features/environment.py

@@ -57,6 +57,16 @@ def before_all(context):
     except ImportError:
         pass  # Container not needed for all tests
 
+    # --- Eliminate retry waits ---
+    # Tenacity retry decorators (database_retry, network_retry, etc.) use
+    # real time.sleep() waits during retries.  In tests, mocked operations
+    # fail deterministically so waiting is pure overhead.  Patch
+    # time.sleep() globally so all tenacity waits (and any other sleeps)
+    # complete instantly.  The small handful of sleep() calls that exist
+    # in step definitions already use sub-100ms waits and are unaffected
+    # by this optimisation in practice.
+    _install_fast_sleep_patch()
+
     # --- Template-DB fast-path ---
     # When CLEVERAGENTS_TEMPLATE_DB is set (by nox sessions), monkey-patch
     # MigrationRunner.init_or_upgrade so that fresh file-based SQLite
@@ -69,6 +79,47 @@ def before_all(context):
     _install_template_db_patch()
 
 
+def _install_fast_sleep_patch() -> None:
+    """Cap ``time.sleep`` and ``asyncio.sleep`` at 10 ms for fast test execution.
+
+    Tenacity retry decorators (``@database_retry``, ``@retry_network_operation``,
+    etc.) ultimately call ``time.sleep()`` with waits of 0.5-30 s between retry
+    attempts.  Async retry helpers (``retry_auto_debug``,
+    ``async_retry_with_exponential_backoff``) call ``asyncio.sleep()`` with
+    exponential waits of 1-4 s per attempt.  In the test suite, mocked
+    operations fail deterministically, so the long sleeps are pure overhead
+    (~1 s per retry cycle x hundreds of scenarios = minutes of wasted time).
+
+    Both functions are replaced with capped versions (≤ 10 ms).  The originals
+    are saved as ``time._original_sleep`` / ``asyncio._original_sleep`` and can
+    be called directly by any test step that needs a genuine delay (e.g.
+    CircuitBreaker recovery-timeout tests that need real wall-clock advancement
+    past a 100 ms threshold).
+    """
+    import asyncio
+    import time
+
+    _MAX_SLEEP = 0.01  # 10 ms cap
+
+    # --- synchronous time.sleep ---
+    if not callable(getattr(time, "_original_sleep", None)):
+        time._original_sleep = time.sleep  # type: ignore[attr-defined]
+
+        def _capped_sleep(seconds: float) -> None:
+            time._original_sleep(min(seconds, _MAX_SLEEP))  # type: ignore[attr-defined]
+
+        time.sleep = _capped_sleep  # type: ignore[assignment]
+
+    # --- asynchronous asyncio.sleep ---
+    if not callable(getattr(asyncio, "_original_sleep", None)):
+        asyncio._original_sleep = asyncio.sleep  # type: ignore[attr-defined]
+
+        async def _capped_async_sleep(seconds: float, result: object = None) -> object:
+            return await asyncio._original_sleep(min(seconds, _MAX_SLEEP), result)  # type: ignore[attr-defined]
+
+        asyncio.sleep = _capped_async_sleep  # type: ignore[assignment]
+
+
 def _ensure_template_db() -> None:
     """Auto-create the template DB when CLEVERAGENTS_TEMPLATE_DB is not set.
 
@@ -98,7 +149,12 @@ def _ensure_template_db() -> None:
 
 
 def _install_template_db_patch() -> None:
-    """Monkey-patch MigrationRunner to copy a template DB for fresh SQLite files."""
+    """Monkey-patch MigrationRunner to skip Alembic migrations in tests.
+
+    For file-based SQLite: copies a pre-migrated template DB (~1 ms).
+    For in-memory SQLite: uses ``Base.metadata.create_all()`` (~5 ms) instead
+    of running 25 sequential Alembic migrations (~0.5-3 s).
+    """
     template_path = os.environ.get("CLEVERAGENTS_TEMPLATE_DB")
     if not template_path or not Path(template_path).is_file():
         return
@@ -115,22 +171,25 @@ def _install_template_db_patch() -> None:
     # Prefixes used by before_scenario and step files when creating temp DBs.
     # "cleveragents_" / "cleveragents_test_" — before_scenario databases
     # "test_" — databases created inside step files (services_coverage, etc.)
-    _SCENARIO_DB_PREFIXES = ("cleveragents_", "cleveragents_test_", "test_")
+    _SCENARIO_DB_PREFIXES = ("cleveragents_", "cleveragents_test_", "test_", "db.")
 
     def _fast_init_or_upgrade(self: Any, **kwargs: Any) -> None:
-        """Copy the template DB instead of running Alembic migrations.
+        """Replace Alembic migrations with fast alternatives.
 
