@@ -1,6 +1,7 @@
 package tools
 
 import (
+	"bytes"
 	"context"
 	"database/sql"
 	"encoding/json"
@@ -36,15 +37,58 @@ var dataAnalysisTool = BaseTool{
 		"the single table \"" + DataAnalysisTableName + "\"; write SQL against that table name and never put " +
 		"the document ID inside the SQL. For Excel files with multiple sheets, non-empty sheets are loaded into " +
 		"the same table and the source sheet name is exposed as a '__sheet_name' column so you can " +
-		"filter/aggregate per sheet. If the user's question requires data statistics, convert the " +
-		"question into SQL and execute it.",
+		"filter/aggregate per sheet. Only the .xlsx container is readable: a legacy binary .xls must be " +
+		"re-saved as .xlsx (or exported as .csv) before it can be analysed. If the user's question requires " +
+		"data statistics, convert the question into SQL and execute it.",
 	schema: utils.GenerateSchema[DataAnalysisInput](),
 }
 
 // excelSheetNameColumn is the name of the synthetic column that identifies
 // which Excel sheet a row came from when multiple sheets are unioned together.
 const excelSheetNameColumn = "__sheet_name"
 
+// legacyXLSMagic is the OLE2/CFB compound-document signature that binary BIFF
+// .xls workbooks start with ("D0 CF 11 E0 A1 B1 1A E1").
+//
+// DuckDB's excel extension only reads the ZIP-based OOXML container
+// (.xlsx/.xlsm), so a genuine .xls has to be rejected up front: handed to
+// read_xlsx it fails with the opaque "No xl/workbook.xml found in xlsx file",
+// which says nothing about the real cause or what to do about it.
+var legacyXLSMagic = []byte{0xD0, 0xCF, 0x11, 0xE0, 0xA1, 0xB1, 0x1A, 0xE1}
+
+// hasLegacyXLSContainer reports whether path holds an OLE2 compound document,
+// i.e. a binary BIFF .xls workbook.
+//
+// The sniff is by content, not by extension: a .xls that is really a renamed
+// .xlsx is a ZIP archive and returns false, so it keeps loading through the
+// normal read_xlsx path. Unreadable or truncated files also return false and
+// are left to DuckDB to report.
+func hasLegacyXLSContainer(path string) bool {
+	f, err := os.Open(path)
+	if err != nil {
+		return false
+	}
+	defer func() { _ = f.Close() }()
+
+	head := make([]byte, len(legacyXLSMagic))
+	if _, err := io.ReadFull(f, head); err != nil {
+		return false
+	}
+	return bytes.Equal(head, legacyXLSMagic)
+}
+
+// errLegacyBinaryXLS is the actionable failure returned for a binary .xls
+// workbook. The data analysis tool cannot read it, and silently reporting an
+// empty table (or leaking DuckDB's xlsx-internals error) leaves the model
+// guessing; naming the conversion to perform is what the caller can act on.
+func errLegacyBinaryXLS() error {
+	return fmt.Errorf(
+		"legacy binary .xls workbooks are not supported by data analysis: " +
+			"only the ZIP-based .xlsx format can be read; " +
+			"re-save the workbook as .xlsx (or export the sheet as .csv) and upload it again",
+	)
+}
+
 // sqlSingleQuoteEscape escapes single quotes in a string so it can be safely
 // embedded inside a single-quoted SQL literal.
 func sqlSingleQuoteEscape(s string) string {
@@ -547,9 +591,23 @@ func (t *DataAnalysisTool) LoadFromCSV(ctx context.Context, filename string, tab
 //
 // Note: requires the DuckDB 'excel' extension (for read_xlsx) and the
 // 'spatial' extension (for st_read_meta used to enumerate sheets).
+//
+// Binary BIFF .xls workbooks are rejected with an actionable error before
+// DuckDB is touched: the excel extension only reads the ZIP-based OOXML
+// container, so a real .xls would otherwise surface as "No xl/workbook.xml
+// found in xlsx file".
 func (t *DataAnalysisTool) LoadFromExcel(ctx context.Context, filename string, tableName string) (*TableSchema, error) {
 	logger.Infof(ctx, "[Tool][DataAnalysis] Loading Excel file '%s' into table '%s' for session %s", filename, tableName, t.sessionID)
 
