fix(runtime): replace tmux pane capture with explicit bounded output (ODY-150)

tests/test_agent_bash_tmux_env.py

@@ -1,74 +1,4 @@
 import asyncio
-from types import SimpleNamespace
-
-
-def test_new_tmux_session_forwards_runtime_python_environment(monkeypatch):
-    from src.agent_tools import subprocess_tools
-
-    calls = []
-
-    checks = 0
-
-    async def fake_has_session(_name):
-        nonlocal checks
-        checks += 1
-        return checks > 1
-
-    async def fake_run_exec(*args, **kwargs):
-        calls.append(args)
-        if args[:2] == ("tmux", "has-session"):
-            return "", "", 0
-        return "", "", 0
-
-    monkeypatch.setattr(subprocess_tools, "_tmux_has_session", fake_has_session)
-    monkeypatch.setattr(subprocess_tools, "_run_exec", fake_run_exec)
-    asyncio.run(subprocess_tools._ensure_tmux_session(
-        "ody-test",
-        "/workspace",
-        {
-            "PATH": "/opt/ody/bin:/usr/bin",
-            "VIRTUAL_ENV": "/opt/ody",
-            "HOME": "/workspace",
-            "TMPDIR": "/workspace/.tmp",
-            "SECRET": "must-not-forward",
-        },
-    ))
-
-    new_session = next(call for call in calls if call[:2] == ("tmux", "new-session"))
-    assert "PATH=/opt/ody/bin:/usr/bin" in new_session
-    assert "VIRTUAL_ENV=/opt/ody" in new_session
-    assert "HOME=/workspace" in new_session
-    assert "TMPDIR=/workspace/.tmp" in new_session
-    assert not any("SECRET=" in arg for arg in new_session)
-
-
-def test_reused_tmux_session_refreshes_runtime_python_environment(monkeypatch):
-    from src.agent_tools import subprocess_tools
-
-    sent = []
-
-    async def fake_has_session(_name):
-        return True
-
-    async def fake_send_line(name, line):
-        sent.append((name, line))
-
-    async def fake_run_exec(*_args, **_kwargs):
-        return "", "", 0
-
-    monkeypatch.setattr(subprocess_tools, "_tmux_has_session", fake_has_session)
-    monkeypatch.setattr(subprocess_tools, "_tmux_send_line", fake_send_line)
-    monkeypatch.setattr(subprocess_tools, "_run_exec", fake_run_exec)
-    asyncio.run(subprocess_tools._ensure_tmux_session(
-        "ody-existing",
-        "/workspace",
-        {"PATH": "/path with spaces/bin:/usr/bin", "VIRTUAL_ENV": "/path with spaces"},
-    ))
-
-    assert sent == [(
-        "ody-existing",
-        "export PATH='/path with spaces/bin:/usr/bin' VIRTUAL_ENV='/path with spaces'",
-    )]
 
 
 def test_workspace_alias_rewrite_does_not_duplicate_absolute_host_path():
@@ -97,93 +27,6 @@ def test_workspace_namespace_preserves_literal_paths_inside_scripts(tmp_path):
     assert (Path(tmp_path) / "result.txt").read_text() == "ok"
 
 
-def test_tmux_bash_runs_tool_command_with_closed_stdin(monkeypatch, tmp_path):
-    from src.agent_tools import subprocess_tools
-
-    sent = []
-    async def fake_ensure(*_args, **_kwargs):
-        return None
-
-    async def fake_send(_name, line):
-        sent.append(line)
-
-    async def fake_capture(_name):
-        start = next(line for line in sent if "__ODYSSEUS_CMD_START_" in line)
-        start = start.split("\\n")[1]
-        end_line = next(line for line in sent if "__ODYSSEUS_CMD_END_" in line)
-        end = end_line.split("\\n")[1].split("%s")[0]
-        return f"{start}\ngot-eof\n{end}0\n"
-
-    monkeypatch.setattr(subprocess_tools, "_ensure_tmux_session", fake_ensure)
-    monkeypatch.setattr(subprocess_tools, "_tmux_send_line", fake_send)
-    monkeypatch.setattr(subprocess_tools, "_tmux_capture", fake_capture)
-    monkeypatch.setattr(subprocess_tools.time, "time", lambda: 0.123456)
-
-    output, _stderr, rc, timed_out = asyncio.run(
-        subprocess_tools._run_tmux_bash(
-            "if read answer; then echo unexpected; else echo got-eof; fi",
-            session_id="test",
-            cwd=str(tmp_path),
-            env={},
-            timeout=2,
-        )
-    )
-
-    assert output == "got-eof"
-    assert rc == 0
-    assert timed_out is False
-    assert any("/bin/bash -lc" in line and "</dev/null" in line for line in sent)
-
-
-def test_tmux_bash_timeout_destroys_session_process_tree(monkeypatch, tmp_path):
-    from src.agent_tools import subprocess_tools
-
-    exec_calls = []
-    sent = []
-    clock = [-2.0]
-
-    async def fake_ensure(*_args, **_kwargs):
-        return None
-
-    async def fake_send(name, line):
-        sent.append((name, line))
-
-    async def fake_capture(_name):
-        return "still running"
-
-    async def fake_run_exec(*args, **kwargs):
-        exec_calls.append(args)
-        return "", "", 0
-
-    async def fake_sleep(_seconds):
-        return None
-
-    monkeypatch.setattr(subprocess_tools, "_ensure_tmux_session", fake_ensure)
-    monkeypatch.setattr(subprocess_tools, "_tmux_send_line", fake_send)
-    monkeypatch.setattr(subprocess_tools, "_tmux_capture", fake_capture)
-    monkeypatch.setattr(subprocess_tools, "_run_exec", fake_run_exec)
-    monkeypatch.setattr(subprocess_tools.asyncio, "sleep", fake_sleep)
-    def fake_time():
-        clock[0] += 2.0
-        return clock[0]
-
-    monkeypatch.setattr(subprocess_tools.time, "time", fake_time)
-
-    _output, _stderr, rc, timed_out = asyncio.run(
-        subprocess_tools._run_tmux_bash(
-            "grep -r needle /large/tree",
-            session_id="timeout-test",
-            cwd=str(tmp_path),
-            env={},
-            timeout=1,
-        )
-    )
-
-    assert rc == 124 and timed_out is True
-    assert any(call[:3] == ("tmux", "send-keys", "-t") for call in exec_calls)
-    assert any(call[:3] == ("tmux", "kill-session", "-t") for call in exec_calls)
-
-
 def test_direct_bash_subprocess_has_closed_stdin(monkeypatch, tmp_path):
     from src.agent_tools import subprocess_tools
     from src import tool_execution

