Merge pull request #201 from SentienceAPI/debugger

Sentience Debugger for any agent

Author: rcholic

sentience/__init__.py

@@ -63,6 +63,7 @@
 from .cloud_tracing import CloudTraceSink, SentienceLogger
 from .conversational_agent import ConversationalAgent
 from .cursor_policy import CursorPolicy
+from .debugger import SentienceDebugger
 from .expect import expect
 from .generator import ScriptGenerator, generate
 from .inspector import Inspector, inspect
@@ -286,6 +287,7 @@
     "AgentAction",
     # Verification (agent assertion loop)
     "AgentRuntime",
+    "SentienceDebugger",
     "AssertContext",
     "AssertOutcome",
     "Predicate",

sentience/agent_runtime.py

@@ -185,6 +185,59 @@ def __init__(
         self._captcha_options: CaptchaOptions | None = None
         self._captcha_retry_count: int = 0
 
+    @classmethod
+    def from_playwright_page(
+        cls,
+        page: Page,
+        tracer: Tracer,
+        snapshot_options: SnapshotOptions | None = None,
+        sentience_api_key: str | None = None,
+        tool_registry: ToolRegistry | None = None,
+    ) -> AgentRuntime:
+        """
+        Create AgentRuntime from a raw Playwright Page (sidecar mode).
+
+        Args:
+            page: Playwright Page for browser interaction
+            tracer: Tracer for emitting verification events
+            snapshot_options: Default options for snapshots
+            sentience_api_key: API key for Pro/Enterprise tier
+            tool_registry: Optional ToolRegistry for LLM-callable tools
+
+        Returns:
+            AgentRuntime instance
+        """
+        from .backends.playwright_backend import PlaywrightBackend
+
+        backend = PlaywrightBackend(page)
+        return cls(
+            backend=backend,
+            tracer=tracer,
+            snapshot_options=snapshot_options,
+            sentience_api_key=sentience_api_key,
+            tool_registry=tool_registry,
+        )
+
+    @classmethod
+    def attach(
+        cls,
+        page: Page,
+        tracer: Tracer,
+        snapshot_options: SnapshotOptions | None = None,
+        sentience_api_key: str | None = None,
+        tool_registry: ToolRegistry | None = None,
+    ) -> AgentRuntime:
+        """
+        Sidecar alias for from_playwright_page().
+        """
+        return cls.from_playwright_page(
+            page=page,
+            tracer=tracer,
+            snapshot_options=snapshot_options,
+            sentience_api_key=sentience_api_key,
+            tool_registry=tool_registry,
+        )
+
     @classmethod
     async def from_sentience_browser(
         cls,

sentience/debugger.py

@@ -0,0 +1,69 @@
+from __future__ import annotations
+
+from collections.abc import AsyncIterator
+from contextlib import asynccontextmanager
+from typing import TYPE_CHECKING, Any
+
+from .agent_runtime import AgentRuntime
+from .models import SnapshotOptions
+from .tools import ToolRegistry
+
+if TYPE_CHECKING:  # pragma: no cover - type hints only
+    from playwright.async_api import Page
+
+    from .tracing import Tracer
+else:  # pragma: no cover - avoid optional runtime imports
+    Page = Any  # type: ignore
+    Tracer = Any  # type: ignore
+
+
+class SentienceDebugger:
+    """
+    Verifier-only sidecar wrapper around AgentRuntime.
+    """
+
+    def __init__(self, runtime: AgentRuntime) -> None:
+        self.runtime = runtime
+        self._step_open = False
+
+    @classmethod
+    def attach(
+        cls,
+        page: Page,
+        tracer: Tracer,
+        snapshot_options: SnapshotOptions | None = None,
+        sentience_api_key: str | None = None,
+        tool_registry: ToolRegistry | None = None,
+    ) -> SentienceDebugger:
+        runtime = AgentRuntime.from_playwright_page(
+            page=page,
+            tracer=tracer,
+            snapshot_options=snapshot_options,
+            sentience_api_key=sentience_api_key,
+            tool_registry=tool_registry,
+        )
+        return cls(runtime=runtime)
+
+    def begin_step(self, goal: str, step_index: int | None = None) -> str:
+        self._step_open = True
+        return self.runtime.begin_step(goal, step_index=step_index)
+
+    async def end_step(self, **kwargs: Any) -> dict[str, Any]:
+        self._step_open = False
+        return await self.runtime.emit_step_end(**kwargs)
+
+    @asynccontextmanager
+    async def step(self, goal: str, step_index: int | None = None) -> AsyncIterator[None]:
+        self.begin_step(goal, step_index=step_index)
+        try:
+            yield
+        finally:
+            await self.end_step()
+
+    async def snapshot(self, **kwargs: Any):
+        return await self.runtime.snapshot(**kwargs)
+
+    def check(self, predicate, label: str, required: bool = False):
+        if not self._step_open:
+            self.begin_step(f"verify:{label}")
+        return self.runtime.check(predicate, label, required=required)

tests/test_agent_runtime.py

@@ -695,6 +695,41 @@ async def test_from_sentience_browser_with_api_key(self) -> None:
             assert runtime._snapshot_options.use_api is True
 
