#5206>"))
+
+;;; ========== 写出端：Herk tree -> UTF-8 tree -> raw 串 ==========
+
+;; 模拟主进程到插件的路径：内部 Herk tree 先转 UTF-8 tree，再由 texmacs->utf8raw 序列化。
+(define (test-write-path-herk-to-utf8raw)
+  (define s (texmacs->utf8raw
+              (herk-tree->utf8-tree
+                (stm-snippet->texmacs "(frac \"<#5206><#5B50>\" \"<#5206><#6BCD>\")"))))
+  (check (string-contains? s "分子") => #t)
+  (check (string-contains? s "分母") => #t)
+  (check (string-contains? s "<#5206>") => #f))
+
+;;; ========== input-done? 旁路：序列化前 Herk stree 转 UTF-8 ==========
+
+;; session-edit.scm / program-edit.scm 在 input-done? 路径上，先把 pre（Herk stree）转成
+;; UTF-8 stree 再 plugin-serialize。下方正向断言验证转换后输出 UTF-8；反向断言证明不转换
+;; 则 Herk 的 <#XXXX> 字面会原样漏给插件（插件收到 8 个 ASCII 字符 "<#4E2D>" 而非"中"），
+;; 即该转换是必需的、不可省略。
+(define (test-input-done-herk-to-utf8)
+  (load-plugin-init "python")
+  (define herk-stree '(document "print('<#4E2D><#6587>')"))
+  (define pre-u8 (tree->stree (herk-tree->utf8-tree (stree->tree herk-stree))))
+  (check (string-contains? (plugin-serialize "python" pre-u8) "中文") => #t)
+  (check (string-contains? (plugin-serialize "python" pre-u8) "<#4E2D>") => #f)
+  (define bad (plugin-serialize "python" herk-stree))
+  (check (string-contains? bad "<#4E2D>") => #t)
+  (check (string-contains? bad "中文") => #f))
+
+;;; ========== 测试入口 ==========
+
+(tm-define (test_0828)
+  (test-utf8raw-roundtrip)
+  (test-utf8raw-from-tree)
+  (test-utf8raw-to-tree)
+  (test-utf8raw-serialize-utf8)
+  (test-generic-serialize-utf8)
+  (test-python-serialize-utf8)
+  (test-julia-serialize-utf8)
+  (test-goldfish-serialize-utf8)
+  (test-gnuplot-serialize-utf8)
+  (test-autosave-serialize-utf8)
+  (test-tikz-serialize-utf8)
+  (test-quiver-serialize-utf8)
+  (test-maxima-serialize-utf8)
+  (test-input-utf8-to-herk)
+  (test-input-emoji-to-herk)
+  (test-utf8raw-mixed-hex-literal)
+  (test-write-path-herk-to-utf8raw)
+  (test-input-done-herk-to-utf8)
+  (check-report))
diff --git a/devel/0828.md b/devel/0828.md
new file mode 100644
index 0000000000..6d8314e2c9
--- /dev/null
+++ b/devel/0828.md
@@ -0,0 +1,56 @@
+# [0828] 插件 I/O 序列化器迁移到 UTF-8 raw
+
+## 1 相关文档
+- [0827.md](0827.md) - 插件 I/O 协议统一为 UTF-8（上一步）
+- [dddd.md](dddd.md) - 任务文档模板
+
+## 2 任务相关的代码文件
+- `src/Data/Convert/Verbatim/verbatim.cpp` — 新增 `tree_to_utf8raw` / `utf8raw_to_tree`
+- `src/Data/Convert/convert.hpp` — 声明新函数
+- `src/Scheme/L4/glue_convert.lua` — 暴露 `cpp-tree->utf8raw` / `cpp-utf8raw->tree`
+- `TeXmacs/progs/convert/rewrite/init-rewrite.scm` — Scheme 包装 `texmacs->utf8raw` / `utf8raw->texmacs`
+- `TeXmacs/progs/init-research.scm` — lazy-define 新包装函数
+- `TeXmacs/progs/utils/plugins/plugin-cmd.scm` — 默认序列化器重命名为 `utf8raw-serialize` 并改用 `texmacs->utf8raw`
