fuli-context 0.7.4-beta.3 → 0.7.5

package/graph-provider/fuli_graph/store_project_agent_task_reads.py

@@ -0,0 +1,458 @@
+"""Task row reconstruction, terminal cleanup, and route-result presentation."""
+
+from __future__ import annotations
+
+import hashlib
+import json
+
+from .project_agent_models import ProjectAgentModelStrategy, ProjectAgentProfile
+from .project_agent_task_models import (
+    ProjectAgentRoutingDecisionRecord,
+    ProjectAgentTaskExecutionSummary,
+    ProjectAgentTaskEventRecord,
+    ProjectAgentTaskParticipantRecord,
+    ProjectAgentTaskRecord,
+    ProjectAgentTaskRouteResult,
+)
+from .provider_values import native_datetime
+
+
+TERMINAL_TASK_STATUSES = {'completed', 'failed', 'cancelled'}
+
+
+class StoreProjectAgentTaskReads:
+    """Read durable task history and derive client-facing route results."""
+
+    async def _find_task_row(self, personal_space_id, task_id):
+        records, _, _ = await self.runtime.driver.execute_query(
+            self._task_read_query('WHERE task.task_id = $task_id'),
+            personal_space_id=personal_space_id,
+            task_id=task_id,
+            routing_='r',
+        )
+        return records[0] if records else None
+
+    @staticmethod
+    def _task_read_query(where_clause):
+        return f'''
+        MATCH (space:FuliSpace {{id: $personal_space_id, kind: 'personal'}})-
+              [:HAS_PROJECT_AGENT_TASK]->
+              (task:FuliProjectAgentTask {{
+                personal_space_id: $personal_space_id
+              }})
+        {where_clause}
+        OPTIONAL MATCH (task)-[participant:HAS_PARTICIPANT]->
+                       (participant_agent:FuliProjectAgent)
+        WITH task, collect(DISTINCT {{
+          agent_id: participant_agent.agent_id,
+          agent_name: participant_agent.name,
+          occupation_emoji: participant_agent.occupation_emoji,
+          role: participant.role,
+          status: participant.status,
+          assignment_summary: participant.assignment_summary,
+          profile_json: participant_agent.profile_json,
+          joined_at: participant.joined_at,
+          updated_at: participant.updated_at,
+          ended_at: participant.ended_at
+        }}) AS participant_rows
+        OPTIONAL MATCH (task)-[:HAS_TASK_EVENT]->
+                       (event:FuliProjectAgentTaskEvent)
+        WITH task, participant_rows, collect(DISTINCT event) AS event_rows
+        OPTIONAL MATCH (task)-[:ROUTED_BY]->
+                       (decision:FuliProjectAgentRoutingDecision)
+        RETURN task, participant_rows, event_rows, decision
+        '''
+
+    def _task_from_row(self, row):
+        raw = dict(row['task'])
+        decision_raw = dict(row['decision']) if row.get('decision') else {}
+        if decision_raw.get('decision_json'):
+            decision = ProjectAgentRoutingDecisionRecord.model_validate_json(
+                decision_raw['decision_json']
+            )
+        else:
+            decision = ProjectAgentRoutingDecisionRecord(
+                decision_id=decision_raw.get('decision_id') or 'legacy',
+                task_id=raw['task_id'],
+                coordinator_agent_id=raw['coordinator_agent_id'],
+                complexity=raw.get('complexity') or 'standard',
+                complexity_basis=list(raw.get('complexity_basis') or []),
+                selected_model_strategy=(
+                    ProjectAgentModelStrategy.model_validate_json(
+                        raw['effective_model_strategy_json']
+                    )
+                    if raw.get('effective_model_strategy_json') else None
+                ),
+                model_strategy_source=raw.get('model_strategy_source') or 'coordinator',
+                outcome=raw.get('routing_outcome') or 'unassigned',
+                reason=raw.get('routing_reason') or 'no_match',
+                match_basis=list(raw.get('match_basis') or []),
+                created_at=native_datetime(raw['created_at']),
+            )
+        events = []
+        for value in row.get('event_rows') or []:
+            if value is None:
+                continue
+            event = dict(value)
+            if not event.get('event_id'):
+                continue
+            fields = {
+                'event_id': event['event_id'],
+                'task_id': event['task_id'],
