claude-smart 0.2.45 → 0.2.47

package/plugin/vendor/reflexio/reflexio/server/services/profile/{profile_extractor.py → components/extractor.py}

@@ -2,6 +2,7 @@ from __future__ import annotations
 
 import logging
 import os
+import uuid
 from datetime import UTC, datetime
 from typing import TYPE_CHECKING
 
@@ -12,11 +13,11 @@ from reflexio.models.api_schema.service_schemas import (
 from reflexio.models.config_schema import ProfileExtractorConfig
 from reflexio.server.api_endpoints.request_context import RequestContext
 from reflexio.server.llm.litellm_client import LiteLLMClient
+from reflexio.server.llm.token_accounting import RunTokenTotals, sum_trace_tokens
 from reflexio.server.services.extraction.outcome import ExtractionOutcome
 from reflexio.server.services.extraction.resumable_agent import (
     run_resumable_extraction_agent,
 )
-from reflexio.server.services.extraction.tools import new_profile_id
 from reflexio.server.services.extractor_config_utils import get_extractor_name
 from reflexio.server.services.extractor_interaction_utils import (
     get_effective_source_filter,
@@ -25,7 +26,7 @@ from reflexio.server.services.extractor_interaction_utils import (
 from reflexio.server.services.operation_state_utils import OperationStateManager
 
 if TYPE_CHECKING:
-    from reflexio.server.services.profile.profile_generation_service import (
+    from reflexio.server.services.profile.service import (
         ProfileGenerationServiceConfig,
     )
 from reflexio.server.llm.model_defaults import ModelRole, resolve_model_name
@@ -50,13 +51,26 @@ PROFILE_EXTRACTION_MAX_RETRIES = 2
 MAX_EXISTING_PROFILES_FOR_CONTEXT = 5
 
 
+def new_profile_id() -> str:
+    """Generate a short (12-char hex) profile id.
+
+    Format chosen for LLM copy fidelity: full ``str(uuid.uuid4())`` is 36
+    characters of hex+dashes, error-prone for smaller LLMs to copy verbatim.
+    Twelve hex chars is short enough for high-fidelity copy and long enough
+    that birthday-paradox collision probability is vanishingly small at any
+    realistic per-user scale (16^12 ~= 2.8e14 unique values; PRIMARY KEY
+    constraint catches the rare collision).
+    """
+    return uuid.uuid4().hex[:12]
+
+
 class ProfileExtractor:
     """
     Extract user profile information from interactions.
 
     This class analyzes user interactions to extract new user profile information.
     It focuses purely on extraction — deduplication against existing profiles
-    is handled separately by ProfileDeduplicator.
+    is handled separately by ProfileConsolidator.
     """
 
     def __init__(
@@ -83,6 +97,7 @@ class ProfileExtractor:
         self.service_config: ProfileGenerationServiceConfig = service_config
         self.agent_context = agent_context
         self._last_resumable_run_id: str | None = None
+        self._last_resumable_token_totals: RunTokenTotals | None = None
 
         # Get LLM config overrides from configuration
         config = self.request_context.configurator.get_config()
@@ -230,20 +245,17 @@ class ProfileExtractor:
             ) from e
 
         logger.info("Generated raw profiles: %s", raw_profiles)
-        if isinstance(raw_profiles, ExtractionOutcome):
-            user_profiles = self._convert_raw_to_user_profiles(
-                raw_profiles=raw_profiles.items,
-                user_id=self.service_config.user_id,
-                request_id=self.service_config.request_id,
-            )
-            self._update_operation_state(request_interaction_data_models)
-            return ExtractionOutcome.completed(
-                user_profiles, run_id=raw_profiles.run_id
-            )
+        source_interaction_ids = [
+            interaction.interaction_id
+            for request_model in request_interaction_data_models
+            for interaction in request_model.interactions
+            if interaction.interaction_id
+        ]
         user_profiles = self._convert_raw_to_user_profiles(
             raw_profiles=raw_profiles or [],
             user_id=self.service_config.user_id,
             request_id=self.service_config.request_id,
+            source_interaction_ids=source_interaction_ids,
         )
         if raw_profiles:
             # Update operation state (bookmark) only when output was produced.
@@ -258,6 +270,7 @@ class ProfileExtractor:
             return ExtractionOutcome.completed(
                 user_profiles,
                 run_id=self._last_resumable_run_id,
+                token_totals=self._last_resumable_token_totals,
             )
         return user_profiles or None
 
