diff --git a/.claude/CLAUDE.md b/.claude/CLAUDE.md
index 6f2efe64c..353df31a6 100644
--- a/.claude/CLAUDE.md
+++ b/.claude/CLAUDE.md
@@ -118,7 +118,7 @@ Three types of tests, all must pass before committing:
 
 - **Unit tests** (`tests/unit/`): C++ tests using the project's own test framework. Test names should be at most 4 words.
 - **Integration tests** (`tests/integration/`): Python pytest tests that start a real clice server and communicate via LSP.
-- **Smoke tests** (`tests/smoke/`): Replay recorded LSP sessions via `tests/replay.py`.
+- **Smoke tests** (`tests/smoke/`): Replay recorded LSP sessions via `tests/tools/replay.py`.
 
 ### Integration Test Style
 
diff --git a/.claude/commands/test.md b/.claude/commands/test.md
index 161ff272e..dceaeb938 100644
--- a/.claude/commands/test.md
+++ b/.claude/commands/test.md
@@ -11,7 +11,7 @@ Filtering specific tests:
 
 - Unit tests: `pixi run unit-test [type] --test-filter=SuiteName.CaseName`
 - Integration tests: `pixi run pytest tests/integration -k "test_name" --executable=./build/[type]/bin/clice`
-- Smoke tests: `pixi run python tests/replay.py tests/smoke/specific.jsonl --clice=./build/[type]/bin/clice`
+- Smoke tests: `pixi run python tests/tools/replay.py tests/smoke/specific.jsonl --clice=./build/[type]/bin/clice`
 
 Example usage:
 
diff --git a/docs/en/dev/test-and-debug.md b/docs/en/dev/test-and-debug.md
index 8b5df6bfd..f9c856457 100644
--- a/docs/en/dev/test-and-debug.md
+++ b/docs/en/dev/test-and-debug.md
@@ -51,7 +51,7 @@ pixi run smoke-test Debug    # debug build
 Equivalent to:
 
 ```bash
-python tests/replay.py tests/smoke/*.jsonl \
+python tests/tools/replay.py tests/smoke/*.jsonl \
     --clice=./build/RelWithDebInfo/bin/clice
 ```
 
diff --git a/docs/zh/dev/test-and-debug.md b/docs/zh/dev/test-and-debug.md
index 9d5a1cd4f..fec65a088 100644
--- a/docs/zh/dev/test-and-debug.md
+++ b/docs/zh/dev/test-and-debug.md
@@ -51,7 +51,7 @@ pixi run smoke-test Debug    # debug 构建
 等价于：
 
 ```bash
-python tests/replay.py tests/smoke/*.jsonl \
+python tests/tools/replay.py tests/smoke/*.jsonl \
     --clice=./build/RelWithDebInfo/bin/clice
 ```
 
diff --git a/editors/vscode/.vscode-test.mjs b/editors/vscode/.vscode-test.mjs
index 5a5e000ae..786cc1a30 100644
--- a/editors/vscode/.vscode-test.mjs
+++ b/editors/vscode/.vscode-test.mjs
@@ -5,7 +5,7 @@ import { fileURLToPath } from "url";
 const repoRoot = path.resolve(path.dirname(fileURLToPath(import.meta.url)), "../..");
 
 // Each fixture under tests/data becomes one test run with that workspace.
-// Run tests/prepare.py first to generate compile databases.
+// Run tests/tools/prepare.py first to generate compile databases.
 // Keep in sync with the editor tasks in pixi.toml.
 const fixtures = ["hello_world", "modules/hover_on_imported_symbol", "header_context"];
 
diff --git a/pixi.toml b/pixi.toml
index f9af87546..26474767c 100644
--- a/pixi.toml
+++ b/pixi.toml
@@ -194,7 +194,7 @@ pytest -n auto --dist loadgroup --timeout=300 --timeout-method=thread \
 [feature.test.tasks.smoke-test]
 args = [{ arg = "type", default = "RelWithDebInfo" }]
 cmd = """
-python tests/replay.py tests/smoke/*.jsonl \
+python tests/tools/replay.py tests/smoke/*.jsonl \
     --clice=./build/{{ type }}/bin/clice
 """
 
@@ -218,7 +218,7 @@ python = ">=3.13"
 
 # Fixture list kept in sync with editors/vscode/.vscode-test.mjs.
 [feature.editor.tasks.editor-prepare]
-cmd = "python tests/prepare.py hello_world modules/hover_on_imported_symbol header_context"
+cmd = "python tests/tools/prepare.py hello_world modules/hover_on_imported_symbol header_context"
 
 # Requires nvim (stable) on PATH.
 [feature.editor.tasks.nvim-e2e]
diff --git a/tests/conftest.py b/tests/conftest.py
index a426576ef..7a7e16518 100644
--- a/tests/conftest.py
+++ b/tests/conftest.py
@@ -1,15 +1,16 @@
-import asyncio
-import os
 import shutil
-import socket
 import sys
 from pathlib import Path
 
 import pytest
 
-from tests.cdb import generate_cdb, generate_test_data_cdbs
-from tests.integration.utils.client import CliceClient
-from tests.integration.utils.assertions import assert_no_anomaly
+from tests.tools.compile_commands import generate_cdb, generate_test_data_cdbs
+from tests.tools.client import CliceClient
+from tests.tools.lifecycle import (
+    check_no_anomaly,
+    find_free_port,
+    shutdown_client,
+)
 
 
 @pytest.hookimpl(tryfirst=True, hookwrapper=True)
@@ -104,20 +105,10 @@ def workspace(request: pytest.FixtureRequest, test_data_dir: Path):
         shutil.rmtree(clice_dir, ignore_errors=True)
 
 
-def build_init_options(request: pytest.FixtureRequest, workspace: Path) -> dict:
-    """Initialization options from @pytest.mark.init_options plus test defaults."""
+def marker_init_options(request: pytest.FixtureRequest) -> dict:
+    """Initialization options from @pytest.mark.init_options, if present."""
     marker = request.node.get_closest_marker("init_options")
-    init_options = dict(marker.args[0]) if marker else {}
-    project = dict(init_options.get("project", {}))
-    # Force cache_dir into the workspace so .clice/ cleanup prevents stale PCH.
-    project["cache_dir"] = str(workspace / ".clice")
-    # One worker of each kind is enough for tests and halves the per-test
-    # process-spawn cost (5 -> 3 processes), which dominates suite time on
-    # macOS Debug. Tests needing more override via @pytest.mark.init_options.
-    project.setdefault("stateless_worker_count", 1)
-    project.setdefault("stateful_worker_count", 1)
-    init_options["project"] = project
-    return init_options
+    return dict(marker.args[0]) if marker else {}
 
 
 @pytest.fixture
@@ -133,8 +124,9 @@ async def client(
     await c.start_io(*cmd)
 
     if workspace is not None:
-        init_options = build_init_options(request, workspace)
-        await c.initialize(workspace, initialization_options=init_options)
+        await c.initialize(
+            workspace, initialization_options=marker_init_options(request)
+        )
 
     yield c
 
@@ -150,43 +142,6 @@ async def client(
         check_no_anomaly(request, c)
 
 
-def check_no_anomaly(request: pytest.FixtureRequest, c: CliceClient) -> None:
-    """Teardown gate: anomalies are internal clice bugs — every test session
-    must end without one. Tests that intentionally trigger anomalies opt out
-    with @pytest.mark.allow_anomaly and assert on them explicitly."""
-    if request.node.get_closest_marker("allow_anomaly") is not None:
-        return
-    assert_no_anomaly(c, c.workspace)
-
-
-next_port_offset = 0
-
-
-def find_free_port() -> int:
-    """Pick a port from a per-xdist-worker range.
-
-    bind(0) draws from the kernel's shared pool: two concurrent xdist
-    workers can grab the same port in the close-then-rebind gap. Disjoint
-    per-worker ranges (below the ephemeral range) remove that race; the
-    advancing offset avoids immediately reusing a just-released port.
-    """
-    global next_port_offset
-    worker = os.environ.get("PYTEST_XDIST_WORKER", "gw0")
-    suffix = worker.removeprefix("gw")
-    index = int(suffix) if suffix.isdigit() else 0
-    base = 21000 + index * 100
-    for _ in range(100):
-        port = base + next_port_offset
-        next_port_offset = (next_port_offset + 1) % 100
-        with socket.socket(socket.AF_INET, socket.SOCK_STREAM) as s:
-            try:
-                s.bind(("127.0.0.1", port))
-                return port
-            except OSError:
-                continue
-    raise RuntimeError(f"no free port in range {base}-{base + 99}")
-
-
 @pytest.fixture
 async def agentic(
     request: pytest.FixtureRequest,
@@ -202,8 +157,9 @@ async def agentic(
     await c.start_io(*cmd)
 
