refactor(core): type EvaluatorProtocol returns as AuditRecord

Closes doc item 1.2 — every ``evaluate_*`` method on
``EvaluatorProtocol`` returned ``-> Any``, forcing callers to downcast
to the type they already knew the concrete evaluator was returning.
The concrete ``GovernanceEvaluator`` in uipath-runtime-python already
declares ``-> AuditRecord`` on each per-hook method, so narrowing the
protocol contract is structurally compatible — no behavior change, no
downstream code change required.

Narrows the six evaluate_* return types (before_agent / after_agent /
before_model / after_model / tool_call / after_tool) from Any to
AuditRecord, imports the type from uipath.core.governance.models, and
refreshes the class docstring (was claiming the protocol is
intentionally Any because the audit record "lives in the plugin
package" — but AuditRecord lives right here in uipath-core).

Verified
- ruff clean, mypy clean (45 source files), 230 passed + 1 skipped in
  uipath-core.
- uipath-runtime-python's test suite (357 passed + 1 skipped) keeps
  green when this version of uipath-core is installed — the protocol-
  conformance tests in test_evaluator.py still pass because the
  concrete GovernanceEvaluator was already returning AuditRecord from
  every evaluate_* method.

Rides on the same 0.5.23 version bump as the previous commit — both
changes ship together as one public-surface change on uipath-core.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

Author: viswa-uipath

packages/uipath-core/src/uipath/core/adapters/evaluator.py

@@ -1,9 +1,9 @@
-"""Structural contract for the policy evaluator an adapter talks to.
+"""Structural contract for the policy evaluator a framework plugin talks to.
 
-Framework adapters call into a policy evaluator at each lifecycle hook.
+Framework plugins call into a policy evaluator at each lifecycle hook.
 Concrete evaluator implementations (the native runtime evaluator, a
 Microsoft AGT bridge, a composite, …) live in packages outside
-``uipath-core`` — adapters depend only on this structural protocol so
+``uipath-core`` — plugins depend only on this structural protocol so
 they can be swapped against any of them without code change.
 
 ``EvaluatorProtocol`` is a :class:`typing.Protocol` so any class whose
@@ -15,15 +15,18 @@
 
 from typing import Any, Protocol, runtime_checkable
 
+from uipath.core.governance.models import AuditRecord
+
 
 @runtime_checkable
 class EvaluatorProtocol(Protocol):
-    """Structural protocol an adapter expects from a policy evaluator.
+    """Structural protocol a framework plugin expects from a policy evaluator.
 
-    Return types are intentionally :class:`typing.Any`: the concrete
-    audit record shape lives in the plugin package that owns the
-    evaluator and the policy model. Adapters in that package cast the
-    return value back to the concrete type they know.
+    Every ``evaluate_*`` method returns an :class:`AuditRecord` — the
+    per-hook audit envelope holding the per-rule
+    :class:`RuleEvaluation` list, the final action, and the trace /
+    agent metadata. Callers get a typed result; no downcasting is
+    required.
     """
 
     def evaluate_before_agent(
@@ -34,7 +37,7 @@ def evaluate_before_agent(
         trace_id: str,
         model_name: str = "",
         **kwargs: Any,
-    ) -> Any:
+    ) -> AuditRecord:
         """Evaluate BEFORE_AGENT rules."""
         ...
 
@@ -45,7 +48,7 @@ def evaluate_after_agent(
         runtime_id: str,
         trace_id: str,
         **kwargs: Any,
-    ) -> Any:
+    ) -> AuditRecord:
         """Evaluate AFTER_AGENT rules."""
         ...
 
@@ -58,7 +61,7 @@ def evaluate_before_model(
         messages: list[dict[str, Any]] | None = None,
         model_name: str = "",
         **kwargs: Any,
-    ) -> Any:
+    ) -> AuditRecord:
         """Evaluate BEFORE_MODEL rules."""
         ...
 
@@ -69,7 +72,7 @@ def evaluate_after_model(
         runtime_id: str,
         trace_id: str,
         **kwargs: Any,
-    ) -> Any:
+    ) -> AuditRecord:
         """Evaluate AFTER_MODEL rules."""
         ...
 
@@ -82,7 +85,7 @@ def evaluate_tool_call(
         trace_id: str,
         session_state: dict[str, Any] | None = None,
         **kwargs: Any,
-    ) -> Any:
+    ) -> AuditRecord:
         """Evaluate TOOL_CALL rules."""
         ...
 
@@ -94,6 +97,6 @@ def evaluate_after_tool(
         runtime_id: str,
         trace_id: str,
         **kwargs: Any,
-    ) -> Any:
+    ) -> AuditRecord:
         """Evaluate AFTER_TOOL rules."""
         ...