 
+class TestAgentRuntimeFromPlaywrightPage:
+    """Tests for from_playwright_page factory method."""
+
+    def test_from_playwright_page_creates_runtime(self) -> None:
+        """Test from_playwright_page creates runtime with PlaywrightBackend."""
+        mock_page = MagicMock()
+        tracer = MockTracer()
+
+        with patch("sentience.backends.playwright_backend.PlaywrightBackend") as MockPWBackend:
+            mock_backend_instance = MagicMock()
+            MockPWBackend.return_value = mock_backend_instance
+
+            runtime = AgentRuntime.from_playwright_page(page=mock_page, tracer=tracer)
+
+            assert runtime.backend is mock_backend_instance
+            assert not hasattr(runtime, "_legacy_browser")
+            assert not hasattr(runtime, "_legacy_page")
+            MockPWBackend.assert_called_once_with(mock_page)
+
+    def test_from_playwright_page_with_api_key(self) -> None:
+        """Test from_playwright_page passes API key."""
+        mock_page = MagicMock()
+        tracer = MockTracer()
+
+        with patch("sentience.backends.playwright_backend.PlaywrightBackend"):
+            runtime = AgentRuntime.from_playwright_page(
+                page=mock_page,
+                tracer=tracer,
+                sentience_api_key="sk_test",
+            )
+
+            assert runtime._snapshot_options.sentience_api_key == "sk_test"
+            assert runtime._snapshot_options.use_api is True
+
+
 class TestAgentRuntimeSnapshot:
     """Tests for snapshot method."""
 

tests/test_debugger.py

@@ -0,0 +1,71 @@
+from __future__ import annotations
+
+from contextlib import asynccontextmanager
+from unittest.mock import AsyncMock, MagicMock, patch
+
+import pytest
+
+
+class MockRuntime:
+    def __init__(self) -> None:
+        self.begin_step = MagicMock(return_value="step-1")
+        self.emit_step_end = AsyncMock(return_value={"ok": True})
+        self.check = MagicMock(return_value="check-handle")
+
+
+class MockTracer:
+    def emit(self, event_type: str, data: dict, step_id: str | None = None) -> None:
+        pass
+
+
+@pytest.mark.asyncio
+async def test_attach_uses_runtime_factory() -> None:
+    mock_page = MagicMock()
+    tracer = MockTracer()
+    runtime = MockRuntime()
+
+    with patch(
+        "sentience.debugger.AgentRuntime.from_playwright_page", return_value=runtime
+    ) as mock_factory:
+        from sentience.debugger import SentienceDebugger
+
+        debugger = SentienceDebugger.attach(page=mock_page, tracer=tracer)
+
+    mock_factory.assert_called_once_with(
+        page=mock_page,
+        tracer=tracer,
+        snapshot_options=None,
+        sentience_api_key=None,
+        tool_registry=None,
+    )
+    assert debugger.runtime is runtime
+
+
+@pytest.mark.asyncio
+async def test_step_context_calls_begin_and_emit() -> None:
+    runtime = MockRuntime()
+
+    from sentience.debugger import SentienceDebugger
+
+    debugger = SentienceDebugger(runtime=runtime)
+
+    async with debugger.step("verify-cart"):
+        pass
+
+    runtime.begin_step.assert_called_once_with("verify-cart", step_index=None)
+    runtime.emit_step_end.assert_awaited_once()
+
+
+def test_check_auto_opens_step_when_missing() -> None:
+    runtime = MockRuntime()
+
+    from sentience.debugger import SentienceDebugger
+
+    debugger = SentienceDebugger(runtime=runtime)
+    predicate = MagicMock()
+
+    handle = debugger.check(predicate=predicate, label="has_cart", required=True)
+
+    runtime.begin_step.assert_called_once_with("verify:has_cart", step_index=None)
+    runtime.check.assert_called_once_with(predicate, "has_cart", required=True)
+    assert handle == "check-handle"