     if workspace is not None:
-        init_options = build_init_options(request, workspace)
-        await c.initialize(workspace, initialization_options=init_options)
+        await c.initialize(
+            workspace, initialization_options=marker_init_options(request)
+        )
 
     yield executable, host, port
 
@@ -211,102 +167,3 @@ async def agentic(
         await shutdown_client(c)
     finally:
         check_no_anomaly(request, c)
-
-
-async def make_client(executable: Path, workspace: Path) -> CliceClient:
-    """Spawn a fresh clice server and initialize it. For multi-session tests."""
-    c = CliceClient()
-    await c.start_io(str(executable), "serve")
-    await c.initialize(workspace)
-    return c
-
-
-SANITIZER_MARKERS = (
-    "AddressSanitizer",
-    "LeakSanitizer",
-    "MemorySanitizer",
-    "ThreadSanitizer",
-    "UndefinedBehaviorSanitizer",
-    "==ERROR:",
-    "runtime error:",
-)
-
-
-def server_stderr_excerpt(stderr_text: str) -> str:
-    interesting = [
-        line
-        for line in stderr_text.splitlines()
-        if "[warn]" in line
-        or "[error]" in line
-        or "Sanitizer" in line
-        or "==ERROR:" in line
-        or "runtime error:" in line
-    ]
-    return "\n".join(interesting[-80:])
-
-
-async def assert_server_exited_cleanly(server, timeout: float = 10.0) -> None:
-    failures: list[str] = []
-
-    if server is None:
-        return
-
-    if server.returncode is None:
-        try:
-            await asyncio.wait_for(server.wait(), timeout=timeout)
-        except asyncio.TimeoutError:
-            server.kill()
-            await server.wait()
-            failures.append(f"server did not exit within {timeout:g}s after shutdown")
-
-    print(f"[server] exit code: {server.returncode}", flush=True)
-
-    stderr_text = ""
-    if server.stderr:
-        try:
-            stderr_data = await asyncio.wait_for(server.stderr.read(), timeout=2.0)
-            stderr_text = stderr_data.decode("utf-8", errors="replace")
-        except Exception as exc:
-            failures.append(f"failed to collect server stderr: {exc!r}")
-
-    for line in server_stderr_excerpt(stderr_text).splitlines():
-        print(f"[server] {line}", flush=True)
-
-    if server.returncode != 0:
-        failures.append(f"server exited with code {server.returncode}")
-
-    if any(marker in stderr_text for marker in SANITIZER_MARKERS):
-        failures.append("server stderr contains sanitizer/runtime error output")
-
-    if failures:
-        excerpt = server_stderr_excerpt(stderr_text)
-        if excerpt:
-            failures.append("server stderr excerpt:\n" + excerpt)
-        pytest.fail("\n".join(failures))
-
-
-async def shutdown_client(c: CliceClient, *, verbose: bool = False) -> None:
-    """Gracefully shut down a client, force-kill if needed."""
-    try:
-        await asyncio.wait_for(c.shutdown_async(None), timeout=10.0)
-    except Exception:
-        pass
-
-    try:
-        c.exit(None)
-    except Exception:
-        pass
-
-    if verbose and c.log_messages:
-        for msg in c.log_messages:
-            level = {1: "ERROR", 2: "WARN", 3: "INFO", 4: "LOG"}.get(msg.type, "?")
-            print(f"[logMessage/{level}] {msg.message}", flush=True)
-
-    try:
-        await assert_server_exited_cleanly(c.server)
-    finally:
-        try:
-            await c.stop_io()
-            await asyncio.sleep(0.1)
-        except Exception:
-            pass
diff --git a/tests/integration/agentic/test_agentic.py b/tests/integration/agentic/test_agentic.py
index 19fad6a07..a1db68117 100644
--- a/tests/integration/agentic/test_agentic.py
+++ b/tests/integration/agentic/test_agentic.py
@@ -8,7 +8,7 @@
 
 import pytest
 
-from tests.integration.utils.wait import wait_for_index
+from tests.tools.checks import wait_for_index
 
 
 class AgenticRpcClient:
@@ -142,8 +142,12 @@ def do_request(_):
 @pytest.fixture
 async def indexed_agentic(request, executable, workspace):
     """Start server with LSP+agentic, compile a file, wait for indexing."""
-    from tests.integration.utils.client import CliceClient
-    from tests.conftest import check_no_anomaly, shutdown_client, find_free_port
+    from tests.tools.client import CliceClient
+    from tests.tools.lifecycle import (
+        check_no_anomaly,
+        shutdown_client,
+        find_free_port,
+    )
 
     host = "127.0.0.1"
     port = find_free_port()
@@ -152,8 +156,7 @@ async def indexed_agentic(request, executable, workspace):
     c = CliceClient()
     await c.start_io(*cmd)
 
-    init_options = {"project": {"cache_dir": str(workspace / ".clice")}}
-    await c.initialize(workspace, initialization_options=init_options)
+    await c.initialize(workspace)
 
     uri, _ = await c.open_and_wait(workspace / "main.cpp")
     assert await wait_for_index(c, uri, "add"), "Index not ready"
@@ -423,8 +426,11 @@ async def test_rpc_status(indexed_agentic, workspace):
 @pytest.mark.workspace("hello_world")
 async def test_rpc_shutdown(executable, workspace):
     """Shutdown notification should cause the server to exit cleanly."""
-    from tests.integration.utils.client import CliceClient
-    from tests.conftest import find_free_port, assert_server_exited_cleanly
+    from tests.tools.client import CliceClient
+    from tests.tools.lifecycle import (
+        find_free_port,
+        assert_server_exited_cleanly,
+    )
 
     host = "127.0.0.1"
     port = find_free_port()
@@ -432,8 +438,7 @@ async def test_rpc_shutdown(executable, workspace):
 
     c = CliceClient()
     await c.start_io(*cmd)
-    init_options = {"project": {"cache_dir": str(workspace / ".clice")}}
-    await c.initialize(workspace, initialization_options=init_options)
+    await c.initialize(workspace)
 
     rpc = AgenticRpcClient(host, port)
     body = json.dumps({"jsonrpc": "2.0", "method": "agentic/shutdown", "params": {}})
@@ -526,8 +531,11 @@ async def test_rpc_impact_analysis_unknown(indexed_agentic, workspace):
 
 async def test_shutdown_during_indexing(executable, tmp_path):
     """Shutdown during active background indexing must exit cleanly."""
-    from tests.integration.utils.client import CliceClient
-    from tests.conftest import find_free_port, assert_server_exited_cleanly
+    from tests.tools.client import CliceClient
+    from tests.tools.lifecycle import (
+        find_free_port,
+        assert_server_exited_cleanly,
+    )
 
     workspace = tmp_path / "ws"
     workspace.mkdir()
@@ -558,12 +566,7 @@ async def test_shutdown_during_indexing(executable, tmp_path):
     await c.start_io(*cmd)
 
     try:
-        init_options = {
-            "project": {
-                "cache_dir": str(workspace / ".clice"),
-                "idle_timeout_ms": 0,
-            }
-        }
+        init_options = {"project": {"idle_timeout_ms": 0}}
         try:
             await c.initialize(workspace, initialization_options=init_options)
         except Exception:
diff --git a/tests/integration/agentic/test_cli.py b/tests/integration/agentic/test_cli.py
index e01cfa277..1449d630a 100644
--- a/tests/integration/agentic/test_cli.py
+++ b/tests/integration/agentic/test_cli.py
@@ -6,7 +6,7 @@
 
 import pytest
 
-from tests.integration.utils.wait import wait_for_index
+from tests.tools.checks import wait_for_index
 
 
 def run_cli(executable, host, port, method, **kwargs):
@@ -30,8 +30,12 @@ def run_cli(executable, host, port, method, **kwargs):
 async def indexed_server(request, executable, workspace):
     """Start server with LSP+agentic, compile a file, wait for indexing."""
     import asyncio
-    from tests.integration.utils.client import CliceClient
-    from tests.conftest import check_no_anomaly, shutdown_client, find_free_port
+    from tests.tools.client import CliceClient
+    from tests.tools.lifecycle import (
+        check_no_anomaly,
+        shutdown_client,
+        find_free_port,
+    )
 
     host = "127.0.0.1"
     port = find_free_port()
@@ -40,8 +44,7 @@ async def indexed_server(request, executable, workspace):
     c = CliceClient()
     await c.start_io(*cmd)
 