+                'agent_id': event.get('agent_id'),
+                'status': event['status'],
+                'actor_kind': event.get('actor_kind') or 'system',
+                'summary': event['summary'],
+                'source_application': event.get('source_application'),
+                'source_session_id': event.get('source_session_id'),
+                'actual_model_provider': event.get('actual_model_provider'),
+                'actual_model': event.get('actual_model'),
+                'created_at': native_datetime(event['created_at']),
+            }
+            for name in (
+                'actual_executor_id',
+                'matched_executor_rule_id',
+                'executor_selection_reason',
+                'executor_fallback_reason',
+                'executor_blocked_reason',
+                'worker_id',
+                'worker_label',
+                'worker_occupation_emoji',
+                'worker_status',
+            ):
+                if name in ProjectAgentTaskEventRecord.model_fields:
+                    fields[name] = event.get(name)
+            events.append(ProjectAgentTaskEventRecord(**fields))
+        events.sort(key=lambda item: item.created_at)
+        fields = {
+            'task_id': raw['task_id'],
+            'personal_space_id': raw['personal_space_id'],
+            'personal_project_id': raw['personal_project_id'],
+            'title': raw['title'],
+            'objective': raw['objective'],
+            'work_kind': raw['work_kind'],
+            'required_capabilities': list(raw.get('required_capabilities') or []),
+            'duration': raw.get('duration') or 'ongoing',
+            'staffing_intent': raw.get('staffing_intent') or 'reuse_preferred',
+            'status': raw['status'],
+            'revision': int(raw.get('revision') or 0),
+            'routing_outcome': raw.get('routing_outcome') or 'unassigned',
+            'routing_reason': raw.get('routing_reason') or 'no_match',
+            'routing_explanation': raw.get('routing_explanation') or '',
+            'match_basis': list(raw.get('match_basis') or []),
+            'coordinator_agent_id': raw['coordinator_agent_id'],
+            'complexity': raw.get('complexity') or 'standard',
+            'complexity_basis': list(raw.get('complexity_basis') or []),
+            'routing_decision': decision,
+            'lead_agent_id': raw.get('lead_agent_id'),
+            'participants': [
+                ProjectAgentTaskParticipantRecord(
+                    agent_id=value['agent_id'],
+                    role=value['role'],
+                    status=value['status'],
+                    assignment_summary=value.get('assignment_summary'),
+                    joined_at=native_datetime(value['joined_at']),
+                    updated_at=native_datetime(value['updated_at']),
+                    ended_at=native_datetime(value.get('ended_at')),
+                )
+                for value in self._participant_maps(row)
+            ],
+            'effective_model_strategy': (
+                ProjectAgentModelStrategy.model_validate_json(
+                    raw['effective_model_strategy_json']
+                )
+                if raw.get('effective_model_strategy_json') else None
+            ),
+            'model_strategy_source': raw.get('model_strategy_source') or 'coordinator',
+            'hr_agent_id': raw.get('hr_agent_id'),
+            'recruitment_id': raw.get('recruitment_id'),
+            'source_application': raw.get('source_application'),
+            'source_session_id': raw.get('source_session_id'),
+            'result_summary': raw.get('result_summary'),
+            'failure_reason': raw.get('failure_reason'),
+            'created_at': native_datetime(raw['created_at']),
+            'updated_at': native_datetime(raw['updated_at']),
+            'completed_at': native_datetime(raw.get('completed_at')),
+            'events': events,
+        }
+        fields['execution_summary'] = self._execution_summary(row, events)
+        executor_raw = json.loads(raw['executor_decision_json']) \
+            if raw.get('executor_decision_json') else None
+        executor_policy_raw = raw.get('executor_policy_json')
+        if executor_policy_raw:
+            fields['executor_policy'] = json.loads(executor_policy_raw)
+        for name, value in {