tests/test_agent_bash_windows.py

@@ -162,7 +162,7 @@ async def test_windows_bash_does_not_use_a_stray_tmux_executable(monkeypatch, tm
     async def fail_tmux(*_args, **_kwargs):
         pytest.fail("native Windows must not enter the POSIX tmux path")
 
-    monkeypatch.setattr(subprocess_tools, "_run_tmux_bash", fail_tmux)
+    monkeypatch.setattr(subprocess_tools.asyncio, "create_subprocess_shell", fail_tmux)
 
     result = await subprocess_tools.BashTool().execute(
         "pwd",

tests/test_agent_tmux_retirement.py

@@ -17,7 +17,7 @@ async def test_a_chat_session_always_uses_the_owned_runner(monkeypatch, tmp_path
     monkeypatch.setattr(subprocess_tools.shutil, "which", lambda name: "/fake/tmux" if name == "tmux" else original(name))
     async def forbidden(*args, **kwargs):
         pytest.fail("native Bash resurrected a persistent tmux shell")
-    monkeypatch.setattr(subprocess_tools, "_run_tmux_bash", forbidden)
+    monkeypatch.setattr(subprocess_tools.asyncio, "create_subprocess_shell", forbidden)
     result = await subprocess_tools.BashTool().execute("printf ok", {"session_id": "same-chat"})
     assert result["output"] == "ok"
     assert result["teardown"]["dead"] is True

tests/test_native_execution_containment.py

@@ -151,3 +151,37 @@ async def test_capture_preserves_multibyte_text_across_chunks(native_boundary):
         [sys.executable, "-c", "import sys; sys.stdout.write('€' * 30000)"], argv=True)
     assert result.stdout == "€" * 30000
     assert result.output_truncated is False
+
+
+async def test_chat_bash_captures_more_than_2000_lines(native_boundary):
+    result = await subprocess_tools.BashTool().execute(
+        "printf 'START\\n'; for i in $(seq 1 3000); do printf 'x\\n'; done; printf 'END\\n'",
+        {"session_id": "large-output-chat"},
+    )
+    assert result["output"].startswith("START\n")
+    assert result["output"].endswith("\nEND")
+    assert len(result["output"].splitlines()) == 3002
+    assert result["output_truncated"] is False
+    assert result["teardown"]["dead"] is True
+
+
+async def test_chat_bash_reports_capture_limit_as_incomplete(native_boundary):
+    result = await subprocess_tools.BashTool().execute(
+        "for i in $(seq 1 12000); do printf 'x\\n'; done", {"session_id": "capped-chat"},
+    )
+    assert result["exit_code"] == 0
+    assert result["output_truncated"] is True
+    assert "truncated" in result["output"]
+    assert result["teardown"]["dead"] is True
+
+
+async def test_repeated_chat_calls_refresh_environment(native_boundary):
+    first = await subprocess_tools.BashTool().execute('printf "%s" "$VALUE"', {
+        "session_id": "same-chat", "subproc_env": {"PATH": "/usr/bin:/bin", "VALUE": "first"},
+    })
+    second = await subprocess_tools.BashTool().execute('printf "%s" "$VALUE"', {
+        "session_id": "same-chat", "subproc_env": {"PATH": "/usr/bin:/bin", "VALUE": "second"},
+    })
+    assert first["output"] == "first"
+    assert second["output"] == "second"
+    assert first["containment"]["id"] != second["containment"]["id"]