-    init_options = {"project": {"cache_dir": str(workspace / ".clice")}}
-    await c.initialize(workspace, initialization_options=init_options)
+    await c.initialize(workspace)
 
     uri, _ = await c.open_and_wait(workspace / "main.cpp")
     assert await wait_for_index(c, uri, "add"), "Index not ready"
@@ -192,13 +195,10 @@ async def test_cli_status(indexed_server, workspace):
     assert isinstance(data["indexed"], int)
 
 
-# --- Server mode and CLI entry point tests ---
-
-
 @pytest.mark.workspace("hello_world")
 async def test_socket_mode_connects(executable, workspace):
-    from tests.conftest import find_free_port, shutdown_client
-    from tests.integration.utils.client import CliceClient
+    from tests.tools.lifecycle import find_free_port, shutdown_client
+    from tests.tools.client import CliceClient
 
     port = find_free_port()
     cmd = [str(executable), "serve", "--mode", "socket", "--port", str(port)]
diff --git a/tests/integration/compilation/test_header_pch.py b/tests/integration/compilation/test_header_pch.py
index ab90189e6..306103df0 100644
--- a/tests/integration/compilation/test_header_pch.py
+++ b/tests/integration/compilation/test_header_pch.py
@@ -6,9 +6,9 @@
 no directives of its own (bound == 0) still gets a PCH.
 """
 
-from tests.integration.utils import write_cdb
-from tests.integration.utils.assertions import assert_clean_compile
-from tests.integration.utils.cache import list_pch_files
+from tests.tools.compile_commands import write_cdb
+from tests.tools.checks import assert_clean_compile
+from tests.tools.workspace import list_pch_files
 
 
 async def test_prefix_not_reprocessed(client, tmp_path):
diff --git a/tests/integration/compilation/test_pch.py b/tests/integration/compilation/test_pch.py
index 3b22788b6..60edb92b0 100644
--- a/tests/integration/compilation/test_pch.py
+++ b/tests/integration/compilation/test_pch.py
@@ -10,10 +10,10 @@
     Position,
 )
 
-from tests.integration.utils import doc
-from tests.integration.utils.workspace import did_change
-from tests.integration.utils.wait import wait_for_recompile
-from tests.integration.utils.assertions import assert_clean_compile, assert_no_errors
+from tests.tools.workspace import doc
+from tests.tools.workspace import did_change
+from tests.tools.checks import wait_for_recompile
+from tests.tools.checks import assert_clean_compile, assert_no_errors
 
 
 @pytest.mark.workspace("pch_test")
diff --git a/tests/integration/compilation/test_persistent_cache.py b/tests/integration/compilation/test_persistent_cache.py
index 44ec1e795..de63a0244 100644
--- a/tests/integration/compilation/test_persistent_cache.py
+++ b/tests/integration/compilation/test_persistent_cache.py
@@ -15,10 +15,11 @@
     Position,
 )
 
-from tests.conftest import make_client, shutdown_client
-from tests.integration.utils import write_cdb, doc
-from tests.integration.utils.wait import MTIME_GRANULARITY, SETTLE_TIME
-from tests.integration.utils.cache import (
+from tests.tools.lifecycle import make_client, shutdown_client
+from tests.tools.compile_commands import write_cdb
+from tests.tools.workspace import doc
+from tests.tools.checks import MTIME_GRANULARITY, SETTLE_TIME
+from tests.tools.workspace import (
     cache_root,
     list_pch_files,
     list_pcm_files,
@@ -26,7 +27,8 @@
     pin_cache_to_workspace,
     read_cache_json,
 )
-from tests.integration.utils.assertions import assert_clean_compile, assert_no_anomaly
+from tests.tools.checks import assert_no_anomaly
+from tests.tools.checks import assert_clean_compile
 
 
 async def test_pch_written_to_cache_dir(client, tmp_path):
@@ -410,7 +412,7 @@ async def test_cache_wiped_while_running(client, tmp_path):
     await asyncio.sleep(1.1)
     (tmp_path / "header.h").write_text("#pragma once\nstruct W { int x; int y; };\n")
 
-    from tests.integration.utils.wait import wait_for_recompile
+    from tests.tools.checks import wait_for_recompile
 
     await wait_for_recompile(client, uri)
     assert_clean_compile(client, uri)
diff --git a/tests/integration/compilation/test_self_containment.py b/tests/integration/compilation/test_self_containment.py
index 9273453b1..9c924491b 100644
--- a/tests/integration/compilation/test_self_containment.py
+++ b/tests/integration/compilation/test_self_containment.py
@@ -7,11 +7,12 @@
 choices persist across server sessions via cache.json.
 """
 
-from tests.conftest import make_client, shutdown_client
-from tests.integration.utils import get_field, write_cdb, write_entries
-from tests.integration.utils.assertions import assert_clean_compile
-from tests.integration.utils.cache import read_cache_json
-from tests.integration.utils.wait import wait_for_recompile
+from tests.tools.lifecycle import make_client, shutdown_client
+from tests.tools.compile_commands import write_cdb, write_entries
+from tests.tools.workspace import get_field
+from tests.tools.checks import assert_clean_compile
+from tests.tools.workspace import read_cache_json
+from tests.tools.checks import wait_for_recompile
 
 
 def prefix_files(workspace):
@@ -149,7 +150,7 @@ async def test_header_save_resets_verdict(executable, tmp_path):
         VersionedTextDocumentIdentifier,
     )
 
-    from tests.integration.utils import doc
+    from tests.tools.workspace import doc
 
     (tmp_path / "types.h").write_text("#pragma once\nstruct Point { int x; int y; };\n")
     utils_h = tmp_path / "utils.h"
diff --git a/tests/integration/compilation/test_staleness.py b/tests/integration/compilation/test_staleness.py
index 8a0cd0d8e..55f68c2c3 100644
--- a/tests/integration/compilation/test_staleness.py
+++ b/tests/integration/compilation/test_staleness.py
@@ -21,19 +21,17 @@
     VersionedTextDocumentIdentifier,
 )
 
-from tests.conftest import make_client, shutdown_client
-from tests.integration.utils import write_cdb, doc
-from tests.integration.utils.cache import list_pch_files, pin_cache_to_workspace
-from tests.integration.utils.wait import (
+from tests.tools.lifecycle import make_client, shutdown_client
+from tests.tools.compile_commands import write_cdb
+from tests.tools.workspace import doc
+from tests.tools.workspace import list_pch_files, pin_cache_to_workspace
+from tests.tools.checks import (
     MTIME_GRANULARITY,
     SETTLE_TIME,
     wait_for_recompile,
 )
-from tests.integration.utils.assertions import (
-    assert_clean_compile,
-    assert_has_errors,
-    assert_no_anomaly,
-)
+from tests.tools.checks import assert_no_anomaly
+from tests.tools.checks import assert_clean_compile, assert_has_errors
 
 
 async def test_header_change_invalidates_ast(client, tmp_path):
@@ -394,7 +392,7 @@ async def test_didsave_with_module_deps(client, test_data_dir, tmp_path):
         if f.is_file():
             shutil.copy2(f, tmp_path / f.name)
 
-    from tests.cdb import generate_cdb
+    from tests.tools.compile_commands import generate_cdb
 
     generate_cdb(tmp_path)
     await client.initialize(tmp_path)
diff --git a/tests/integration/extensions/test_context_switching.py b/tests/integration/extensions/test_context_switching.py
index 25917a28b..7e2d43988 100644
--- a/tests/integration/extensions/test_context_switching.py
+++ b/tests/integration/extensions/test_context_switching.py
@@ -7,9 +7,10 @@
 
 import asyncio
 
-from tests.integration.utils import get_field, write_cdb, write_entries
-from tests.integration.utils.assertions import assert_clean_compile, assert_has_errors
-from tests.integration.utils.wait import MTIME_GRANULARITY, wait_for_recompile
+from tests.tools.compile_commands import write_cdb, write_entries
+from tests.tools.workspace import get_field
+from tests.tools.checks import assert_clean_compile, assert_has_errors
+from tests.tools.checks import MTIME_GRANULARITY, wait_for_recompile
 
 
 async def test_source_command_switch(client, tmp_path):
@@ -193,7 +194,7 @@ async def test_stale_epoch_rejected(client, tmp_path):
 
     from lsprotocol.types import DidSaveTextDocumentParams
 
-    from tests.integration.utils import doc
+    from tests.tools.workspace import doc
 
     (tmp_path / "shared.h").write_text("VALUE_TYPE get_value();\n")
     (tmp_path / "main.cpp").write_text(
@@ -282,7 +283,7 @@ async def test_saved_include_updates_hosts(client, tmp_path):
         VersionedTextDocumentIdentifier,
     )
 