+	if hasLegacyXLSContainer(filename) {
+		logger.Warnf(ctx,
+			"[Tool][DataAnalysis] Rejecting legacy binary .xls workbook '%s' (session=%s): "+
+				"DuckDB can only read the ZIP-based .xlsx container",
+			filename, t.sessionID,
+		)
+		return nil, errLegacyBinaryXLS()
+	}
+
 	if !t.loadedTables[tableName] {
 		t.recordCreatedTable(tableName)
 		sheetNames, enumErr := t.listExcelSheets(ctx, filename)
 
@@ -0,0 +1,91 @@
+package tools
+
+import (
+	"context"
+	"os"
+	"path/filepath"
+	"strings"
+	"testing"
+)
+
+// ole2Header is the compound-document signature a binary BIFF .xls starts with.
+var ole2Header = []byte{0xD0, 0xCF, 0x11, 0xE0, 0xA1, 0xB1, 0x1A, 0xE1}
+
+// writeLegacyXLS writes a file that looks like a binary BIFF .xls: the OLE2
+// signature followed by opaque payload bytes. DuckDB's read_xlsx cannot open
+// this (there is no xl/workbook.xml inside), which is exactly the case #4239
+// reports.
+func writeLegacyXLS(t *testing.T, path string) {
+	t.Helper()
+	payload := append([]byte{}, ole2Header...)
+	payload = append(payload, make([]byte, 512)...)
+	if err := os.WriteFile(path, payload, 0o600); err != nil {
+		t.Fatalf("write legacy xls fixture: %v", err)
+	}
+}
+
+func TestHasLegacyXLSContainerDetectsOLE2Workbook(t *testing.T) {
+	path := filepath.Join(t.TempDir(), "legacy.xls")
+	writeLegacyXLS(t, path)
+
+	if !hasLegacyXLSContainer(path) {
+		t.Fatalf("expected %s to be detected as a binary BIFF workbook", path)
+	}
+}
+
+func TestHasLegacyXLSContainerIgnoresZipBasedWorkbooks(t *testing.T) {
+	path := filepath.Join(t.TempDir(), "modern.xlsx")
+	writeWorkbook(t, path, map[string][][]any{
+		"Data": {{"id", "amount"}, {1, 100}},
+	}, []string{"Data"})
+
+	// A .xls that is really a renamed .xlsx is a ZIP archive and must keep
+	// loading through the normal read_xlsx path.
+	if hasLegacyXLSContainer(path) {
+		t.Fatalf("expected the ZIP-based workbook %s not to be flagged as legacy", path)
+	}
+}
+
+func TestHasLegacyXLSContainerLeavesShortAndMissingFilesAlone(t *testing.T) {
+	dir := t.TempDir()
+
+	empty := filepath.Join(dir, "empty.xls")
+	if err := os.WriteFile(empty, nil, 0o600); err != nil {
+		t.Fatalf("write empty fixture: %v", err)
+	}
+	if hasLegacyXLSContainer(empty) {
+		t.Fatalf("an empty file is not a legacy workbook")
+	}
+
+	if hasLegacyXLSContainer(filepath.Join(dir, "missing.xls")) {
+		t.Fatalf("a missing file must not be reported as a legacy workbook")
+	}
+}
+
+func TestLoadFromExcelRejectsLegacyBinaryXLSWithActionableError(t *testing.T) {
+	path := filepath.Join(t.TempDir(), "legacy.xls")
+	writeLegacyXLS(t, path)
+
+	// No DuckDB handle: the rejection has to happen before the tool touches
+	// the database, otherwise a nil db would panic instead of failing cleanly.
+	tool := &DataAnalysisTool{sessionID: "legacy-xls"}
+
+	_, err := tool.LoadFromExcel(context.Background(), path, "dataset")
+	if err == nil {
+		t.Fatal("expected a binary .xls workbook to be rejected")
+	}
+
+	message := err.Error()
+	for _, want := range []string{"legacy binary .xls", ".xlsx", ".csv"} {
+		if !strings.Contains(message, want) {
+			t.Errorf("error %q does not mention %q", message, want)
+		}
+	}
+}
+
+func TestDataAnalysisToolDescriptionNamesTheXLSConversion(t *testing.T) {
+	description := dataAnalysisTool.description
+	if !strings.Contains(description, ".xlsx") || !strings.Contains(description, "binary .xls") {
+		t.Fatalf("tool description must tell the model how to handle a legacy .xls, got: %s", description)
+	}
+}