+            'selected_executor_id': raw.get('selected_executor_id'),
+            'actual_executor_id': raw.get('actual_executor_id'),
+            'actual_run_id': raw.get('actual_run_id'),
+            'actual_model_provider': raw.get('actual_model_provider'),
+            'actual_model': raw.get('actual_model'),
+            'matched_executor_rule_id': raw.get('matched_executor_rule_id'),
+            'executor_selection_reason': raw.get('executor_selection_reason'),
+            'executor_fallback_outcome': raw.get('executor_fallback_outcome'),
+            'executor_fallback_reason': raw.get('executor_fallback_reason'),
+            'executor_blocked_reason': raw.get('executor_blocked_reason'),
+            'executor_decision': executor_raw,
+        }.items():
+            if name in ProjectAgentTaskRecord.model_fields:
+                fields[name] = value
+        return ProjectAgentTaskRecord(**fields)
+
+    @staticmethod
+    def _participant_maps(row):
+        return [
+            dict(value)
+            for value in (row.get('participant_rows') or [])
+            if value and value.get('agent_id') and value.get('role')
+        ]
+
+    @staticmethod
+    def _participant_profile(participant):
+        """Return the durable display fields, tolerating legacy profiles."""
+
+        profile_json = participant.get('profile_json')
+        profile = None
+        if profile_json:
+            try:
+                profile = ProjectAgentProfile.model_validate_json(profile_json)
+            except (TypeError, ValueError):
+                # A malformed legacy profile must not make an otherwise
+                # readable task invent identity data or fail the whole read.
+                profile = None
+        agent_name = (
+            profile.name if profile else participant.get('agent_name')
+        )
+        occupation_emoji = (
+            profile.occupation_emoji if profile else None
+        ) or participant.get('occupation_emoji')
+        if profile is None and profile_json:
+            try:
+                legacy = json.loads(profile_json)
+            except (TypeError, ValueError):
+                legacy = {}
+            agent_name = agent_name or legacy.get('name')
+            occupation_emoji = occupation_emoji or legacy.get('occupationEmoji')
+            occupation_emoji = occupation_emoji or legacy.get('occupation_emoji')
+        return agent_name, occupation_emoji
+
+    @classmethod
+    def _execution_summary(cls, row, events):
+        """Project only observed events onto identities actually assigned."""
+
+        result = []
+        participants = cls._participant_maps(row)
+        task_raw = dict(row.get('task') or {})
+        task_fallback = task_raw if len(participants) == 1 else {}
+        for participant in participants:
+            agent_id = participant['agent_id']
+            participant_events = [
+                event for event in events
+                if event.agent_id == agent_id
+                and cls._is_execution_event(event)
+            ]
+            agent_name, occupation_emoji = cls._participant_profile(participant)
+            worker_groups = {}
+            for event in participant_events:
+                if event.worker_id:
+                    worker_groups.setdefault(event.worker_id, []).append(event)
+
+            # A worker id is an explicit observation of a concrete execution
+            # worker.  Preserve one row per observed worker, even when several
+            # workers share one durable Agent identity.
+            if worker_groups:
+                for worker_id, worker_events in worker_groups.items():
+                    result.append(
+                        cls._execution_summary_row(
+                            participant,
+                            agent_name,
+                            occupation_emoji,
+                            worker_events,
+                            worker_id=worker_id,
+                            task_fallback={},
+                        )
+                    )
+                continue
+
+            # A configured participant is not execution evidence. Legacy
+            # clients may omit worker_id, but they must still have recorded an
+            # Agent-linked running/terminal event or concrete executor/model
+            # evidence before a summary row exists.