-    from tests.integration.utils import doc
+    from tests.tools.workspace import doc
 
     (tmp_path / "lonely.h").write_text("inline int lonely() { return 1; }\n")
     main_cpp = tmp_path / "main.cpp"
diff --git a/tests/integration/extensions/test_header_context.py b/tests/integration/extensions/test_header_context.py
index 0ccc9bbbe..17718fca6 100644
--- a/tests/integration/extensions/test_header_context.py
+++ b/tests/integration/extensions/test_header_context.py
@@ -17,7 +17,8 @@
     TextDocumentIdentifier,
 )
 
-from tests.integration.utils import doc, get_field, write_entries
+from tests.tools.compile_commands import write_entries
+from tests.tools.workspace import doc, get_field
 
 
 @pytest.mark.workspace("header_context")
diff --git a/tests/integration/features/test_completion.py b/tests/integration/features/test_completion.py
index 44574f941..183e90360 100644
--- a/tests/integration/features/test_completion.py
+++ b/tests/integration/features/test_completion.py
@@ -9,8 +9,8 @@
     TextDocumentIdentifier,
 )
 
-from tests.integration.utils import doc
-from tests.integration.utils.workspace import did_change
+from tests.tools.workspace import doc
+from tests.tools.workspace import did_change
 
 
 @pytest.mark.workspace("include_completion")
diff --git a/tests/integration/features/test_file_tracker.py b/tests/integration/features/test_file_tracker.py
index e1a82b8fe..9e9ff0499 100644
--- a/tests/integration/features/test_file_tracker.py
+++ b/tests/integration/features/test_file_tracker.py
@@ -4,19 +4,19 @@
 
 import asyncio
 
-from tests.integration.utils import write_cdb
-from tests.integration.utils.assertions import (
+from tests.tools.compile_commands import write_cdb
+from tests.tools.checks import (
     assert_has_errors,
     assert_no_errors,
     get_errors,
 )
-from tests.integration.utils.wait import (
+from tests.tools.checks import (
     MTIME_GRANULARITY,
     wait_for_index,
     wait_for_recompile,
     wait_for_reference,
 )
-from tests.integration.utils.workspace import get_field
+from tests.tools.workspace import get_field
 
 GATED_MAIN = """\
 #ifndef FEATURE
diff --git a/tests/integration/features/test_formatting.py b/tests/integration/features/test_formatting.py
index 57d23efef..14691aa60 100644
--- a/tests/integration/features/test_formatting.py
+++ b/tests/integration/features/test_formatting.py
@@ -1,7 +1,7 @@
 import pytest
 from lsprotocol.types import Position, Range
 
-from tests.integration.utils.workspace import did_change
+from tests.tools.workspace import did_change
 
 UNFORMATTED = "int    add(   int   a  ,  int   b  ) {\nreturn   a+b ;\n}\n"
 FORMATTED = "int add(int a, int b) { return a + b; }\n"
diff --git a/tests/integration/features/test_guidance_diagnostics.py b/tests/integration/features/test_guidance_diagnostics.py
index 6a3903ec8..c7dd6fa8f 100644
--- a/tests/integration/features/test_guidance_diagnostics.py
+++ b/tests/integration/features/test_guidance_diagnostics.py
@@ -7,9 +7,10 @@
 
 from lsprotocol.types import DiagnosticSeverity
 
-from tests.conftest import make_client, shutdown_client
-from tests.integration.utils import write_cdb
-from tests.integration.utils.assertions import assert_no_anomaly, guidance_messages
+from tests.tools.lifecycle import make_client, shutdown_client
+from tests.tools.compile_commands import write_cdb
+from tests.tools.checks import assert_no_anomaly
+from tests.tools.checks import guidance_messages
 
 GUIDANCE_CODE = "inferred-compile-command"
 
diff --git a/tests/integration/features/test_header_reindex.py b/tests/integration/features/test_header_reindex.py
index 389eb7f55..5d4b65f99 100644
--- a/tests/integration/features/test_header_reindex.py
+++ b/tests/integration/features/test_header_reindex.py
@@ -10,8 +10,8 @@
     VersionedTextDocumentIdentifier,
 )
 
-from tests.integration.utils import write_cdb
-from tests.integration.utils.wait import (
+from tests.tools.compile_commands import write_cdb
+from tests.tools.checks import (
     MTIME_GRANULARITY,
     reference_uris,
     wait_for_reference,
diff --git a/tests/integration/features/test_inactive_regions.py b/tests/integration/features/test_inactive_regions.py
index 50a771d4d..e0b88a0aa 100644
--- a/tests/integration/features/test_inactive_regions.py
+++ b/tests/integration/features/test_inactive_regions.py
@@ -8,7 +8,7 @@
 
 import asyncio
 
-from tests.integration.utils import write_cdb, write_entries
+from tests.tools.compile_commands import write_cdb, write_entries
 
 
 async def wait_regions(captured, timeout=15.0):
diff --git a/tests/integration/features/test_index.py b/tests/integration/features/test_index.py
index aa0706c06..808d355eb 100644
--- a/tests/integration/features/test_index.py
+++ b/tests/integration/features/test_index.py
@@ -16,8 +16,8 @@
     WorkspaceSymbolParams,
 )
 
-from tests.integration.utils import doc
-from tests.integration.utils.wait import wait_for_index
+from tests.tools.workspace import doc
+from tests.tools.checks import wait_for_index
 
 
 @pytest.mark.workspace("index_features")
diff --git a/tests/integration/features/test_index_staleness.py b/tests/integration/features/test_index_staleness.py
index 7df1f674b..26bdf3d43 100644
--- a/tests/integration/features/test_index_staleness.py
+++ b/tests/integration/features/test_index_staleness.py
@@ -3,10 +3,10 @@
 
 import asyncio
 
-from tests.conftest import make_client, shutdown_client
-from tests.integration.utils import write_cdb
-from tests.integration.utils.cache import cache_root, pin_cache_to_workspace
-from tests.integration.utils.wait import MTIME_GRANULARITY
+from tests.tools.lifecycle import make_client, shutdown_client
+from tests.tools.compile_commands import write_cdb
+from tests.tools.workspace import cache_root, pin_cache_to_workspace
+from tests.tools.checks import MTIME_GRANULARITY
 
 HEADER = "#pragma once\ninline int alpha() { return 1; }\n"
 CLOSED_TU = '#include "header.h"\nint use() { return alpha(); }\n'
diff --git a/tests/integration/features/test_query_freshness.py b/tests/integration/features/test_query_freshness.py
index 81b1d3d9b..a65e87b53 100644
--- a/tests/integration/features/test_query_freshness.py
+++ b/tests/integration/features/test_query_freshness.py
@@ -1,8 +1,8 @@
 """Navigation right after an edit must resolve against the edited buffer:
 the server settles the file's compile before answering, with no timeout."""
 
-from tests.integration.utils import write_cdb
-from tests.integration.utils.workspace import did_change
+from tests.tools.compile_commands import write_cdb
+from tests.tools.workspace import did_change
 
 SOURCE_V1 = "int foo() { return 1; }\nint main() { return foo(); }\n"
 
diff --git a/tests/integration/features/test_server.py b/tests/integration/features/test_server.py
index 879fa4309..6b28c5028 100644
--- a/tests/integration/features/test_server.py
+++ b/tests/integration/features/test_server.py
@@ -10,9 +10,9 @@
     Range,
 )
 
-from tests.integration.utils import doc
-from tests.integration.utils.wait import SETTLE_TIME
-from tests.integration.utils.workspace import did_change
+from tests.tools.workspace import doc
+from tests.tools.checks import SETTLE_TIME
+from tests.tools.workspace import did_change
 
 
 @pytest.mark.workspace("hello_world")
diff --git a/tests/integration/lifecycle/test_anomaly.py b/tests/integration/lifecycle/test_anomaly.py
index 3c2948d25..908ee7dd9 100644
--- a/tests/integration/lifecycle/test_anomaly.py
+++ b/tests/integration/lifecycle/test_anomaly.py
@@ -13,9 +13,9 @@
 
 import pytest
 
-from tests.conftest import make_client, shutdown_client
-from tests.integration.utils import write_cdb
-from tests.integration.utils.assertions import (
+from tests.tools.lifecycle import make_client, shutdown_client
+from tests.tools.compile_commands import write_cdb
+from tests.tools.checks import (
     anomalies_in_log_files,
     anomalies_in_log_messages,
 )
diff --git a/tests/integration/lifecycle/test_config.py b/tests/integration/lifecycle/test_config.py
index 6359d3d09..a60afbe35 100644
--- a/tests/integration/lifecycle/test_config.py
+++ b/tests/integration/lifecycle/test_config.py
@@ -8,11 +8,11 @@
 import pytest
 from lsprotocol.types import DiagnosticSeverity
 
