Skip to content

Your lightweight open-source personal agent.

fix(documents): preserve complete PDF pages across reads

takiAA5f6e8a6baab8
diff --git a/nanobot/agent/tools/filesystem.py b/nanobot/agent/tools/filesystem.py
index 0510b72d0..60eedd7f7 100644
--- a/nanobot/agent/tools/filesystem.py
+++ b/nanobot/agent/tools/filesystem.py
@@ -446,6 +446,11 @@ class ReadFileTool(_FsTool):
return f"(PDF has no extractable text: {fp})"
result = extraction.text
+ if extraction.truncated:
+ result += (
+ f"\n\n(Page {extraction.end_page + 1} was truncated; "
+ "the remaining text on this page is not shown.)"
+ )
if extraction.end_page < extraction.total_pages - 1:
next_start = extraction.end_page + 2
next_end = min(extraction.end_page + 1 + self._MAX_PDF_PAGES, extraction.total_pages)
diff --git a/nanobot/utils/document.py b/nanobot/utils/document.py
index 34cfd9f78..7039a0ac3 100644
--- a/nanobot/utils/document.py
+++ b/nanobot/utils/document.py
@@ -77,6 +77,7 @@ class PdfExtraction:
total_pages: int
start_page: int
end_page: int
+ truncated: bool = False
@dataclass(frozen=True, slots=True)
@@ -485,10 +486,17 @@ def extract_pdf_pages(
f"{_MAX_PDF_CONTENT_STREAM_SIZE // (1024 * 1024)} MB limit"
)
text = (page.extract_text() or "").strip()
- if text and not collector.add(f"--- Page {index + 1} ---\n{text}", separator="\n\n"):
+ if not text:
+ continue
+ page_text = f"--- Page {index + 1} ---\n{text}"
+ if collector.parts and collector.length + 2 + len(page_text) > max_chars:
+ # Leave this page for the next read instead of returning only its prefix.
+ end = index - 1
+ break
+ if not collector.add(page_text, separator="\n\n"):
end = index
break
- return PdfExtraction(collector.render(), total_pages, start, end)
+ return PdfExtraction(collector.render(), total_pages, start, end, truncated=collector.truncated)
def _parse_pdf_page_range(pages: str | None, total_pages: int) -> tuple[int, int]:
diff --git a/tests/tools/test_read_enhancements.py b/tests/tools/test_read_enhancements.py
index 3445b42b2..bb809ad0d 100644
--- a/tests/tools/test_read_enhancements.py
+++ b/tests/tools/test_read_enhancements.py
@@ -212,9 +212,70 @@ class TestReadDedupSessionIsolation:
class TestReadPdf:
@pytest.fixture()
- def tool(self, tmp_path):
+ def tool(self, tmp_path, monkeypatch):
+ monkeypatch.setattr("nanobot.config.loader._current_config_path", tmp_path / "config.json")
return ReadFileTool(workspace=tmp_path)
+ @pytest.mark.asyncio
+ @pytest.mark.parametrize("page_count", [2, 3])
+ async def test_pdf_character_limit_preserves_pages_across_continuation(
+ self, tool, tmp_path, monkeypatch, page_count,
+ ):
+ import re
+
+ fitz = pytest.importorskip("pymupdf")
+ pdf_path = tmp_path / "report.pdf"
+ doc = fitz.open()
+ markers = []
+ for number in range(1, page_count + 1):
+ marker = f"END_OF_PAGE_{number}"
+ markers.append(marker)
+ page = doc.new_page()
+ page.insert_text((72, 72), "Report details " * 4 + "\n" + marker)
+ doc.save(pdf_path)
+ doc.close()
+ monkeypatch.setattr(tool, "_MAX_CHARS", 130)
+
+ outputs = []
+ pages = None
+ for _ in range(page_count):
+ result = await tool.execute(path=str(pdf_path), pages=pages)
+ outputs.append(result)
+ continuation = re.search(r"Use pages='([^']+)'", result)
+ if continuation is None:
+ break
+ pages = continuation[1]
+
+ combined = "\n".join(outputs)
+ for marker in markers:
+ assert combined.count(marker) == 1
+ assert "truncated" not in combined
+ assert "Use pages=" not in outputs[-1]
+
+ @pytest.mark.asyncio
+ async def test_pdf_oversized_page_reports_unreturned_text(
+ self, tool, tmp_path, monkeypatch,
+ ):
+ fitz = pytest.importorskip("pymupdf")
+ pdf_path = tmp_path / "dense.pdf"
+ doc = fitz.open()
+ page = doc.new_page()
+ page.insert_text((72, 72), "\n".join(["Dense page content " * 3] * 8))
+ page = doc.new_page()
+ page.insert_text((72, 72), "Second page complete")
+ doc.save(pdf_path)
+ doc.close()
+ monkeypatch.setattr(tool, "_MAX_CHARS", 130)
+
+ result = await tool.execute(path=str(pdf_path))
+
+ assert "Page 1 was truncated" in result
+ assert "remaining text on this page is not shown" in result
+ assert "Use pages='2-2'" in result
+ second = await tool.execute(path=str(pdf_path), pages="2")
+ assert "Second page complete" in second
+ assert "truncated" not in second
+
@pytest.mark.asyncio
async def test_pdf_returns_text_content(self, tool, tmp_path):
fitz = pytest.importorskip("fitz")