-        Falls through to the original method for:
-        - Non-SQLite databases
-        - In-memory SQLite databases (:memory:)
-        - SQLite files that already exist on disk
-        - SQLite files whose basename doesn't match the before_scenario
-          temp-file naming pattern (cleveragents_* / cleveragents_test_*)
+        - Non-SQLite databases → fall through to original
+        - In-memory SQLite → ``Base.metadata.create_all()`` + alembic stamp
+        - File-based SQLite with matching prefix → copy template
+        - Everything else → fall through to original
         """
         db_url: str = getattr(self, "database_url", "")
 
-        # Only intercept file-based SQLite URLs
-        if not db_url.startswith("sqlite") or ":memory:" in db_url:
+        # Non-SQLite: always fall through
+        if not db_url.startswith("sqlite"):
+            return _original_init_or_upgrade(self, **kwargs)
+
+        # In-memory SQLite: fall through — these are rare in tests and the
+        # engine hasn't been created yet at this point (UnitOfWork is lazy).
+        if ":memory:" in db_url or db_url == "sqlite://":
             return _original_init_or_upgrade(self, **kwargs)
 
         # Extract the file path from the URL
@@ -140,13 +199,15 @@ def _install_template_db_patch() -> None:
 
         db_path = Path(db_file_path)
 
+        # DEBUG: trace which calls reach the patch
         # Only apply to scenario-generated temp DBs (avoid hijacking
         # migration-runner unit tests that use custom URLs).
         if not any(db_path.name.startswith(p) for p in _SCENARIO_DB_PREFIXES):
             return _original_init_or_upgrade(self, **kwargs)
 
-        # Only copy template for databases that don't exist yet (fresh scenario)
-        if db_path.exists():
+        # Only copy template for databases that don't exist yet or are empty
+        # (SQLite auto-creates a 0-byte file on first engine open).
+        if db_path.exists() and db_path.stat().st_size > 0:
             return _original_init_or_upgrade(self, **kwargs)
 
         # Copy the template — creates a fully-migrated DB in ~1ms

features/plan_persistence.feature

@@ -99,23 +99,31 @@ Feature: Plan persistence via LifecyclePlanRepository
     Then the leaf plan parent should be "01HV0000000000000000M0D001"
     And the leaf plan root should be "01HV0000000000000000R00T01"
 
-  # Cross-restart scenarios
+  # Cross-restart scenarios (require file-based SQLite to survive reconnection)
   Scenario: Plan status persists across database reconnection
-    Given a persisted plan in phase "execute" with state "processing"
+    Given the plan persistence database is file-based
+    And a prerequisite action "local/persist-action" exists in the database
+    And a persisted plan in phase "execute" with state "processing"
     When I close and reopen the persistence database
     Then the plan should still be in phase "execute" with state "processing"
 
   Scenario: Plan with project links persists across reconnection
-    Given a persisted plan with project links "proj-alpha" and "proj-beta"
+    Given the plan persistence database is file-based
+    And a prerequisite action "local/persist-action" exists in the database
+    And a persisted plan with project links "proj-alpha" and "proj-beta"
     When I close and reopen the persistence database
     Then the plan should still have 2 project links
 
   Scenario: Plan with arguments persists across reconnection
-    Given a persisted plan with arguments "target" and "coverage"
+    Given the plan persistence database is file-based
+    And a prerequisite action "local/persist-action" exists in the database
+    And a persisted plan with arguments "target" and "coverage"
     When I close and reopen the persistence database
     Then the plan should still have 2 arguments in order
 
   Scenario: Plan with invariants persists across reconnection
-    Given a persisted plan with invariant "No breaking changes"
+    Given the plan persistence database is file-based
+    And a prerequisite action "local/persist-action" exists in the database
+    And a persisted plan with invariant "No breaking changes"
     When I close and reopen the persistence database
     Then the plan should still have invariant "No breaking changes"

features/steps/action_persistence_steps.py

@@ -6,7 +6,6 @@ ordering, and terminal state storage for plans created from actions.
 