-from tests.conftest import make_client, shutdown_client
-from tests.integration.utils.assertions import (
+from tests.tools.lifecycle import make_client, shutdown_client
+from tests.tools.checks import assert_no_anomaly
+from tests.tools.checks import (
     assert_clean_compile,
     assert_has_errors,
-    assert_no_anomaly,
     get_errors,
 )
 
diff --git a/tests/integration/lifecycle/test_file_operation.py b/tests/integration/lifecycle/test_file_operation.py
index c74406ff5..01091f18b 100644
--- a/tests/integration/lifecycle/test_file_operation.py
+++ b/tests/integration/lifecycle/test_file_operation.py
@@ -12,9 +12,9 @@
     VersionedTextDocumentIdentifier,
 )
 
-from tests.integration.utils import doc
-from tests.integration.utils.wait import IDLE_TIMEOUT
-from tests.integration.utils.workspace import did_change
+from tests.tools.workspace import doc
+from tests.tools.checks import IDLE_TIMEOUT
+from tests.tools.workspace import did_change
 
 
 @pytest.mark.workspace("hello_world")
diff --git a/tests/integration/lifecycle/test_protocol_edges.py b/tests/integration/lifecycle/test_protocol_edges.py
index 1f2c31e53..5fa8a6770 100644
--- a/tests/integration/lifecycle/test_protocol_edges.py
+++ b/tests/integration/lifecycle/test_protocol_edges.py
@@ -16,10 +16,11 @@
     VersionedTextDocumentIdentifier,
 )
 
-from tests.conftest import check_no_anomaly, shutdown_client
-from tests.integration.utils.assertions import get_errors, guidance_messages
-from tests.integration.utils.client import CliceClient
-from tests.integration.utils.workspace import did_change, write_cdb, write_source
+from tests.tools.lifecycle import check_no_anomaly, shutdown_client
+from tests.tools.checks import get_errors, guidance_messages
+from tests.tools.client import CliceClient
+from tests.tools.compile_commands import write_cdb
+from tests.tools.workspace import did_change, write_source
 
 TEST_TOML = (
     '[project]\ncache_dir = "${workspace}/.clice"\nenable_indexing = false\n'
diff --git a/tests/integration/lifecycle/test_protocol_robustness.py b/tests/integration/lifecycle/test_protocol_robustness.py
index d3d00f29b..f9a718b41 100644
--- a/tests/integration/lifecycle/test_protocol_robustness.py
+++ b/tests/integration/lifecycle/test_protocol_robustness.py
@@ -6,8 +6,12 @@
 import pytest
 from lsprotocol.types import InitializeParams
 
-from tests.integration.utils.injection import build_params
-from tests.replay import SERVER_REQUEST_DEFAULTS, read_lsp_message, write_lsp_message
+from tests.tools.injection import build_params
+from tests.tools.replay import (
+    SERVER_REQUEST_DEFAULTS,
+    read_lsp_message,
+    write_lsp_message,
+)
 
 INJECTION_FLOOR = 5
 
diff --git a/tests/integration/modules/test_modules.py b/tests/integration/modules/test_modules.py
index fdb9fdd65..44e5bfde5 100644
--- a/tests/integration/modules/test_modules.py
+++ b/tests/integration/modules/test_modules.py
@@ -4,7 +4,7 @@
 import shutil
 
 import pytest
-from tests.cdb import generate_cdb
+from tests.tools.compile_commands import generate_cdb
 from lsprotocol.types import (
     DidOpenTextDocumentParams,
     HoverParams,
@@ -13,8 +13,8 @@
     TextDocumentItem,
 )
 
-from tests.integration.utils.assertions import assert_clean_compile, assert_has_errors
-from tests.integration.utils.wait import IDLE_TIMEOUT, wait_for_index
+from tests.tools.checks import assert_clean_compile, assert_has_errors
+from tests.tools.checks import IDLE_TIMEOUT, wait_for_index
 
 
 @pytest.mark.workspace("modules/single_module_no_deps")
diff --git a/tests/integration/stress/test_eviction.py b/tests/integration/stress/test_eviction.py
index 35b14593f..b80a8e31c 100644
--- a/tests/integration/stress/test_eviction.py
+++ b/tests/integration/stress/test_eviction.py
@@ -1,7 +1,7 @@
 """Worker document eviction: opening more files than a stateful worker's
 LRU cap must not silently break features on the evicted documents."""
 
-from tests.integration.utils import write_cdb
+from tests.tools.compile_commands import write_cdb
 
 FILE_COUNT = 18  # one stateful worker holds at most 16 compiled documents
 
diff --git a/tests/integration/stress/test_rapid_edit.py b/tests/integration/stress/test_rapid_edit.py
index 7c32a08f4..16a38bc0c 100644
--- a/tests/integration/stress/test_rapid_edit.py
+++ b/tests/integration/stress/test_rapid_edit.py
@@ -9,9 +9,9 @@
     Position,
 )
 
-from tests.integration.utils import doc
-from tests.integration.utils.wait import SETTLE_TIME
-from tests.integration.utils.workspace import did_change
+from tests.tools.workspace import doc
+from tests.tools.checks import SETTLE_TIME
+from tests.tools.workspace import did_change
 