@@ -266,6 +279,7 @@ class ProfileExtractor:
         raw_profiles: list[dict],
         user_id: str,
         request_id: str,
+        source_interaction_ids: list[int] | None = None,
     ) -> list[UserProfile]:
         """
         Convert raw profile dicts from LLM to UserProfile objects.
@@ -274,11 +288,13 @@ class ProfileExtractor:
             raw_profiles: List of profile dicts with content, time_to_live, and optional metadata
             user_id: User ID
             request_id: Request ID
+            source_interaction_ids: Stored interaction ids used as the extraction window
 
         Returns:
             List of UserProfile objects
         """
         new_profiles = []
+        profile_source_interaction_ids = list(source_interaction_ids or [])
         for profile_content in raw_profiles:
             if (
                 not isinstance(profile_content, dict)
@@ -307,6 +323,7 @@ class ProfileExtractor:
                 expiration_timestamp=calculate_expiration_timestamp(now_ts, ttl),
                 custom_features=custom_features or None,
                 extractor_names=None,
+                source_interaction_ids=profile_source_interaction_ids,
             )
 
             new_profiles.append(added_profile)
@@ -342,11 +359,6 @@ class ProfileExtractor:
                 self.config.context_prompt.strip() if self.config.context_prompt else ""
             ),
             extraction_definition_prompt=self.config.extraction_definition_prompt.strip(),
-            metadata_definition_prompt=(
-                self.config.metadata_definition_prompt.strip()
-                if self.config.metadata_definition_prompt
-                else None
-            ),
             existing_profiles=existing_profiles,
         )
 
@@ -392,6 +404,7 @@ class ProfileExtractor:
             log_label="Profile extraction",
         )
         self._last_resumable_run_id = result.run_id
+        self._last_resumable_token_totals = sum_trace_tokens(result.trace)
         if not isinstance(result.output, StructuredProfilesOutput):
             logger.warning(
                 "Profile extraction did not finish: %s", result.finished_reason
@@ -455,8 +468,4 @@ class ProfileExtractor:
             "time_to_live": "one_month",
         }
 
-        # If metadata definition exists, add mock metadata
-        if self.config.metadata_definition_prompt:
-            mock_profile["metadata"] = "mock_metadata_value"
-
         return [mock_profile]

package/plugin/vendor/reflexio/reflexio/server/services/profile/profile_generation_service_utils.py

@@ -14,6 +14,7 @@ from reflexio.models.api_schema.service_schemas import (
     ProfileTimeToLive,
     UserProfile,
 )
+from reflexio.models.structured_output import StrictStructuredOutput
 from reflexio.server.prompt.prompt_manager import PromptManager
 from reflexio.server.services.service_utils import (
     MessageConstructionConfig,
@@ -74,7 +75,6 @@ class ProfileAddItem(BaseModel):
     Attributes:
         content (str): The profile content based on content definition
         time_to_live (str): Time to live for the profile - one of: 'one_day', 'one_week', 'one_month', 'one_quarter', 'one_year', 'infinity'
-        metadata (str, optional): Metadata extracted for the profile based on metadata definition
     """
 
     content: str = Field(description="Profile content based on content definition")
@@ -83,10 +83,6 @@ class ProfileAddItem(BaseModel):
     ] = Field(
         description="Time to live for the profile - determines when the profile expires"
     )
-    metadata: str | None = Field(
-        default=None,
-        description="Metadata extracted for the profile based on metadata definition",
-    )
     source_span: str | None = Field(
         default=None,
         description="Verbatim excerpt from the source that most directly supports this profile item",
@@ -107,7 +103,7 @@ class ProfileAddItem(BaseModel):
     )
 
