fix: make the remaining test suite pass on Windows

tests/test_coding_tools.py

@@ -1,5 +1,6 @@
 import asyncio
 import shlex
+import sys
 from pathlib import Path
 from time import monotonic
 
@@ -160,6 +161,14 @@ async def test_bash_tool_captures_stdout_and_exit_code(tmp_path: Path) -> None:
     assert result.data["timed_out"] is False
 
 
+# These tests rely on bash behavior (shell_command_prefix routes through bash,
+# and job control like `sleep 1 & wait`) that only applies on POSIX.
+requires_posix_shell = pytest.mark.skipif(
+    sys.platform == "win32", reason="bash-specific shell behavior is POSIX-only"
+)
+
+
+@requires_posix_shell
 @pytest.mark.anyio
 async def test_create_coding_tools_applies_shell_command_prefix(
     tmp_path: Path,
@@ -178,6 +187,7 @@ async def test_create_coding_tools_applies_shell_command_prefix(
     assert result.data["shell_command_prefix_applied"] is True
 
 
+@requires_posix_shell
 @pytest.mark.anyio
 async def test_bash_tool_applies_opt_in_shell_command_prefix(tmp_path: Path) -> None:
     rc_path = tmp_path / ".zshrc"
@@ -227,6 +237,7 @@ async def test_bash_tool_timeout_kills_shell_children(tmp_path: Path) -> None:
     assert not marker.exists()
 
 
+@requires_posix_shell
 @pytest.mark.anyio
 async def test_bash_tool_cancellation_kills_shell_children(tmp_path: Path) -> None:
     tool = create_bash_tool(cwd=tmp_path)

tests/test_credentials.py

@@ -1,3 +1,4 @@
+import os
 from stat import S_IMODE
 
 import pytest
@@ -12,7 +13,9 @@ def test_file_credential_store_round_trips_and_sets_private_permissions(tmp_path
     store.set("openai", "test-key")
 
     assert store.get("openai") == "test-key"
-    assert S_IMODE(path.stat().st_mode) == 0o600
+    if os.name == "posix":
+        # chmod(0o600) is a no-op for these bits on Windows filesystems.
+        assert S_IMODE(path.stat().st_mode) == 0o600
 
 
 def test_file_credential_store_deletes_key(tmp_path) -> None:

tests/test_system_prompt.py

@@ -34,7 +34,9 @@ def test_default_prompt_includes_tools_guidelines_date_and_cwd(tmp_path: Path) -
     assert "Available tools:\n- read: Read file contents" in prompt
     assert "- Use bash for file operations like ls, rg, find" in prompt
     assert "- Use read to examine files instead of cat or sed." in prompt
-    assert prompt.endswith(f"Current date: 2026-06-17\nCurrent working directory: {tmp_path}")
+    # build_system_prompt normalizes the cwd to forward slashes on every platform.
+    expected_cwd = str(tmp_path).replace("\\", "/")
+    assert prompt.endswith(f"Current date: 2026-06-17\nCurrent working directory: {expected_cwd}")
 
 
 def test_tool_without_prompt_snippet_is_hidden_from_available_tools() -> None:

tests/test_tui_app.py

@@ -470,9 +470,10 @@ def test_session_sidebar_lists_multiple_context_files() -> None:
     console.print(render_session_sidebar(session))
 
     output = console.export_text()
+    # The sidebar renders labels via str(Path(...)), so separators are native.
     assert "AGENTS.md" in output
-    assert ".agents/AGENTS.md" in output
-    assert "docs/AGENTS.md" in output
+    assert str(Path(".agents") / "AGENTS.md") in output
+    assert str(Path("docs") / "AGENTS.md") in output
 
 
 def test_compact_session_info_renders_sidebar_facts() -> None:
@@ -481,7 +482,7 @@ def test_compact_session_info_renders_sidebar_facts() -> None:
     console.print(render_compact_session_info(FakeSession()))
 
     output = console.export_text()
-    assert "/workspace/project (--)" in output
+    assert f"{Path('/workspace/project')} (--)" in output
     assert "12k/200k context" in output
     assert "openai:fake-model" in output
     assert "(medium)" in output
@@ -2648,7 +2649,9 @@ def fake_notify(message: str, **kwargs: object) -> None:
         await pilot.press("enter")
 
         assert session.export_calls == [(Path("out.jsonl"), "jsonl")]
-        assert notifications == ["Exported session to /workspace/project/session.html"]
+        assert notifications == [
+            f"Exported session to {Path('/workspace/project/session.html')}"
+        ]
         assert session.prompt_texts == []
 
 
@@ -2911,7 +2914,7 @@ async def test_tui_app_completes_resume_session_argument() -> None:
 
         assert app._completion_state.selected is not None
         assert app._completion_state.selected.description == (
-            "Session - fake-model - /workspace/project"
+            f"Session - fake-model - {Path('/workspace/project')}"
         )
 
         await pilot.press("tab")