test(audit): close the four remaining minor items from the May 1 audit

CHANGELOG.md

@@ -5,6 +5,26 @@ All notable changes to this project will be documented here.
 ## [Unreleased]
 
 ### Added
+- Public `clipboard.reset_backend_cache()` helper. Replaces the previous
+  pattern of poking the module-private `_backend` global directly (which
+  the autouse fixtures in `tests/test_server.py` and
+  `tests/test_integration_x11.py` were already doing). Production code
+  has no reason to call this; it exists for tests that switch backends
+  or re-read `MCP_CLIPBOARD_BACKEND` mid-process. (#104)
+- Direct unit tests for `parser._has_header_row` covering the
+  `len(rows) < 2` early-return path and the all-text-no-header path,
+  plus a positive case for text-headers-over-numeric-data. Previously
+  exercised only indirectly via `format_table` JSON output. (#104)
+- Coverage for `clipboard_paste` `include_schema=True` when the data
+  rows are wider than the header row -- the padding loop at
+  `server.py:222-228` now has a regression test that asserts synthetic
+  `Col 3` / `Col 4` labels are emitted. (#104)
+- Coverage for `MCP_CLIPBOARD_MAX_WRITE_BYTES` and
+  `MCP_CLIPBOARD_MAX_IMAGE_BYTES` env-var validation. Both vars are
+  parsed via `int(os.environ.get(...))` at module import; a non-integer
+  value raises `ValueError` before anything else runs. Tests exercise
+  each in a subprocess so the partial-import state stays out of the
+  in-process module cache. (#104)
 - Headless X11 integration tests (`tests/test_integration_x11.py`) and a
   matching CI `integration-x11` job that runs them against a real `xclip`
   process under Xvfb. Five round-trip tests exercise plain text, unicode,

src/mcp_clipboard/clipboard.py

@@ -705,6 +705,17 @@ def _get_backend() -> str:
     return _backend
 
 
+def reset_backend_cache() -> None:
+    """Clear the cached backend so the next call to _get_backend() re-detects.
+
+    Intended for tests that need to switch backends or re-read
+    MCP_CLIPBOARD_BACKEND mid-process. Production code should not need
+    this -- the cache is a process-lifetime decision.
+    """
+    global _backend
+    _backend = None
+
+
 _READERS = {
     "wayland": _wayland_read,
     "x11": _x11_read,

tests/test_integration_x11.py

@@ -21,11 +21,11 @@
 
 import pytest
 
-import mcp_clipboard.clipboard as cb
 from mcp_clipboard.clipboard import (
     list_clipboard_formats,
     read_clipboard,
     read_clipboard_image,
+    reset_backend_cache,
     write_clipboard,
     write_clipboard_typed,
 )
@@ -47,9 +47,9 @@
 def _force_x11_backend(monkeypatch):
     """Force the X11 backend for every test in this module."""
     monkeypatch.setenv("MCP_CLIPBOARD_BACKEND", "x11")
-    cb._backend = None
+    reset_backend_cache()
     yield
-    cb._backend = None
+    reset_backend_cache()
 
 
 @pytest.mark.asyncio

tests/test_parser.py

@@ -3,6 +3,7 @@
 import re
 
 from mcp_clipboard.parser import (
+    _has_header_row,
     detect_content_type,
     extract_html_text,
     format_table,
@@ -461,6 +462,34 @@ def test_detect_code_python_def_still_detected():
     assert detect_content_type("def add(a, b):") == "code"
 
 
+# ---------------------------------------------------------------------------
+# _has_header_row direct coverage
+# ---------------------------------------------------------------------------
+
+
+def test_has_header_row_single_row():
+    """Single row triggers the early return; can't be a header without data."""
+    assert _has_header_row([["Name", "Age", "City"]]) is False
+
+
+def test_has_header_row_empty():
+    """Zero rows also takes the early-return path."""
+    assert _has_header_row([]) is False
+
+
+def test_has_header_row_uniform_text_no_header():
+    """When every row (header included) is text and data is also text, no
+    column type differs, so no header is detected."""
+    rows = [["Alice", "Bob"], ["Carol", "Dan"], ["Eve", "Frank"]]
+    assert _has_header_row(rows) is False
+
+
+def test_has_header_row_text_header_over_numeric_data():
+    """Classic header case: text labels over numeric data rows."""
+    rows = [["Count", "Total"], ["1", "100"], ["2", "200"], ["3", "300"]]
+    assert _has_header_row(rows) is True
+
+
 # ---------------------------------------------------------------------------
 # Schema inference
 # ---------------------------------------------------------------------------