 
-class ProfileUpdateOutput(BaseModel):
+class ProfileUpdateOutput(StrictStructuredOutput):
     """
     Legacy output schema for profile_update_main prompt (kept for backward compatibility).
     Represents the complete set of profile updates including additions, deletions, and mentions.
@@ -120,7 +116,7 @@ class ProfileUpdateOutput(BaseModel):
 
     add: list[ProfileAddItem] | None = Field(
         default=None,
-        description="List of new profiles to be added with their content, time to live, and optional metadata",
+        description="List of new profiles to be added with their content and time to live",
     )
     delete: list[str] | None = Field(
         default=None,
@@ -138,18 +134,18 @@ class ProfileUpdateOutput(BaseModel):
     )
 
 
-class StructuredProfilesOutput(BaseModel):
+class StructuredProfilesOutput(StrictStructuredOutput):
     """
     Output schema for extraction-only profile extraction.
     Only extracts profiles — no delete/mention operations.
 
     Attributes:
-        profiles (list[ProfileAddItem], optional): List of extracted profiles with content, time_to_live, and optional metadata
+        profiles (list[ProfileAddItem], optional): List of extracted profiles with content and time_to_live
     """
 
     profiles: list[ProfileAddItem] | None = Field(
         default=None,
-        description="List of extracted profiles with content, time_to_live, and optional metadata",
+        description="List of extracted profiles with content and time_to_live",
     )
 
     model_config = ConfigDict(
@@ -225,7 +221,6 @@ def construct_profile_extraction_messages_from_sessions(
     agent_context_prompt: str,
     context_prompt: str,
     extraction_definition_prompt: str,
-    metadata_definition_prompt: str | None = None,
 ) -> list[dict]:
     """
     Construct LLM messages for profile extraction from sessions.
@@ -241,7 +236,6 @@ def construct_profile_extraction_messages_from_sessions(
         agent_context_prompt: Context about the agent for system message
         context_prompt: Additional context for system message
         extraction_definition_prompt: Definition of what profiles should contain
-        metadata_definition_prompt: Optional definition for profile metadata
 
     Returns:
         list[dict]: List of messages ready for profile extraction
@@ -259,7 +253,6 @@ def construct_profile_extraction_messages_from_sessions(
             "agent_context_prompt": agent_context_prompt,
             "context_prompt": context_prompt,
             "extraction_definition_prompt": extraction_definition_prompt,
-            "metadata_definition_prompt": metadata_definition_prompt,
         },
     )
 

package/plugin/vendor/reflexio/reflexio/server/services/profile/{profile_generation_service.py → service.py}

@@ -5,7 +5,6 @@ from __future__ import annotations
 import logging
 import uuid
 from dataclasses import dataclass
-from datetime import UTC, datetime
 from typing import TYPE_CHECKING, Any
 
 if TYPE_CHECKING:
@@ -14,11 +13,9 @@ if TYPE_CHECKING:
 
 from reflexio.models.api_schema.internal_schema import RequestInteractionDataModel
 from reflexio.models.api_schema.service_schemas import (
-    DeleteUserProfileRequest,
     DowngradeProfilesResponse,
     ManualProfileGenerationRequest,
     ManualProfileGenerationResponse,
-    ProfileChangeLog,
     RerunProfileGenerationRequest,
     RerunProfileGenerationResponse,
     Status,
@@ -30,7 +27,7 @@ from reflexio.server.services.base_generation_service import (
     BaseGenerationService,
     StatusChangeOperation,
 )
-from reflexio.server.services.profile.profile_extractor import ProfileExtractor
+from reflexio.server.services.profile.components.extractor import ProfileExtractor
 from reflexio.server.services.profile.profile_generation_service_utils import (
     ProfileGenerationRequest,
     ProfileGenerationServiceConstants,
@@ -38,7 +35,7 @@ from reflexio.server.services.profile.profile_generation_service_utils import (
 from reflexio.server.services.service_utils import (
     format_sessions_to_history_string,
 )
-from reflexio.server.tracing import sentry_tags
+from reflexio.server.tracing import capture_anomaly, sentry_tags
 
 logger = logging.getLogger(__name__)
 
