Phase 3: PDF/DOCX extraction, chunking, LLM client with mock interface

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

internal/parse/docx.go

@@ -0,0 +1,74 @@
+package parse
+
+import (
+	"archive/zip"
+	"bytes"
+	"encoding/xml"
+	"errors"
+	"fmt"
+	"io"
+	"strings"
+)
+
+// ExtractDOCX reads a .docx file and returns its plain text.
+// DOCX is a ZIP archive; we unzip word/document.xml, walk <w:t> nodes
+// for text, and emit a newline at each <w:p> boundary.
+func ExtractDOCX(r io.Reader) (string, error) {
+	data, err := io.ReadAll(r)
+	if err != nil {
+		return "", fmt.Errorf("read docx: %w", err)
+	}
+	zr, err := zip.NewReader(bytes.NewReader(data), int64(len(data)))
+	if err != nil {
+		return "", fmt.Errorf("open docx zip: %w", err)
+	}
+	var docFile *zip.File
+	for _, f := range zr.File {
+		if f.Name == "word/document.xml" {
+			docFile = f
+			break
+		}
+	}
+	if docFile == nil {
+		return "", errors.New("word/document.xml not found in docx")
+	}
+	rc, err := docFile.Open()
+	if err != nil {
+		return "", fmt.Errorf("open document.xml: %w", err)
+	}
+	defer rc.Close()
+	return parseDocXML(rc)
+}
+
+func parseDocXML(r io.Reader) (string, error) {
+	dec := xml.NewDecoder(r)
+	var sb strings.Builder
+	var inText bool
+	for {
+		tok, err := dec.Token()
+		if err == io.EOF {
+			break
+		}
+		if err != nil {
+			return "", fmt.Errorf("parse document.xml: %w", err)
+		}
+		switch t := tok.(type) {
+		case xml.StartElement:
+			if t.Name.Local == "t" {
+				inText = true
+			}
+		case xml.EndElement:
+			if t.Name.Local == "t" {
+				inText = false
+			}
+			if t.Name.Local == "p" {
+				sb.WriteByte('\n')
+			}
+		case xml.CharData:
+			if inText {
+				sb.Write([]byte(t))
+			}
+		}
+	}
+	return strings.TrimSpace(sb.String()), nil
+}