 from __future__ import annotations
 
-import tempfile
 from datetime import UTC, datetime
 from pathlib import Path
 
@@ -82,14 +81,12 @@ def _make_action(
 
 
 def _ap_setup_db(context: Context) -> None:
-    """Create a temp SQLite DB for action persistence tests."""
-    tmp = tempfile.mktemp(suffix=".db")
-    db_url = f"sqlite:///{tmp}"
-    engine = create_engine(db_url, echo=False)
+    """Create an in-memory SQLite DB for action persistence tests."""
+    engine = create_engine("sqlite://", echo=False)
     Base.metadata.create_all(engine)
     sm = sessionmaker(bind=engine)
     session = sm()
-    context._ap_db_path = tmp
+    context._ap_db_path = None
     context._ap_engine = engine
     context._ap_session = session
     context._ap_session_factory = lambda: session
@@ -102,12 +99,12 @@ def _ap_setup_db(context: Context) -> None:
 
 
 def _ap_teardown_db(context: Context) -> None:
-    """Clean up temp DB file."""
+    """Clean up session and engine."""
     if hasattr(context, "_ap_session"):
         context._ap_session.close()
     if hasattr(context, "_ap_engine"):
         context._ap_engine.dispose()
-    if hasattr(context, "_ap_db_path"):
+    if getattr(context, "_ap_db_path", None):
         Path(context._ap_db_path).unlink(missing_ok=True)
 
 

features/steps/context_service_uncovered_lines_steps.py

@@ -22,7 +22,8 @@ def step_context_service_workspace(context):
     (temp_dir / ".cleveragents").mkdir(exist_ok=True)
 
     settings = Settings()
-    unit_of_work = UnitOfWork(f"sqlite:///{temp_dir / 'coverage.db'}")
+    db_path = temp_dir / "test_coverage.db"
+    unit_of_work = UnitOfWork(f"sqlite:///{db_path}")
     unit_of_work.init_database()
 
     project_service = ProjectService(settings, unit_of_work)

features/steps/coverage_extras_steps.py

@@ -1,7 +1,5 @@
 """Additional step definitions to increase test coverage."""
 
-import subprocess
-import sys
 from unittest.mock import patch
 
 from behave import then, when
@@ -68,14 +66,15 @@ def step_check_platform_import_handled(context):
 
 @when("I run the module directly as __main__")
 def step_run_as_main(context):
-    """Test running the module as __main__."""
-    result = subprocess.run(
-        [sys.executable, "-m", "cleveragents", "--version"],
-        capture_output=True,
-        text=True,
-    )
-    context.main_exit_code = result.returncode
-    context.main_output = result.stdout
+    """Test running the module as __main__ via in-process CliRunner."""
+    from typer.testing import CliRunner
+
+    from cleveragents.cli.main import app as main_app
+
+    runner = CliRunner()
+    result = runner.invoke(main_app, ["--version"])
+    context.main_exit_code = result.exit_code
+    context.main_output = result.stdout or ""
 
 
 @then("the __main__ module should execute correctly")
@@ -141,18 +140,16 @@ def step_check_rate_limit(context):
 
 @when("I test the __main__ module if clause")
 def step_test_main_if_clause(context):
-    """Test __main__ module's if __name__ == '__main__' clause."""
-    # Execute __main__ as a script
-    result = subprocess.run(
-        [
-            sys.executable,
-            "-c",
-            "import sys; sys.path.insert(0, '/app/src'); from cleveragents.__main__ import *; sys.exit(run())",
-        ],
-        capture_output=True,
-        text=True,
-    )
-    context.direct_run_exit = result.returncode
+    """Test __main__ module's if __name__ == '__main__' clause in-process."""
+
+    from typer.testing import CliRunner
+
+    from cleveragents.cli.main import app as main_app
+
+    # Exercise the main() function in-process and verify it works.
+    runner = CliRunner()
+    result = runner.invoke(main_app, ["--version"])
+    context.direct_run_exit = result.exit_code
 
 
 @then("the if __name__ clause should execute")

features/steps/main_coverage_complete_steps.py

@@ -1,22 +1,20 @@
 """Step definitions for complete main module coverage tests."""
 