@@ -81,6 +78,9 @@ class ProfileGenerationService(
 ):
     """Service to generate user profiles from interactions"""
 
+    # Profile generation produces learnings — opt in to ② Learning billing.
+    EMITS_LEARNING_BILLING: bool = True
+
     def __init__(
         self,
         llm_client: LiteLLMClient,
@@ -154,24 +154,23 @@ class ProfileGenerationService(
         request_id = self.service_config.request_id  # type: ignore[reportOptionalMemberAccess]
 
         existing_ids_to_delete: list[str] = []
-        superseded_profiles: list[UserProfile] = []
 
         # Always run deduplicator when enabled and there are new profiles
         if all_new_profiles:
             from reflexio.server.site_var.feature_flags import is_deduplicator_enabled
 
             if is_deduplicator_enabled(self.org_id):
-                from reflexio.server.services.profile.profile_deduplicator import (
-                    ProfileDeduplicator,
+                from reflexio.server.services.profile.components.consolidator import (
+                    ProfileConsolidator,
                 )
 
-                deduplicator = ProfileDeduplicator(
+                consolidator = ProfileConsolidator(
                     request_context=self.request_context,
                     llm_client=self.client,
                     output_pending_status=self.output_pending_status,
                 )
-                all_new_profiles, existing_ids_to_delete, superseded_profiles = (
-                    deduplicator.deduplicate(all_new_profiles, user_id, request_id)
+                all_new_profiles, existing_ids_to_delete, _superseded_profiles = (
+                    consolidator.deduplicate(all_new_profiles, user_id, request_id)
                 )
                 logger.info(
                     "Profile updates after deduplication: %d profiles, %d existing to delete",
@@ -203,59 +202,42 @@ class ProfileGenerationService(
                     )
                 return
 
-        # Delete superseded existing profiles
+        # Always soft-supersede superseded existing profiles (never hard-delete on
+        # the dedup path). This emits the lineage events that the profile change log
+        # is reconstructed from (the legacy `profile_change_logs` table is no longer
+        # written — see reconstruct_profile_change_log).
         if existing_ids_to_delete:
-            for profile_id in existing_ids_to_delete:
+            if not request_id:
+                # An empty request_id makes the removal unreconstructable (the lineage
+                # events are keyed on it). Fail loud and skip removal entirely — never
+                # silently hard-delete.
+                capture_anomaly(
+                    "lineage.dedup.missing_request_id",
+                    level="error",
+                    org_id=self.org_id,
+                    user_id=user_id,
+                )
+            else:
                 try:
-                    self.storage.delete_user_profile(  # type: ignore[reportOptionalMemberAccess]
-                        DeleteUserProfileRequest(
-                            user_id=user_id,
-                            profile_id=profile_id,
-                        )
+                    self.storage.supersede_profiles_by_ids(  # type: ignore[reportOptionalMemberAccess]
+                        user_id=user_id,
+                        profile_ids=existing_ids_to_delete,
+                        request_id=request_id,
                     )
-                except Exception as e:  # noqa: PERF203
+                except Exception as e:
                     with sentry_tags(
                         subsystem="profile_generation",
-                        op="delete_superseded_profile",
+                        op="supersede_profiles",
                         org_id=self.org_id,
                         user_id=user_id,
                         request_id=request_id,
-                        profile_id=profile_id,
                         error_type=type(e).__name__,
                     ):
                         logger.exception(
-                            "Failed to delete superseded profile %s for user %s",
-                            profile_id,
+                            "Failed to soft-delete superseded profiles for user %s",
                             user_id,
                         )
 