+- `TeXmacs/plugins/{python,julia,goldfish,autosave,tikz,quiver,gnuplot,maxima}/progs/init-*.scm` — 自定义 serializer 改用 `texmacs->utf8raw`
+- `src/Data/Convert/Generic/input.cpp` — `verbatim_flush` / `utf8_flush` 改用 `utf8raw_to_tree` + `tree_utf8_to_herk`
+- `src/System/Link/connection.cpp` — `connection_write(tree)` 入口加入 `tree_herk_to_utf8`
+- `src/Edit/Interface/edit_complete.cpp` — `session_complete_command` 在调用 `utf8raw-serialize` 前加入 `tree_herk_to_utf8`
+- `TeXmacs/progs/dynamic/session-edit.scm` — `input-done?` 路径在调用 `plugin-serialize` 前把 `pre` 从 Herk 转成 UTF-8
+- `TeXmacs/progs/dynamic/program-edit.scm` — 同上，针对 program session
+- `TeXmacs/tests/0828.scm` — 新增 UTF-8 raw 往返、各插件 serializer UTF-8、输入/输出路径转换测试
+
+## 3 如何测试
+
+### 3.1 确定性测试（单元测试）
+```bash
+xmake b stem
+xmake r 0801
+xmake r 0804
+xmake r 0827
+xmake r 0828
+```
+
+### 3.2 非确定性测试（文档验证）
+
+## 4 What
+- 新增无 cork/SourceCode 转换的 UTF-8 raw 序列化/反序列化函数。
+- 插件 I/O 主路径（`connection_write(tree)`）先把内部 Herk tree 转成 UTF-8 tree。
+- 所有插件 serializer 改用 `texmacs->utf8raw`，不再使用 `texmacs->code ... "SourceCode"` 或 `"utf-8"`。
+- 输入端 `verbatim_flush` / `utf8_flush` 改用 `utf8raw_to_tree`，再转回内部 Herk。
+- `tree_to_verbatim` / `verbatim_to_tree` 保留给文件格式转换等其他用途，plugin I/O 不再使用。
+
+## 5 How
+1. 在 `verbatim.cpp` 新增 `tree_to_utf8raw(tree)`，直接调用 `as_verbatim` 并处理 Windows 换行，跳过 `var_cork_to_utf8` / `var_cork_to_sourcecode`。
+2. 在 `verbatim.cpp` 新增 `utf8raw_to_tree(string)`，复用 `mac_to_unix` / `dos_to_unix` 与拆分逻辑，但不经过 `encode`（即不做 `utf8_to_cork`）。把原有拆分逻辑抽成静态 helper `verbatim_raw_to_tree` 供旧函数复用。两函数均不设 `wrap` 形参——`tree_to_verbatim` / `verbatim_to_tree` 保留了 `wrap` 是为了文件格式转换等需要折行的场景，而插件 I/O 一律传 `false`，故 raw 版本直接省去该形参，避免留下一处无调用者、无测试的折行分支。
+3. `convert.hpp` 声明两个新函数；`glue_convert.lua` 暴露给 Scheme。
+4. `init-rewrite.scm` 提供 `texmacs->utf8raw` / `utf8raw->texmacs` 包装；`init-research.scm` 加入 lazy-define。
+5. `plugin-cmd.scm` 的 `verbatim-serialize` 重命名为 `utf8raw-serialize`，`generic-serialize` 改用 `texmacs->utf8raw`。
+6. Python、Julia、Goldfish、Autosave、TikZ、Quiver、Gnuplot、Maxima 自定义 serializer 全部替换为 `texmacs->utf8raw (stree->tree u)`。
+7. `input.cpp` 的 `verbatim_flush` / `utf8_flush` 改为 `tree_utf8_to_herk (utf8raw_to_tree (buf))`。`texmacs_input_rep` 是插件连接的"输入端"——站在 TeXmacs 视角，插件进程经管道发回的数据要"输入"进来；`write(tree)` 把解析结果喂给**编辑器内部的文档树**（不是写文件，也不是发回插件）。编辑器内部所有字符串标签统一用 Herk 编码，所以 `write` 的实参必须是 Herk tree，而 `buf` 是插件发来的 UTF-8 字节，分两步翻译：`utf8raw_to_tree(buf)` 先切成 `(document ...)` 但标签仍是 UTF-8 原始字节（raw，不做编码转换），`tree_utf8_to_herk(...)` 再把每个 atomic 标签从 UTF-8 解码成 Herk。两步合起来等价于旧版 `verbatim_to_tree(buf, false, "utf-8")`（内部走 `utf8_to_cork`），但编码目标从 cork 升级为 Herk——cork 表之外的码点，旧的 `utf8_to_cork` 无映射就丢弃，新的 `utf8_to_herk` 用 `<#XXXX>` 保留，emoji、BMP 外字符因此不再丢失，这是相对 0827 的关键修正。此外 `utf8raw_to_tree` 引入了 `dos_to_unix` / `mac_to_unix` 换行归一化（旧 `verbatim_to_tree(buf,"utf-8")` 路径没有），插件输出里的 `\r\n` / 裸 `\r` 会被统一成 `\n`。写出端是对称的逆过程：第 8–10 条里编辑器内部的 Herk tree 先经 `tree_herk_to_utf8` 转成 UTF-8 tree，再交给 serializer 输出给插件。
+8. `connection.cpp` 的 `connection_write(tree)` 在 `tree_to_stree` 之前调用 `tree_herk_to_utf8`。
+9. `edit_complete.cpp` 的 `session_complete_command` 在调用 `utf8raw-serialize` 前加入 `tree_herk_to_utf8`，处理 tab completion 这条旁路。
+10. `session-edit.scm` 与 `program-edit.scm` 的 `input-done?` 路径在调用 `plugin-serialize` 前把 `pre` 从 Herk 转成 UTF-8，避免该查询路径发送 cork 文本给插件。**该转换不可省略：`plugin-serialize` 约定接收 UTF-8 stree**，而编辑器内部字符串为 Herk/cork，省略会让 Herk 的 `<#XXXX>` 字面原样漏给插件——插件收到的是 8 个 ASCII 字符 `<#4E2D>` 而非"中"。`0828.scm` 的 `test-input-done-herk-to-utf8` 用同一 Herk stree 对比了转换与不转换两种结果，断言不转换时 `<#XXXX>` 会漏出，以此证明该转换是必需的而非冗余。当前实现有 `stree->tree->stree` 往返开销，但 `input-done?` 输入通常很小，影响可忽略；若后续出现瓶颈，可改为直接 stree 级转换。
+11. 新增 `TeXmacs/tests/0828.scm`，覆盖 `texmacs->utf8raw` / `utf8raw->texmacs` 往返、各插件 serializer（含 Maxima 的 `;`/`$` 终止符与空输入特例）的 UTF-8 输出、输入端 Herk 转换（含 emoji 落到 `<#XXXX>` 的保留路径）、`<#XXXX>` 与普通文本混合的往返、主写出入口 Herk->UTF-8 转换、以及 `input-done?` 旁路 Herk stree 转 UTF-8 后再序列化（含不转换的反向对照）等场景。
+
+## PS
+- `TeXmacs/plugins/*/progs/data/*.scm` 中出现的 `(texmacs->verbatim x (acons "texmacs->verbatim:encoding" "SourceCode" '()))` 用于文件格式导入/导出（如 `.py`、`.cpp`、`.scm`），不属于插件 I/O 路径；SourceCode 编码用于文件持久化与往返，0828 不涉及这些路径，保持原样不改。
diff --git a/src/Data/Convert/Generic/input.cpp b/src/Data/Convert/Generic/input.cpp
index 1f03d44d6a..20d62482f9 100644
--- a/src/Data/Convert/Generic/input.cpp
+++ b/src/Data/Convert/Generic/input.cpp
@@ -260,7 +260,7 @@ texmacs_input_rep::flush (bool force) {
 void
 texmacs_input_rep::verbatim_flush (bool force) {
   if (force || ends (buf, "\n")) {
-    if (!ignore_verb) write (verbatim_to_tree (buf, false, "utf-8"));
+    if (!ignore_verb) write (tree_utf8_to_herk (utf8raw_to_tree (buf)));
     else if (DEBUG_IO) debug_io << "ignore verbatim (aborted input)" << LF;
     buf= "";
   }
@@ -270,7 +270,7 @@ void
 texmacs_input_rep::utf8_flush (bool force) {
   if (force || ends (buf, "\n")) {
     if (!ignore_verb) {
-      write (verbatim_to_tree (buf, false, "utf-8"));
+      write (tree_utf8_to_herk (utf8raw_to_tree (buf)));
     }
     else if (DEBUG_IO) debug_io << "ignore verbatim (aborted input)" << LF;
     buf= "";
diff --git a/src/Data/Convert/Verbatim/verbatim.cpp b/src/Data/Convert/Verbatim/verbatim.cpp
index 167f4de815..2da56782ad 100644
--- a/src/Data/Convert/Verbatim/verbatim.cpp
+++ b/src/Data/Convert/Verbatim/verbatim.cpp
@@ -271,6 +271,16 @@ tree_to_verbatim (tree t, bool wrap, string enc) {
 #endif
 }
 
+string
+tree_to_utf8raw (tree t) {
+  string buf= as_verbatim (t, false);
+#ifdef OS_WIN
+  return unix_to_dos (buf);
+#else
+  return buf;
+#endif
+}
+
 /******************************************************************************
  * Verbatim to TeXmacs
  ******************************************************************************/
@@ -302,9 +312,8 @@ encode (string s, string enc) {
   else return tm_encode (s);
 }
 
-tree
-verbatim_to_tree (string s, string enc) {
-  s= encode (s, enc);
+static tree
+verbatim_raw_to_tree (string s) {
   int i, j;
   int s_N= N (s);
   for (i= 0; i < s_N; i++)
@@ -321,6 +330,12 @@ verbatim_to_tree (string s, string enc) {
   return un_special (s);
 }
 
+tree
+verbatim_to_tree (string s, string enc) {
+  s= encode (s, enc);
+  return verbatim_raw_to_tree (s);
+}
+
 string
 dos_to_unix (string s) {
   int    i, n= N (s);
@@ -344,6 +359,12 @@ mac_to_unix (string s) {
   return r;
 }
 
+tree
+utf8raw_to_tree (string s) {
+  s= mac_to_unix (dos_to_unix (s));
+  return verbatim_raw_to_tree (s);
+}
+
 tree
 verbatim_to_tree (string s, bool wrap, string enc) {
   if (enc == "default") enc= "auto";
diff --git a/src/Data/Convert/convert.hpp b/src/Data/Convert/convert.hpp
index 50942a7417..d45996dd66 100644
--- a/src/Data/Convert/convert.hpp
+++ b/src/Data/Convert/convert.hpp
@@ -61,4 +61,8 @@ tree   verbatim_to_tree (string s, bool wrap= false, string enc= "default");
 tree   verbatim_document_to_tree (string s, bool w= false, string e= "default");
 bool   is_verbatim (tree t);
 
+/*** UTF-8 raw (plugin I/O) ***/
+string tree_to_utf8raw (tree t);
+tree   utf8raw_to_tree (string s);
+
 #endif // defined CONVERT_H
diff --git a/src/Edit/Interface/edit_complete.cpp b/src/Edit/Interface/edit_complete.cpp
index 4d0d0c71c2..5a419205c9 100644
--- a/src/Edit/Interface/edit_complete.cpp
+++ b/src/Edit/Interface/edit_complete.cpp
@@ -19,6 +19,7 @@
 #include "observers.hpp"
 #include "preferences.hpp"
 #include "tree_observer.hpp"
+#include "tree_traverse.hpp"
 
 #include <moebius/data/scheme.hpp>
 
@@ -355,7 +356,8 @@ edit_interface_rep::session_complete_command (tree tt) {
   (void) eval ("(use-modules (utils plugins plugin-cmd))");
   string lan= get_env_string (PROG_LANGUAGE);
   string ses= get_env_string (PROG_SESSION);
-  string s  = as_string (call ("verbatim-serialize", lan, tree_to_stree (t)));
+  tree   u  = tree_herk_to_utf8 (t);
+  string s  = as_string (call ("utf8raw-serialize", lan, tree_to_stree (u)));
   s         = s (0, N (s) - 1);
 
   int pos= search_forwards (cursor_symbol, s);
diff --git a/src/Scheme/L4/glue_convert.lua b/src/Scheme/L4/glue_convert.lua
index 3844ef8a27..4a374da662 100644
--- a/src/Scheme/L4/glue_convert.lua
+++ b/src/Scheme/L4/glue_convert.lua
@@ -130,6 +130,14 @@ function main()
                     "string"
                 }
             },
+            {
+                scm_name = "cpp-tree->utf8raw",
+                cpp_name = "tree_to_utf8raw",
+                ret_type = "string",
+                arg_list = {
+                    "tree"
+                }
+            },
             {
                 scm_name = "cpp-verbatim-snippet->texmacs",
                 cpp_name = "verbatim_to_tree",
@@ -140,6 +148,14 @@ function main()
                     "string"
                 }
             },
+            {
+                scm_name = "cpp-utf8raw->tree",
+                cpp_name = "utf8raw_to_tree",
+                ret_type = "tree",
+                arg_list = {
+                    "string"
+                }
+            },
             {
                 scm_name = "cpp-verbatim->texmacs",
                 cpp_name = "verbatim_document_to_tree",
diff --git a/src/System/Link/connection.cpp b/src/System/Link/connection.cpp
index f93cd059ee..ea84b53de2 100644
--- a/src/System/Link/connection.cpp
+++ b/src/System/Link/connection.cpp
@@ -26,6 +26,7 @@
 #include "socket_notifier.hpp"
 #include "tm_debug.hpp"
 #include "tree_helper.hpp"
+#include "tree_traverse.hpp"
 
 using namespace moebius;
 
@@ -258,7 +259,8 @@ connection_write (string name, string session, string s) {
 void
 connection_write (string name, string session, tree t) {
   // cout << "Write " << name << ", " << session << ", " << t << "\n";
-  string s= as_string (call ("plugin-serialize", name, tree_to_stree (t)));
+  tree   u= tree_herk_to_utf8 (t);
+  string s= as_string (call ("plugin-serialize", name, tree_to_stree (u)));
   connection_write (name, session, s);
 }