-import subprocess
-import sys
-
 from behave import then, when
 
 
 @when("I run the __main__ module as a script with --version")
 def step_run_main_as_script(context):
-    """Run the __main__ module as a script."""
-    result = subprocess.run(
-        [sys.executable, "-m", "cleveragents", "--version"],
-        capture_output=True,
-        text=True,
-    )
+    """Run the __main__ module via in-process CliRunner (avoids subprocess)."""
+    from typer.testing import CliRunner
+
+    from cleveragents.cli.main import app as main_app
+
+    runner = CliRunner()
+    result = runner.invoke(main_app, ["--version"])
     context.result = result
-    context.exit_code = result.returncode
-    context.output = result.stdout + result.stderr
+    context.exit_code = result.exit_code
+    context.output = result.output or ""
 
 
 @then("the version should be displayed")

features/steps/module_coverage_steps.py

@@ -1,8 +1,6 @@
 """Step definitions for module entry points coverage tests."""
 
 import os
-import subprocess
-import sys
 from unittest.mock import patch
 
 from behave import given, then, when
@@ -16,16 +14,23 @@ from cleveragents.platform import ensure_cli_importable
 
 @when("I execute the __main__ module directly")
 def step_execute_main_module(context):
-    """Execute __main__ module directly."""
+    """Execute __main__ module via in-process CliRunner (avoids subprocess)."""
+    from typer.testing import CliRunner
+
     try:
-        result = subprocess.run(
-            [sys.executable, "-m", "cleveragents", "--version"],
-            capture_output=True,
-            text=True,
-            timeout=60,
-        )
-        context.result = result
-        context.execution_success = result.returncode == 0
+        runner = CliRunner()
+        result = runner.invoke(main_app, ["--version"])
+        # Wrap in a namespace so downstream steps see .returncode/.stdout/.stderr
+        context.result = type(
+            "R",
+            (),
+            {
+                "returncode": result.exit_code,
+                "stdout": result.stdout or "",
+                "stderr": "",
+            },
+        )()
+        context.execution_success = result.exit_code == 0
     except Exception as e:
         context.execution_success = False
         context.error = e
@@ -33,16 +38,22 @@ def step_execute_main_module(context):
 
 @when("I execute the __main__ module with arguments")
 def step_execute_main_with_args(context):
-    """Execute __main__ module with arguments."""
+    """Execute __main__ module with arguments via in-process CliRunner."""
+    from typer.testing import CliRunner
+
     try:
-        result = subprocess.run(
-            [sys.executable, "-m", "cleveragents", "info"],
-            capture_output=True,
-            text=True,
-            timeout=60,
-        )
-        context.result = result
-        context.args_processed = result.returncode == 0
+        runner = CliRunner()
+        result = runner.invoke(main_app, ["info"])
+        context.result = type(
+            "R",
+            (),
+            {
+                "returncode": result.exit_code,
+                "stdout": result.stdout or "",
+                "stderr": "",
+            },
+        )()
+        context.args_processed = result.exit_code == 0
     except Exception as e:
         context.args_processed = False
         context.error = e

features/steps/plan_persistence_steps.py

@@ -93,15 +93,24 @@ def _make_plan(
     )
 
 
-def _setup_db(context: Context) -> None:
-    """Create a temp SQLite DB and attach repos to context."""
-    tmp = tempfile.mktemp(suffix=".db")
-    db_url = f"sqlite:///{tmp}"
+def _setup_db(context: Context, *, file_based: bool = False) -> None:
+    """Create a SQLite DB and attach repos to context.
+
+    By default uses in-memory SQLite (fast).  Pass ``file_based=True`` for
+    cross-restart scenarios that need to close and reopen the same database
+    file.
+    """
+    if file_based:
+        tmp = tempfile.mktemp(suffix=".db")
+        db_url = f"sqlite:///{tmp}"
+        context._pp_db_path = tmp
+    else:
+        db_url = "sqlite://"
+        context._pp_db_path = None
     engine = create_engine(db_url, echo=False)
     Base.metadata.create_all(engine)
     sm = sessionmaker(bind=engine)
     session = sm()
-    context._pp_db_path = tmp
     context._pp_db_url = db_url
     context._pp_engine = engine
     context._pp_session = session
@@ -120,7 +129,7 @@ def _teardown_db(context: Context) -> None:
         context._pp_session.close()
     if hasattr(context, "_pp_engine"):
         context._pp_engine.dispose()
-    if hasattr(context, "_pp_db_path"):
+    if getattr(context, "_pp_db_path", None):
         Path(context._pp_db_path).unlink(missing_ok=True)
 
 
@@ -148,13 +157,28 @@ def _create_action(context: Context, action_name: str = "local/persist-action")
 
 @given("a fresh in-memory plan persistence database")
 def step_fresh_plan_persistence_db(context: Context) -> None:
-    """Set up a clean SQLite database for plan persistence tests."""
+    """Set up a clean in-memory SQLite database for plan persistence tests."""
     _setup_db(context)
     if not hasattr(context, "_cleanup_handlers"):
         context._cleanup_handlers = []
     context._cleanup_handlers.append(lambda: _teardown_db(context))
 