-        # Create profile changelog post-deduplication
-        if all_new_profiles or superseded_profiles:
-            try:
-                profile_change_log = ProfileChangeLog(
-                    id=0,  # Auto-generated by storage
-                    user_id=user_id,
-                    request_id=request_id,
-                    created_at=int(datetime.now(UTC).timestamp()),
-                    added_profiles=all_new_profiles,
-                    removed_profiles=superseded_profiles,
-                    mentioned_profiles=[],
-                )
-                self.storage.add_profile_change_log(profile_change_log)  # type: ignore[reportOptionalMemberAccess]
-            except Exception as e:
-                with sentry_tags(
-                    subsystem="profile_generation",
-                    op="add_profile_change_log",
-                    org_id=self.org_id,
-                    user_id=user_id,
-                    request_id=request_id,
-                    error_type=type(e).__name__,
-                ):
-                    logger.exception(
-                        "Failed to add profile change log for user %s",
-                        user_id,
-                    )
-
     def check_and_update_profiles(self, profiles: list[UserProfile]) -> None:
         """check if the profiles are expired and update them if they are"""
         raise NotImplementedError

package/plugin/vendor/reflexio/reflexio/server/services/reflection/__init__.py

@@ -1,17 +1,13 @@
-"""Reflection service: critique-and-revise of cited memories after publish."""
+"""Reflection service package."""
 
-from reflexio.server.services.reflection.reflection_service import ReflectionService
 from reflexio.server.services.reflection.reflection_service_utils import (
-    ReflectionDecision,
-    ReflectionOutput,
-    ReflectionResult,
+    REFLECTION_OPERATION_NAME,
     ReflectionServiceRequest,
 )
+from reflexio.server.services.reflection.service import ReflectionService
 
 __all__ = [
+    "REFLECTION_OPERATION_NAME",
     "ReflectionService",
     "ReflectionServiceRequest",
-    "ReflectionDecision",
-    "ReflectionOutput",
-    "ReflectionResult",
 ]

package/plugin/vendor/reflexio/reflexio/server/services/reflection/components/__init__.py

@@ -0,0 +1,7 @@
+"""Reflection service internal components."""
+
+from reflexio.server.services.reflection.components.extractor import (
+    ReflectionExtractor,
+)
+
+__all__ = ["ReflectionExtractor"]

package/plugin/vendor/reflexio/reflexio/server/services/reflection/reflection_service_utils.py

@@ -11,11 +11,14 @@ in light of how they were applied across the window.
 
 from __future__ import annotations
 
+import uuid
 from typing import Literal
 
 from pydantic import BaseModel, Field
 
 from reflexio.models.api_schema.domain.enums import ProfileTimeToLive
+from reflexio.models.api_schema.validators import NonEmptyStr
+from reflexio.models.structured_output import StrictStructuredOutput
 
 REFLECTION_OPERATION_NAME = "reflection"
 
@@ -29,6 +32,13 @@ class ReflectionServiceRequest(BaseModel):
 
     Args:
         user_id (str): User to scope the bookmark and window to.
+        request_id (NonEmptyStr): The publish pass's own request id; used as the
+            lineage event ``request_id`` on revise events so B3
+            reconstruction can link revisions back to the triggering pass.
+            Defaults to a fresh UUID hex so two passes on the same profile
+            with no explicit request_id produce distinct lineage events.
+            Empty strings and whitespace-only values are rejected at
+            construction (``ValidationError``), before any storage write.
         agent_version (str): Agent version of the current publish; copied
             into replacement playbooks.
         source (str | None): Optional source filter for the window.
@@ -37,6 +47,7 @@ class ReflectionServiceRequest(BaseModel):
     """
 
     user_id: str
+    request_id: NonEmptyStr = Field(default_factory=lambda: uuid.uuid4().hex)
     agent_version: str = ""
     source: str | None = None
 
@@ -88,7 +99,7 @@ class ReflectionDecision(BaseModel):
     reason: str = ""
 
 
-class ReflectionOutput(BaseModel):
+class ReflectionOutput(StrictStructuredOutput):
     """Structured LLM output for one reflection pass."""
 
     decisions: list[ReflectionDecision] = Field(default_factory=list)

package/plugin/vendor/reflexio/reflexio/server/services/reflection/{reflection_service.py → service.py}

