fix(agent): read document attachments on demand (#5122)

tests/test_context_documents.py

@@ -1,8 +1,8 @@
 """Tests for context builder media handling.
 
-The ContextBuilder._build_user_content method should ONLY handle images.
-Document text extraction is the responsibility of the processing layer
-(AgentLoop._process_message and _drain_pending).
+The ContextBuilder.build_user_content method should ONLY handle images.
+The processing layer turns non-image media into attachment path references;
+document contents are read on demand through ``read_file``.
 """
 
 from __future__ import annotations
@@ -10,7 +10,6 @@ from __future__ import annotations
 from pathlib import Path
 
 from nanobot.agent.context import ContextBuilder
-from nanobot.utils.document import extract_documents
 
 
 def _make_builder(tmp_path: Path) -> ContextBuilder:
@@ -20,7 +19,7 @@ def _make_builder(tmp_path: Path) -> ContextBuilder:
 
 def test_build_user_content_with_no_media_returns_string(tmp_path: Path) -> None:
     builder = _make_builder(tmp_path)
-    result = builder._build_user_content("hello", None)
+    result = builder.build_user_content("hello", None)
     assert result == "hello"
 
 
@@ -29,7 +28,7 @@ def test_build_user_content_with_image_returns_list(tmp_path: Path) -> None:
     builder = _make_builder(tmp_path)
     png = tmp_path / "test.png"
     png.write_bytes(b"\x89PNG\r\n\x1a\n" + b"\x00" * 100)
-    result = builder._build_user_content("describe this", [str(png)])
+    result = builder.build_user_content("describe this", [str(png)])
     assert isinstance(result, list)
     types = [b["type"] for b in result]
     assert "image_url" in types
@@ -41,7 +40,7 @@ def test_build_user_content_ignores_non_image_files(tmp_path: Path) -> None:
     builder = _make_builder(tmp_path)
     txt = tmp_path / "notes.txt"
     txt.write_text("some text", encoding="utf-8")
-    result = builder._build_user_content("summarize", [str(txt)])
+    result = builder.build_user_content("summarize", [str(txt)])
     assert result == "summarize"
 
 
@@ -53,81 +52,8 @@ def test_build_user_content_mixed_image_and_non_image(tmp_path: Path) -> None:
     txt = tmp_path / "report.txt"
     txt.write_text("report text", encoding="utf-8")
 
-    result = builder._build_user_content("analyze", [str(png), str(txt)])
+    result = builder.build_user_content("analyze", [str(png), str(txt)])
     assert isinstance(result, list)
     assert any(b["type"] == "image_url" for b in result)
     text_parts = [b.get("text", "") for b in result if b.get("type") == "text"]
     assert all("report text" not in t for t in text_parts)
-
-
-# ---------------------------------------------------------------------------
-# Bug detection: extract_documents must be called BEFORE _build_user_content
-# to prevent document media from being silently dropped.
-# This simulates the _drain_pending code path.
-# ---------------------------------------------------------------------------
-
-def test_drain_pending_path_preserves_document_text(tmp_path: Path) -> None:
-    """Simulates the _drain_pending path: a pending follow-up message
-    with a document attachment must have its text extracted before being
-    passed to _build_user_content.  Without extract_documents, the
-    document is silently dropped."""
-    from docx import Document
-
-    doc = Document()
-    doc.add_paragraph("Quarterly revenue is $5M")
-    docx_path = tmp_path / "report.docx"
-    doc.save(docx_path)
-
-    content = "summarize"
-    media = [str(docx_path)]
-
-    # Step 1: extract_documents separates docs from images
-    new_content, image_only = extract_documents(content, media)
-
-    # Step 2: _build_user_content handles only images (none left here)
-    builder = _make_builder(tmp_path)
-    result = builder._build_user_content(new_content, image_only if image_only else None)
-
-    # The document text should be present in the final content
-    assert "Quarterly revenue" in result
-    assert "summarize" in result
-
-
-def test_drain_pending_path_preserves_docx_table_text(tmp_path: Path) -> None:
-    """Uploaded Word forms must retain content stored in table cells."""
-    from docx import Document
-
-    doc = Document()
-    table = doc.add_table(rows=2, cols=2)
-    table.cell(0, 0).text = "Applicant"
-    table.cell(0, 1).text = "Ada Lovelace"
-    table.cell(1, 0).text = "Research area"
-    table.cell(1, 1).text = "Analytical engines"
-    docx_path = tmp_path / "application.docx"
-    doc.save(docx_path)
-
-    content, image_only = extract_documents("summarize", [str(docx_path)])
-
-    assert image_only == []
-    assert "Applicant\tAda Lovelace" in content
-    assert "Research area\tAnalytical engines" in content
-
-
-def test_drain_pending_path_without_extract_loses_document(tmp_path: Path) -> None:
-    """Demonstrates the BUG: if _drain_pending calls _build_user_content
-    directly without extract_documents, document content is lost."""
-    from docx import Document
-
-    doc = Document()
-    doc.add_paragraph("Secret data in document")
-    docx_path = tmp_path / "report.docx"
-    doc.save(docx_path)
-
-    builder = _make_builder(tmp_path)
-
-    # Bug path: call _build_user_content directly with document media
-    result = builder._build_user_content("summarize", [str(docx_path)])
-
-    # The document text is LOST — _build_user_content ignores non-images
-    assert result == "summarize"  # only the original text, no doc content
-    assert "Secret data" not in result