 
 @pytest.mark.workspace("hello_world")
diff --git a/tests/integration/utils/__init__.py b/tests/integration/utils/__init__.py
deleted file mode 100644
index e51135231..000000000
--- a/tests/integration/utils/__init__.py
+++ /dev/null
@@ -1,40 +0,0 @@
-"""Shared utilities for clice integration tests."""
-
-from tests.integration.utils.client import CliceClient
-from tests.integration.utils.workspace import (
-    doc,
-    get_field,
-    write_cdb,
-    write_entries,
-    write_source,
-)
-from tests.integration.utils.assertions import (
-    assert_no_errors,
-    assert_has_errors,
-    assert_diagnostics_count,
-)
-from tests.integration.utils.wait import wait_for_recompile, wait_for_index
-from tests.integration.utils.cache import (
-    list_pch_files,
-    list_pcm_files,
-    list_tmp_files,
-    read_cache_json,
-)
-
-__all__ = [
-    "get_field",
-    "write_entries",
-    "CliceClient",
-    "doc",
-    "write_cdb",
-    "write_source",
-    "assert_no_errors",
-    "assert_has_errors",
-    "assert_diagnostics_count",
-    "wait_for_recompile",
-    "wait_for_index",
-    "list_pch_files",
-    "list_pcm_files",
-    "list_tmp_files",
-    "read_cache_json",
-]
diff --git a/tests/integration/utils/wait.py b/tests/integration/utils/wait.py
deleted file mode 100644
index b0807eb94..000000000
--- a/tests/integration/utils/wait.py
+++ /dev/null
@@ -1,78 +0,0 @@
-"""Wait and polling helpers for integration tests."""
-
-import asyncio
-
-from lsprotocol.types import (
-    HoverParams,
-    Position,
-    TextDocumentIdentifier,
-    WorkspaceSymbolParams,
-)
-
-# Standard timing constants — use these instead of hardcoded sleep values.
-MTIME_GRANULARITY = 1.1  # Filesystem mtime precision (1s on some FSes, +0.1 margin)
-SETTLE_TIME = 0.5  # Time for the server to stabilize after an operation
-IDLE_TIMEOUT = 5.0  # Idle soak time in lifecycle tests
-
-
-async def wait_for_recompile(client, uri: str, *, timeout: float = 60.0) -> None:
-    """Trigger recompilation via hover and wait for fresh diagnostics.
-
-    Useful after didChange or on-disk file modifications. Sends a hover
-    request at (0,0) to trigger ensure_compiled(), then waits for the
-    resulting diagnostics notification.
-    """
-    event = client.wait_for_diagnostics(uri)
-    await client.text_document_hover_async(
-        HoverParams(
-            text_document=TextDocumentIdentifier(uri=uri),
-            position=Position(line=0, character=0),
-        )
-    )
-    await asyncio.wait_for(event.wait(), timeout=timeout)
-
-
-async def wait_for_index(
-    client,
-    uri: str,
-    symbol_name: str = "add",
-    *,
-    timeout: int = 30,
-) -> bool:
-    """Poll workspace/symbol until a specific symbol appears in the index.
-
-    Sends a hover to trigger compilation/indexing, then polls every second.
-    Returns True if the symbol was found, False on timeout.
-    """
-    await client.text_document_hover_async(
-        HoverParams(
-            text_document=TextDocumentIdentifier(uri=uri),
-            position=Position(line=0, character=0),
-        )
-    )
-
-    for _ in range(timeout):
-        result = await client.workspace_symbol_async(
-            WorkspaceSymbolParams(query=symbol_name)
-        )
-        if result and any(s.name == symbol_name for s in result):
-            return True
-        await asyncio.sleep(1)
-    return False
-
-
-async def reference_uris(client, uri: str, line: int, character: int) -> list[str]:
-    """URIs of the references at a position (declaration excluded)."""
-    refs = await client.references_at(uri, line, character, include_declaration=False)
-    return [ref.uri for ref in (refs or [])]
-
-
-async def wait_for_reference(
-    client, uri: str, line: int, character: int, expected_uri: str, timeout: int = 30
-) -> bool:
-    """Poll references at a position until expected_uri shows up."""
-    for _ in range(timeout):
-        if expected_uri in await reference_uris(client, uri, line, character):
-            return True
-        await asyncio.sleep(1)
-    return False
diff --git a/tests/integration/utils/workspace.py b/tests/integration/utils/workspace.py
deleted file mode 100644
index 4333a8032..000000000
--- a/tests/integration/utils/workspace.py
+++ /dev/null
@@ -1,97 +0,0 @@
-"""Workspace and file utilities for integration tests."""
-
-import json
-from pathlib import Path
-
-from lsprotocol.types import (
-    DidChangeTextDocumentParams,
-    TextDocumentContentChangeWholeDocument,
-    TextDocumentIdentifier,
-    VersionedTextDocumentIdentifier,
-)
-
-
-def doc(uri: str) -> TextDocumentIdentifier:
-    """Create a TextDocumentIdentifier from a URI string."""
-    return TextDocumentIdentifier(uri=uri)
-
-
-def write_cdb(
-    workspace: Path,
-    files: list[str],
-    *,
-    extra_args: list[str] | None = None,
-    std: str = "c++17",
-) -> None:
-    """Write a compile_commands.json for the given source files.
-
-    Args:
-        workspace: Root directory of the workspace.
-        files: List of source file names (relative to workspace).
-        extra_args: Additional compiler arguments (e.g. ["-DFOO", "-I/bar"]).
-        std: C++ standard version (default: c++17).
-    """
-    entries = []
-    for f in files:
-        args = ["clang++", f"-std={std}", "-fsyntax-only"]
-        if extra_args:
-            args.extend(extra_args)
-        args.append(str(workspace / f))
-        entries.append(
-            {
-                "directory": str(workspace),
-                "file": str(workspace / f),
-                "arguments": args,
-            }
-        )
-    (workspace / "compile_commands.json").write_text(json.dumps(entries, indent=2))
-
-
-def write_source(workspace: Path, name: str, content: str) -> Path:
-    """Write a source file to the workspace directory. Returns the file path."""
-    path = workspace / name
-    path.parent.mkdir(parents=True, exist_ok=True)
-    path.write_text(content)
-    return path
-
-
-def did_change(client, uri: str, version: int, text: str) -> None:
-    """Send a didChange notification with whole-document replacement."""
-    client.text_document_did_change(
-        DidChangeTextDocumentParams(
-            text_document=VersionedTextDocumentIdentifier(uri=uri, version=version),
-            content_changes=[TextDocumentContentChangeWholeDocument(text=text)],
-        )
-    )
-
-
-def get_field(obj, key, default=None):
-    """Read a field from a dict or attribute-style LSP response object."""
-    if isinstance(obj, dict):
-        return obj.get(key, default)
-    return getattr(obj, key, default)
-
-
-def write_entries(workspace, entries):
-    """Write a compile_commands.json with per-file extra arguments.
-
-    Args:
-        workspace: Root directory of the workspace.
-        entries: List of (file_name, extra_args) pairs; a file may appear
-            multiple times to model multi-configuration projects.
-    """
-    data = [
-        {
-            "directory": str(workspace),
-            "file": str(workspace / f),
-            "arguments": [
-                "clang++",
-                "-std=c++17",
-                "-fsyntax-only",
-                *args,
-                str(workspace / f),
-            ],
-        }
-        for f, args in entries
-    ]
-    (workspace / "compile_commands.json").write_text(json.dumps(data))
diff --git a/tests/tools/__init__.py b/tests/tools/__init__.py
new file mode 100644
index 000000000..e69de29bb
diff --git a/tests/integration/utils/assertions.py b/tests/tools/checks.py
similarity index 55%
rename from tests/integration/utils/assertions.py
rename to tests/tools/checks.py
index 73da2c183..384aef84c 100644
--- a/tests/integration/utils/assertions.py
+++ b/tests/tools/checks.py
@@ -1,9 +1,86 @@
-"""Diagnostic and anomaly assertion helpers for integration tests."""
+"""Observation and assertion helpers: diagnostics, anomalies, waits."""
 
+import asyncio
 import re
 from pathlib import Path
 
-from lsprotocol.types import Diagnostic, DiagnosticSeverity
+from lsprotocol.types import (
+    Diagnostic,
+    DiagnosticSeverity,
+    HoverParams,
+    Position,
+    TextDocumentIdentifier,
+    WorkspaceSymbolParams,
+)
+
+# Standard timing constants — use these instead of hardcoded sleep values.
+MTIME_GRANULARITY = 1.1  # Filesystem mtime precision (1s on some FSes, +0.1 margin)
+SETTLE_TIME = 0.5  # Time for the server to stabilize after an operation
+IDLE_TIMEOUT = 5.0  # Idle soak time in lifecycle tests
+
+
+async def wait_for_recompile(client, uri: str, *, timeout: float = 60.0) -> None:
+    """Trigger recompilation via hover and wait for fresh diagnostics.
+
+    Useful after didChange or on-disk file modifications. Sends a hover
+    request at (0,0) to trigger ensure_compiled(), then waits for the
+    resulting diagnostics notification.
+    """
+    event = client.wait_for_diagnostics(uri)
+    await client.text_document_hover_async(
+        HoverParams(
+            text_document=TextDocumentIdentifier(uri=uri),
+            position=Position(line=0, character=0),
+        )
+    )
+    await asyncio.wait_for(event.wait(), timeout=timeout)
+
+
+async def wait_for_index(
+    client,
+    uri: str,
+    symbol_name: str = "add",
+    *,
+    timeout: int = 30,
+) -> bool:
+    """Poll workspace/symbol until a specific symbol appears in the index.
+
+    Sends a hover to trigger compilation/indexing, then polls every second.
+    Returns True if the symbol was found, False on timeout.
+    """
+    await client.text_document_hover_async(
+        HoverParams(
+            text_document=TextDocumentIdentifier(uri=uri),
+            position=Position(line=0, character=0),
+        )
+    )
+
+    for _ in range(timeout):
+        result = await client.workspace_symbol_async(
+            WorkspaceSymbolParams(query=symbol_name)
+        )
+        if result and any(s.name == symbol_name for s in result):
+            return True
+        await asyncio.sleep(1)
+    return False
+
+
+async def reference_uris(client, uri: str, line: int, character: int) -> list[str]:
+    """URIs of the references at a position (declaration excluded)."""
+    refs = await client.references_at(uri, line, character, include_declaration=False)
+    return [ref.uri for ref in (refs or [])]
+
+
+async def wait_for_reference(
+    client, uri: str, line: int, character: int, expected_uri: str, timeout: int = 30
+) -> bool:
+    """Poll references at a position until expected_uri shows up."""
+    for _ in range(timeout):
+        if expected_uri in await reference_uris(client, uri, line, character):
+            return True
+        await asyncio.sleep(1)
+    return False
+
 
 ANOMALY_PATTERN = re.compile(r"\[anomaly:([A-Za-z]+)\]")
 
diff --git a/tests/integration/utils/client.py b/tests/tools/client.py
similarity index 92%
rename from tests/integration/utils/client.py
rename to tests/tools/client.py
index 573f4797e..fda469e83 100644
--- a/tests/integration/utils/client.py
+++ b/tests/tools/client.py
@@ -89,8 +89,6 @@ def on_progress(params: ProgressParams) -> None:
             token = str(params.token) if isinstance(params.token, int) else params.token
             self.progress_events.append({"token": token, "value": params.value})
 
-    # ── URI helpers ──────────────────────────────────────────────────
-
     @staticmethod
     def normalize_uri(uri: str) -> str:
         return unquote(uri)
@@ -98,8 +96,6 @@ def normalize_uri(uri: str) -> str:
     def path_to_uri(self, filepath: Path) -> str:
         return self.normalize_uri(filepath.as_uri())
 
-    # ── Lifecycle ────────────────────────────────────────────────────
-
     async def initialize(
         self,
         workspace: Path,
@@ -109,7 +105,15 @@ async def initialize(
         if initialization_options is None:
             initialization_options = {}
         project = dict(initialization_options.get("project", {}))
-        project.setdefault("cache_dir", str(workspace / ".clice"))
+        # Force cache_dir into the workspace so .clice/ cleanup prevents
+        # stale PCH.
+        project["cache_dir"] = str(workspace / ".clice")
+        # One worker of each kind is enough for tests and halves the
+        # per-test process-spawn cost (5 -> 3 processes), which dominates
+        # suite time on macOS Debug. Tests needing more pass their own
+        # counts via initialization_options.
+        project.setdefault("stateless_worker_count", 1)
+        project.setdefault("stateful_worker_count", 1)
         initialization_options["project"] = project
         # Disable the stat-polling loops: tests drive ticks deterministically
         # through the clice/internal/poll hook instead.
@@ -130,7 +134,6 @@ async def initialize(
         self.workspace = workspace
         return result
 
-    # ── Server process control ───────────────────────────────────────
     # Single home for the pygls internals these wrap; tests must not poke
     # at _server/_stop_event/_async_tasks directly.
 