@@ -34,12 +34,14 @@ from typing import TYPE_CHECKING
 from reflexio.models.api_schema.domain.entities import (
     Citation,
     Interaction,
+    LineageContext,
     UserPlaybook,
     UserProfile,
 )
 from reflexio.server.llm.litellm_client import LiteLLMClient
 from reflexio.server.services.operation_state_utils import OperationStateManager
-from reflexio.server.services.reflection.reflection_extractor import (
+from reflexio.server.services.playbook.playbook_edit_apply import apply_playbook_edit
+from reflexio.server.services.reflection.components.extractor import (
     ReflectionExtractor,
 )
 from reflexio.server.services.reflection.reflection_service_utils import (
@@ -446,15 +448,19 @@ class ReflectionService:
         decision: ReflectionDecision,
         cited: UserProfile,
     ) -> bool:
-        """Insert the replacement profile, then archive the cited row.
+        """Insert the replacement profile, then supersede the cited row.
 
         Insert-first ordering means that if ``add_user_profile`` raises,
         the cited row stays current and the per-decision exception
         handler reports ``failed_count``. Only after the new row is
-        durable do we flip the cited row to ARCHIVED — and if *that*
-        fails we log at ERROR rather than silently dropping the user's
-        data, leaving a transient duplicate that downstream dedup can
-        clean up.
+        durable do we atomically supersede the cited row via
+        ``supersede_record``, which sets ``status=SUPERSEDED`` with a
+        ``superseded_by`` pointer and appends a ``revise`` lineage event.
+        If the CAS guard fails (incumbent no longer CURRENT), the
+        just-inserted successor is deleted (not audited); if that delete
+        itself raises the exception propagates to the caller. If
+        ``supersede_record`` raises, log at ERROR and accept a transient
+        duplicate rather than silently dropping user data.
         """
         storage = self.request_context.storage
         if storage is None:
@@ -481,14 +487,22 @@ class ReflectionService:
             new_content=new_profile.content,
         )
         storage.add_user_profile(cited.user_id, [new_profile])
+        ctx = LineageContext(
+            op_kind="revise",
+            actor="reflection",
+            request_id=request.request_id,
+        )
         try:
-            archived = storage.archive_profile_by_id(
-                user_id=request.user_id, profile_id=cited.profile_id
+            superseded = storage.supersede_record(
+                entity_type="profile",
+                incumbent_id=str(cited.profile_id),
+                successor_id=str(new_profile.profile_id),
+                context=ctx,
             )
         except Exception as exc:  # noqa: BLE001
             with sentry_tags(
                 subsystem="reflection",
-                op="archive_after_insert",
+                op="supersede_after_insert",
                 kind="profile",
                 org_id=self.request_context.org_id,
                 user_id=cited.user_id,
@@ -503,7 +517,12 @@ class ReflectionService:
                     new_profile.profile_id,
                 )
             return True
-        if not archived:
+        if not superseded:
+            # lost the CAS race: incumbent was no longer CURRENT — drop the
+            # just-added successor (never live) without emitting an audit event.
+            storage.delete_profiles_by_ids(
+                [new_profile.profile_id], emit_hard_delete=False
+            )
             with sentry_tags(
                 subsystem="reflection",
                 op="archive_after_insert_noop",
@@ -579,11 +598,13 @@ class ReflectionService:
                 (cited.content or "")[:120].replace('"', "'"),
                 (decision.new_rationale or "")[:120].replace('"', "'"),
             )
-        storage.save_user_playbooks([new_playbook])
         try:
-            archived = storage.archive_user_playbook_by_id(
-                user_id=owning_user_id,
-                user_playbook_id=cited.user_playbook_id,
+            archived = apply_playbook_edit(
+                storage,
+                incumbent_id=cited.user_playbook_id,
+                new_playbook=new_playbook,
+                source=new_playbook.source or "reflection",
+                request_id=request.request_id,
             )
         except Exception as exc:  # noqa: BLE001
             with sentry_tags(
@@ -603,7 +624,7 @@ class ReflectionService:
                     new_playbook.user_playbook_id,
                 )
             return True
-        if not archived:
+        if archived < 0:
             with sentry_tags(
                 subsystem="reflection",
                 op="archive_after_insert_noop",
@@ -799,5 +820,4 @@ def _filter_citations_by_horizon(
                     citation=cite, position=position, has_full_horizon=False
                 )
             )
-        # else: deferred — not included
     return out