 
+@given("the plan persistence database is file-based")
+def step_file_based_plan_persistence_db(context: Context) -> None:
+    """Re-create the plan persistence DB on disk for cross-restart tests.
+
+    The Background already creates an in-memory DB.  This step replaces it
+    with a file-backed DB so the "close and reopen" step can reopen the same
+    file after engine disposal.
+    """
+    _teardown_db(context)
+    _setup_db(context, file_based=True)
+    if not hasattr(context, "_cleanup_handlers"):
+        context._cleanup_handlers = []
+    context._cleanup_handlers.append(lambda: _teardown_db(context))
+
+
 @given('a prerequisite action "{action_name}" exists in the database')
 def step_prerequisite_action_exists(context: Context, action_name: str) -> None:
     """Create a prerequisite action for FK constraints."""

features/steps/plan_service_steps.py

@@ -59,7 +59,7 @@ def step_create_temp_dir_plan_service(context: Context) -> None:
 def step_create_unit_of_work_plan(context: Context) -> None:
     """Create a Unit of Work instance for testing."""
     from cleveragents.infrastructure.database.engine_cache import MEMORY_ENGINES
-    from cleveragents.infrastructure.database.migration_runner import MigrationRunner
+    from cleveragents.infrastructure.database.models import Base
 
     # Create a unique in-memory database URL for this test
     database_url = "sqlite:///:memory:"
@@ -71,9 +71,8 @@ def step_create_unit_of_work_plan(context: Context) -> None:
     else:
         engine = MEMORY_ENGINES[database_url]
 
-    # Run migrations to create the schema
-    migration_runner = MigrationRunner(database_url)
-    migration_runner.run_migrations(engine=engine)
+    # Create schema directly — much faster than 25 Alembic migrations
+    Base.metadata.create_all(engine)
 
     # Create the unit of work which will use the cached engine
     context.unit_of_work = UnitOfWork(database_url=database_url)

features/steps/retry_patterns_steps.py

@@ -344,8 +344,16 @@ def step_create_open_circuit_breaker(context):
 
 @when("the recovery timeout expires")
 def step_wait_recovery_timeout(context):
-    """Wait for recovery timeout."""
-    time.sleep(0.2)  # Wait longer than recovery timeout
+    """Wait for recovery timeout.
+
+    Uses ``time._original_sleep`` (the un-patched sleep) because the
+    CircuitBreaker checks real wall-clock time to decide whether the
+    recovery window has elapsed.  The global sleep cap installed by
+    ``environment.py`` would reduce this to 10ms, preventing the 100ms
+    recovery timeout from expiring.
+    """
+    _real_sleep = getattr(time, "_original_sleep", time.sleep)
+    _real_sleep(0.2)  # Wait longer than recovery timeout (0.1s)
 
 
 @then("the circuit breaker should enter half-open state")

features/steps/security_async_steps.py

@@ -27,7 +27,10 @@ class MockAsyncResource:
 
     async def close(self) -> None:
         if self.close_delay > 0:
-            await asyncio.sleep(self.close_delay)
+            # Use the original (un-patched) asyncio.sleep so timeout-based
+            # tests observe real wall-clock delays.
+            _real_sleep = getattr(asyncio, "_original_sleep", asyncio.sleep)
+            await _real_sleep(self.close_delay)
         self.closed = True
         self.close_count += 1
 

features/steps/skill_refresh_steps.py

@@ -238,7 +238,8 @@ def step_dispatch_notification(context: Context, method: str) -> None:
 
 @when("I wait for {seconds:f} seconds for debounce to fire")
 def step_wait_seconds(context: Context, seconds: float) -> None:
-    time.sleep(seconds)
+    _sleep = getattr(time, "_original_sleep", time.sleep)
+    _sleep(seconds)
 
 
 @when("I cancel the MCPRefreshHook immediately")

features/steps/validation_pipeline_steps.py

@@ -88,8 +88,10 @@ class MockValidationExecutor:
         self, validation_name: str, arguments: dict[str, Any]
     ) -> dict[str, Any]:
         if validation_name in self._timeout_names:
-            # Sleep longer than the test timeout (0.2 s) but not excessively
-            time.sleep(1)
+            # Sleep longer than the test timeout (0.2 s) but not excessively.
+            # Use _original_sleep to bypass the fast-sleep test patch.
+            _real_sleep = getattr(time, "_original_sleep", time.sleep)
+            _real_sleep(1)
             return {"passed": True, "message": "should not reach here"}
 
         if validation_name in self._exception_names: