Werkzeugkasten: Anhaenge lesen - Word, Excel, Text (#692)

- Core/Documents/DocumentText: text of .docx (paragraphs, tables as rows),
  .xlsx (sheet by sheet, every cell at its column, stored values not
  formulas) and plain text/CSV. No DTD, no external references, every part
  bounded against zip bombs. PDF is named as not readable yet.
- Mail: read_message lists attachments; read_attachment returns the text;
  save_attachment keeps the file in the agent's workspace (never over an
  existing file, never outside).
- FileRW read/read_many return Word and Excel files as text; writing them
  stays refused.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

src/ClawdDotNet.Tools.FileRW/FileRWTool.cs

@@ -592,7 +592,12 @@ public sealed class FileRWTool : IAgentTool, IToolStateProbe
                 ? je.EnumerateArray().Select(x => x.GetString()?.ToLowerInvariant()).ToList()
                 : new List<string?> { ".txt", ".json", ".md", ".html", ".js", ".css" }; // Default
 
-            if (!allowedExtensions.Contains(extension))
+            // #692: A Word file or a spreadsheet is read as its extracted text, never raw and never
+            // written by this tool — so reading it needs no place on the list.
+            var reading = input.ValueKind == JsonValueKind.Object && input.TryGetProperty("action", out var act)
+                          && act.ValueKind == JsonValueKind.String && act.GetString() is "read" or "read_many";
+            if (!allowedExtensions.Contains(extension)
+                && !(reading && ClawdDotNet.Core.Documents.DocumentText.IsOfficeDocument(fullPath)))
             {
                 throw new UnauthorizedAccessException($"File extension '{extension}' is not allowed in workspace '{area.Name}'.");
             }
@@ -668,7 +673,7 @@ public sealed class FileRWTool : IAgentTool, IToolStateProbe
         await fileLock.WaitAsync(ct);
         try
         {
-            var content = await File.ReadAllTextAsync(path, ct);
+            var content = await ReadContentAsync(path, ct);
             NoteRead(path, area, context);
             var relativePath = Path.GetRelativePath(rootPath, path);
             return ToolResult.Ok($"[{area.Name}:/{relativePath}]\n{content}");
@@ -679,6 +684,20 @@ public sealed class FileRWTool : IAgentTool, IToolStateProbe
         }
     }
 
+    /// <summary>
+    /// #692: A Word file or a spreadsheet is read as its text — a price list saved from a mail is
+    /// a table, not a heap of zip bytes. Everything else as before.
+    /// </summary>
+    private static async Task<string> ReadContentAsync(string path, CancellationToken ct)
+    {
+        if (!ClawdDotNet.Core.Documents.DocumentText.IsOfficeDocument(path))
+            return await File.ReadAllTextAsync(path, ct);
+        var extracted = ClawdDotNet.Core.Documents.DocumentText.Extract(await File.ReadAllBytesAsync(path, ct), path, MaxZeichenJeAufruf);
+        return extracted.Ok
+            ? $"(text of a {extracted.Kind}{(extracted.Truncated ? $", first {MaxZeichenJeAufruf} characters" : "")})\n{extracted.Text}"
+            : $"({extracted.Problem})";
+    }
+
     /// <summary>Hoechstzahl Dateien je <c>read_many</c>.</summary>
     public const int MaxDateienJeAufruf = 50;
 
@@ -730,7 +749,7 @@ public sealed class FileRWTool : IAgentTool, IToolStateProbe
             string inhalt;
             try
             {
-                inhalt = await File.ReadAllTextAsync(pfad, ct);
+                inhalt = await ReadContentAsync(pfad, ct);
                 NoteRead(pfad, area, context);
             }
             finally { sperre.Release(); }