@@ -152,8 +155,6 @@ async def stop_io(self) -> None:
         # Wait the cancellations out so no task outlives the test teardown.
         await asyncio.gather(*self._async_tasks, return_exceptions=True)
 
-    # ── Document operations ──────────────────────────────────────────
-
     def open(self, filepath: Path, version: int = 0) -> tuple[str, str]:
         """Open a text document. Returns (normalized_uri, content)."""
         content = filepath.read_bytes().decode("utf-8")
@@ -173,8 +174,6 @@ def close(self, uri: str) -> None:
             DidCloseTextDocumentParams(text_document=TextDocumentIdentifier(uri=uri))
         )
 
-    # ── Diagnostics ──────────────────────────────────────────────────
-
     def wait_for_diagnostics(self, uri: str) -> asyncio.Event:
         uri = self.normalize_uri(uri)
         if uri not in self.diagnostics_events:
@@ -192,8 +191,6 @@ async def wait_diagnostics(self, uri: str, timeout: float = 30.0) -> None:
             return
         await asyncio.wait_for(event.wait(), timeout=timeout)
 
-    # ── Compile & wait ───────────────────────────────────────────────
-
     async def open_and_wait(
         self, filepath: Path, timeout: float = 60.0
     ) -> tuple[str, str]:
@@ -209,8 +206,6 @@ async def open_and_wait(
         await asyncio.wait_for(event.wait(), timeout=timeout)
         return uri, content
 
-    # ── Feature request shortcuts ────────────────────────────────────
-
     async def hover_at(
         self, uri: str, line: int, character: int, *, timeout: float = 30.0
     ):
@@ -420,8 +415,6 @@ async def format_range(self, uri: str, range_: Range, *, timeout: float = 30.0):
             timeout=timeout,
         )
 
-    # ── Extension protocol ───────────────────────────────────────────
-
     async def query_context(
         self, uri: str, *, offset: int | None = None, timeout: float = 30.0
     ):
diff --git a/tests/cdb.py b/tests/tools/compile_commands.py
similarity index 66%
rename from tests/cdb.py
rename to tests/tools/compile_commands.py
index 9c3f5bc49..06ded833f 100644
--- a/tests/cdb.py
+++ b/tests/tools/compile_commands.py
@@ -1,7 +1,7 @@
 """Compilation database generation for test fixtures.
 
 Stdlib-only so it can be used both by pytest (conftest.py) and by
-standalone scripts (tests/editors/prepare.py) without pulling in
+standalone scripts (tests/tools/prepare.py) without pulling in
 pytest/pygls dependencies.
 """
 
@@ -16,7 +16,7 @@ def generate_cdb(workspace: Path) -> None:
     cmake = shutil.which("cmake")
     if cmake is None:
         raise RuntimeError("cmake executable not found in PATH")
-    toolchain = Path(__file__).resolve().parent.parent / "cmake" / "toolchain.cmake"
+    toolchain = Path(__file__).resolve().parents[2] / "cmake" / "toolchain.cmake"
     cmd = [
         cmake,
         "-G",
@@ -113,3 +113,59 @@ def entry(directory: Path, source: Path, extra_args: list[str] | None = None):
                 entries.append(entry(pt_dir, src))
         if entries:
             write(pt_dir, entries)
+
+
+def write_cdb(
+    workspace: Path,
+    files: list[str],
+    *,
+    extra_args: list[str] | None = None,
+    std: str = "c++17",
+) -> None:
+    """Write a compile_commands.json for the given source files.
+
+    Args:
+        workspace: Root directory of the workspace.
+        files: List of source file names (relative to workspace).
+        extra_args: Additional compiler arguments (e.g. ["-DFOO", "-I/bar"]).
+        std: C++ standard version (default: c++17).
+    """
+    entries = []
+    for f in files:
+        args = ["clang++", f"-std={std}", "-fsyntax-only"]
+        if extra_args:
+            args.extend(extra_args)
+        args.append(str(workspace / f))
+        entries.append(
+            {
+                "directory": str(workspace),
+                "file": str(workspace / f),
+                "arguments": args,
+            }
+        )
+    (workspace / "compile_commands.json").write_text(json.dumps(entries, indent=2))
+
+
+def write_entries(workspace, entries):
+    """Write a compile_commands.json with per-file extra arguments.
+
+    Args:
+        workspace: Root directory of the workspace.
+        entries: List of (file_name, extra_args) pairs; a file may appear
+            multiple times to model multi-configuration projects.
+    """
+    data = [
+        {
+            "directory": str(workspace),
+            "file": str(workspace / f),
+            "arguments": [
+                "clang++",
+                "-std=c++17",
+                "-fsyntax-only",
+                *args,
+                str(workspace / f),
+            ],
+        }
+        for f, args in entries
+    ]
+    (workspace / "compile_commands.json").write_text(json.dumps(data))
diff --git a/tests/integration/utils/injection.py b/tests/tools/injection.py
similarity index 100%
rename from tests/integration/utils/injection.py
rename to tests/tools/injection.py
diff --git a/tests/tools/lifecycle.py b/tests/tools/lifecycle.py
new file mode 100644
index 000000000..0242d0f88
--- /dev/null
+++ b/tests/tools/lifecycle.py
@@ -0,0 +1,145 @@
+"""Server lifecycle helpers: spawn, graceful shutdown, clean-exit gates."""
+
+import asyncio
+import os
+import socket
+from pathlib import Path
+
+import pytest
+
+from tests.tools.checks import assert_no_anomaly
+from tests.tools.client import CliceClient
+
+SANITIZER_MARKERS = (
+    "AddressSanitizer",
+    "LeakSanitizer",
+    "MemorySanitizer",
+    "ThreadSanitizer",
+    "UndefinedBehaviorSanitizer",
+    "==ERROR:",
+    "runtime error:",
+)
+
+next_port_offset = 0
+
+
+def find_free_port() -> int:
+    """Pick a port from a per-xdist-worker range.
+
+    bind(0) draws from the kernel's shared pool: two concurrent xdist
+    workers can grab the same port in the close-then-rebind gap. Disjoint
+    per-worker ranges (below the ephemeral range) remove that race; the
+    advancing offset avoids immediately reusing a just-released port.
+    """
+    global next_port_offset
+    worker = os.environ.get("PYTEST_XDIST_WORKER", "gw0")
+    suffix = worker.removeprefix("gw")
+    index = int(suffix) if suffix.isdigit() else 0
+    base = 21000 + index * 100
+    for _ in range(100):
+        port = base + next_port_offset
+        next_port_offset = (next_port_offset + 1) % 100
+        with socket.socket(socket.AF_INET, socket.SOCK_STREAM) as s:
+            try:
+                s.bind(("127.0.0.1", port))
+                return port
+            except OSError:
+                continue
+    raise RuntimeError(f"no free port in range {base}-{base + 99}")
+
+
+async def make_client(executable: Path, workspace: Path) -> CliceClient:
+    """Spawn a fresh clice server and initialize it. For multi-session tests."""
+    c = CliceClient()
+    await c.start_io(str(executable), "serve")
+    await c.initialize(workspace)
+    return c
+
+
+def check_no_anomaly(request: pytest.FixtureRequest, c: CliceClient) -> None:
+    """Teardown gate: anomalies are internal clice bugs — every test session
+    must end without one. Tests that intentionally trigger anomalies opt out
+    with @pytest.mark.allow_anomaly and assert on them explicitly."""
+    if request.node.get_closest_marker("allow_anomaly") is not None:
+        return
+    assert_no_anomaly(c, c.workspace)
+
+
+def server_stderr_excerpt(stderr_text: str) -> str:
+    interesting = [
+        line
+        for line in stderr_text.splitlines()
+        if "[warn]" in line
+        or "[error]" in line
+        or "Sanitizer" in line
+        or "==ERROR:" in line
+        or "runtime error:" in line
+    ]
+    return "\n".join(interesting[-80:])
+
+
+async def assert_server_exited_cleanly(server, timeout: float = 10.0) -> None:
+    failures: list[str] = []
+
+    if server is None:
+        return
+
+    if server.returncode is None:
+        try:
+            await asyncio.wait_for(server.wait(), timeout=timeout)
+        except asyncio.TimeoutError:
+            server.kill()
+            await server.wait()
+            failures.append(f"server did not exit within {timeout:g}s after shutdown")
+
+    print(f"[server] exit code: {server.returncode}", flush=True)
+
+    stderr_text = ""
+    if server.stderr:
+        try:
+            stderr_data = await asyncio.wait_for(server.stderr.read(), timeout=2.0)
+            stderr_text = stderr_data.decode("utf-8", errors="replace")
+        except Exception as exc:
+            failures.append(f"failed to collect server stderr: {exc!r}")
+
+    for line in server_stderr_excerpt(stderr_text).splitlines():
+        print(f"[server] {line}", flush=True)
+
+    if server.returncode != 0:
+        failures.append(f"server exited with code {server.returncode}")
+
+    if any(marker in stderr_text for marker in SANITIZER_MARKERS):
+        failures.append("server stderr contains sanitizer/runtime error output")
+
+    if failures:
+        excerpt = server_stderr_excerpt(stderr_text)
+        if excerpt:
+            failures.append("server stderr excerpt:\n" + excerpt)
+        pytest.fail("\n".join(failures))
+
+
+async def shutdown_client(c: CliceClient, *, verbose: bool = False) -> None:
+    """Gracefully shut down a client, force-kill if needed."""
+    try:
+        await asyncio.wait_for(c.shutdown_async(None), timeout=10.0)
+    except Exception:
+        pass
+
+    try:
+        c.exit(None)
+    except Exception:
+        pass
+
+    if verbose and c.log_messages:
+        for msg in c.log_messages:
+            level = {1: "ERROR", 2: "WARN", 3: "INFO", 4: "LOG"}.get(msg.type, "?")
+            print(f"[logMessage/{level}] {msg.message}", flush=True)
+
+    try:
+        await assert_server_exited_cleanly(c.server)
+    finally:
+        try:
+            await c.stop_io()
+            await asyncio.sleep(0.1)
+        except Exception:
+            pass
diff --git a/tests/prepare.py b/tests/tools/prepare.py
similarity index 93%
rename from tests/prepare.py
rename to tests/tools/prepare.py
index 299b0ebc2..852455070 100644
--- a/tests/prepare.py
+++ b/tests/tools/prepare.py
@@ -1,6 +1,6 @@
 """Prepare test data fixtures for editor E2E tests.
 