tests/test_server.py

@@ -6,6 +6,7 @@
 from __future__ import annotations
 
 import asyncio
+import os
 from unittest.mock import AsyncMock, patch
 
 import pytest
@@ -167,6 +168,70 @@ async def test_paste_schema_not_appended_for_non_table():
     assert "hello world" in result
 
 
+def test_max_write_bytes_non_integer_raises_at_import():
+    """A non-integer MCP_CLIPBOARD_MAX_WRITE_BYTES raises ValueError at module
+    load (server.py runs ``int(os.environ.get(...))`` at import time).
+
+    Run as a subprocess so the bad env var and partial-import state are
+    contained -- reloading mcp_clipboard.server in-process would break
+    every test that holds a reference to the module's exception classes.
+    """
+    import subprocess
+    import sys
+
+    env = {**os.environ, "MCP_CLIPBOARD_MAX_WRITE_BYTES": "not-a-number"}
+    proc = subprocess.run(
+        [sys.executable, "-c", "import mcp_clipboard.server"],
+        env=env,
+        capture_output=True,
+        text=True,
+    )
+    assert proc.returncode != 0
+    assert "ValueError" in proc.stderr
+    assert "not-a-number" in proc.stderr
+
+
+def test_max_image_bytes_non_integer_raises_at_import():
+    """Same for MCP_CLIPBOARD_MAX_IMAGE_BYTES on the clipboard module."""
+    import subprocess
+    import sys
+
+    env = {**os.environ, "MCP_CLIPBOARD_MAX_IMAGE_BYTES": "ten-megs"}
+    proc = subprocess.run(
+        [sys.executable, "-c", "import mcp_clipboard.clipboard"],
+        env=env,
+        capture_output=True,
+        text=True,
+    )
+    assert proc.returncode != 0
+    assert "ValueError" in proc.stderr
+    assert "ten-megs" in proc.stderr
+
+
+@pytest.mark.asyncio
+async def test_paste_with_schema_pads_short_header_row():
+    """When data rows are wider than the header row, the schema table
+    pads with synthetic Col N labels (server.py:222-228 padding loop)."""
+    # Header has 2 cells, data rows have 4. Padding should add Col 3, Col 4.
+    html = (
+        "<table>"
+        "<tr><th>Name</th><th>Age</th></tr>"
+        "<tr><td>Alice</td><td>30</td><td>Portland</td><td>Engineer</td></tr>"
+        "<tr><td>Bob</td><td>25</td><td>Seattle</td><td>Designer</td></tr>"
+        "</table>"
+    )
+    with patch("mcp_clipboard.server.read_clipboard", side_effect=_mock_read(html=html)):
+        result = await clipboard_paste(output_format="markdown", include_schema=True)
+
+    assert "Column types" in result
+    # Original headers preserved
+    assert "| Name" in result
+    assert "| Age" in result
+    # Synthetic labels filled in for the wider data rows
+    assert "Col 3" in result
+    assert "Col 4" in result
+
+
 @pytest.mark.asyncio
 async def test_paste_slack_format():
     """clipboard_paste returns Slack-formatted table."""
@@ -1050,16 +1115,14 @@ def _reset_backend_cache():
 
     Isolates tests from each other: any test that mutates ``cb._backend``
     (directly or via ``_get_backend``) must not leak state to the next test.
-    Prior code did this manually with trailing ``cb._backend = None`` lines,
-    which was brittle (easy to forget; no enforcement).
     """
-    import mcp_clipboard.clipboard as cb
+    from mcp_clipboard.clipboard import reset_backend_cache
 
-    cb._backend = None
+    reset_backend_cache()
     try:
         yield
     finally:
-        cb._backend = None
+        reset_backend_cache()
 
 
 def test_detect_backend_darwin():