-Usage: python tests/prepare.py <fixture> [<fixture> ...]
+Usage: python tests/tools/prepare.py <fixture> [<fixture> ...]
 
 Each fixture is a subdirectory of tests/data. Fixtures with a
 CMakeLists.txt get compile_commands.json generated via CMake; plain
@@ -13,10 +13,10 @@
 import sys
 from pathlib import Path
 
-REPO_ROOT = Path(__file__).resolve().parents[1]
+REPO_ROOT = Path(__file__).resolve().parents[2]
 sys.path.insert(0, str(REPO_ROOT))
 
-from tests.cdb import generate_cdb, generate_test_data_cdbs  # noqa: E402
+from tests.tools.compile_commands import generate_cdb, generate_test_data_cdbs  # noqa: E402
 
 
 def xdg_cache_dir(workspace: Path) -> Path | None:
diff --git a/tests/replay.py b/tests/tools/replay.py
similarity index 99%
rename from tests/replay.py
rename to tests/tools/replay.py
index ee484b920..6e0032aa4 100644
--- a/tests/replay.py
+++ b/tests/tools/replay.py
@@ -19,7 +19,7 @@
 from pathlib import Path
 from urllib.parse import quote, unquote
 
-REPO_ROOT = Path(__file__).resolve().parent.parent
+REPO_ROOT = Path(__file__).resolve().parents[2]
 
 SERVER_REQUEST_DEFAULTS: dict[str, object] = {
     "window/workDoneProgress/create": None,
diff --git a/tests/stress.py b/tests/tools/stress.py
similarity index 95%
rename from tests/stress.py
rename to tests/tools/stress.py
index 4730f6f71..9c5c18700 100644
--- a/tests/stress.py
+++ b/tests/tools/stress.py
@@ -3,16 +3,16 @@
 
 Usage:
     # Full indexing test on LLVM (wait for completion):
-    pixi run python tests/stress.py /home/ykiko/workspace/llvm-project \
+    pixi run python tests/tools/stress.py /home/ykiko/workspace/llvm-project \
         --executable build/RelWithDebInfo/bin/clice
 
     # Time-limited run (just see how far it gets in 5 minutes):
-    pixi run python tests/stress.py /home/ykiko/workspace/llvm-project \
+    pixi run python tests/tools/stress.py /home/ykiko/workspace/llvm-project \
         --executable build/RelWithDebInfo/bin/clice \
         --timeout 300
 
     # Custom worker limits:
-    pixi run python tests/stress.py /home/ykiko/workspace/llvm-project \
+    pixi run python tests/tools/stress.py /home/ykiko/workspace/llvm-project \
         --executable build/RelWithDebInfo/bin/clice \
         --max-stateless 16
 """
@@ -27,8 +27,8 @@
 import time
 from pathlib import Path
 
-sys.path.insert(0, str(Path(__file__).parent.parent))
-from tests.integration.utils.client import CliceClient
+sys.path.insert(0, str(Path(__file__).resolve().parents[2]))
+from tests.tools.client import CliceClient
 
 
 def parse_args():
@@ -266,6 +266,11 @@ async def run_stress_test(args):
             "cache_dir": str(cache_dir),
             "logging_dir": str(log_dir),
             "enable_indexing": True,
+            # 0 = server-side auto. Explicit so CliceClient.initialize's
+            # 1-worker test default doesn't apply — this tool exists to
+            # stress real worker pools.
+            "stateless_worker_count": 0,
+            "stateful_worker_count": 0,
         }
     }
     if args.max_stateless > 0:
diff --git a/tests/integration/utils/cache.py b/tests/tools/workspace.py
similarity index 57%
rename from tests/integration/utils/cache.py
rename to tests/tools/workspace.py
index fd0e641f2..9fb78a640 100644
--- a/tests/integration/utils/cache.py
+++ b/tests/tools/workspace.py
@@ -1,8 +1,46 @@
-"""Cache inspection helpers for persistent cache tests."""
+"""On-disk workspace helpers: sources, document edits, cache inspection."""
 
 import json
 from pathlib import Path
 
+from lsprotocol.types import (
+    DidChangeTextDocumentParams,
+    TextDocumentContentChangeWholeDocument,
+    TextDocumentIdentifier,
+    VersionedTextDocumentIdentifier,
+)
+
+
+def doc(uri: str) -> TextDocumentIdentifier:
+    """Create a TextDocumentIdentifier from a URI string."""
+    return TextDocumentIdentifier(uri=uri)
+
+
+def write_source(workspace: Path, name: str, content: str) -> Path:
+    """Write a source file to the workspace directory. Returns the file path."""
+    path = workspace / name
+    path.parent.mkdir(parents=True, exist_ok=True)
+    path.write_text(content)
+    return path
+
+
+def did_change(client, uri: str, version: int, text: str) -> None:
+    """Send a didChange notification with whole-document replacement."""
+    client.text_document_did_change(
+        DidChangeTextDocumentParams(
+            text_document=VersionedTextDocumentIdentifier(uri=uri, version=version),
+            content_changes=[TextDocumentContentChangeWholeDocument(text=text)],
+        )
+    )
+
+
+def get_field(obj, key, default=None):
+    """Read a field from a dict or attribute-style LSP response object."""
+    if isinstance(obj, dict):
+        return obj.get(key, default)
+    return getattr(obj, key, default)
+
+
 # Versioned root of the unified cache store; bump together with
 # cache_format_version in src/server/state/workspace.h.
 CACHE_ROOT = Path(".clice") / "cache" / "v3"