+            if not participant_events:
+                continue
+            result.append(
+                cls._execution_summary_row(
+                    participant,
+                    agent_name,
+                    occupation_emoji,
+                    participant_events,
+                    task_fallback=task_fallback,
+                )
+            )
+        return result
+
+    @staticmethod
+    def _is_execution_event(event):
+        return bool(
+            event.worker_id
+            or event.worker_status
+            or event.actual_executor_id
+            or event.actual_model_provider
+            or event.actual_model
+            or event.status in {
+                'running', 'paused', 'awaiting_review', 'blocked',
+                'completed', 'failed', 'cancelled',
+            }
+        )
+
+    @staticmethod
+    def _execution_summary_row(
+        participant,
+        agent_name,
+        occupation_emoji,
+        events,
+        *,
+        worker_id=None,
+        task_fallback=None,
+    ):
+        task_fallback = task_fallback or {}
+        latest = events[-1] if events else None
+
+        def latest_value(name):
+            return next(
+                (
+                    getattr(event, name)
+                    for event in reversed(events)
+                    if getattr(event, name, None) is not None
+                ),
+                None,
+            )
+
+        actual_executor_id = (
+            latest_value('actual_executor_id')
+            or task_fallback.get('actual_executor_id')
+        )
+        worker_label = latest_value('worker_label')
+        worker_occupation_emoji = latest_value('worker_occupation_emoji')
+        worker_status = latest_value('worker_status')
+        status = worker_status or (
+            latest.status
+            if latest
+            else participant['status']
+        )
+        return ProjectAgentTaskExecutionSummary(
+            agent_id=participant['agent_id'],
+            agent_name=agent_name,
+            occupation_emoji=occupation_emoji,
+            participant_role=participant['role'],
+            executor=actual_executor_id,
+            executor_id=actual_executor_id,
+            source_application=latest_value('source_application'),
+            actual_model_provider=(
+                latest_value('actual_model_provider')
+                or task_fallback.get('actual_model_provider')
+            ),
+            actual_model=(
+                latest_value('actual_model')
+                or task_fallback.get('actual_model')
+            ),
+            work_summary=(
+                latest.summary
+                if latest
+                else task_fallback.get('result_summary')
+                or task_fallback.get('failure_reason')
+            ),
+            status=status,
+            worker_id=worker_id,
+            worker_label=worker_label,
+            worker_occupation_emoji=worker_occupation_emoji,
+        )
+
+    async def _archive_finished_temporary_agent(self, request, raw_task, updated_at):
+        if request.status not in TERMINAL_TASK_STATUSES:
+            return
+        await self.runtime.driver.execute_query(
+            '''
+            MATCH (space:FuliSpace {
+              id: $personal_space_id, kind: 'personal'
+            })-[:HAS_PROJECT_AGENT_TASK]->
+                  (task:FuliProjectAgentTask {task_id: $task_id})-
+                  [:HAS_PARTICIPANT]->
+                  (agent:FuliProjectAgent {
+                    agent_type: 'temporary',
+                    temporary_task_id: $task_id
+                  })
+            SET agent.status = 'archived',
+                agent.updated_at = $updated_at
+            WITH space, agent
+            MATCH (space)-[:CONTAINS_PROJECT]->(:FuliPersonalProject)-
+                  [:HAS_PROJECT_AGENT_ASSIGNMENT]->
+                  (assignment:FuliProjectAgentAssignment {status: 'active'})-
+                  [:ASSIGNED_AGENT]->(agent)
+            SET assignment.status = 'ended',
+                assignment.end_reason = 'temporary task reached terminal state',
+                assignment.ended_at = $updated_at,
+                assignment.updated_at = $updated_at,
+                assignment.revision = assignment.revision + 1
+            ''',
+            personal_space_id=request.personal_space_id,
+            task_id=request.task_id,
+            updated_at=updated_at,
+        )
+
+    @staticmethod
+    def _routing_explanation(reason, match_basis, executor_decision):
+        basis = '; '.join(match_basis) if match_basis else 'no exact assignment match'
+        value = f'{reason}: {basis}'
+        if executor_decision:
+            raw = (
+                executor_decision.model_dump(mode='json')
+                if hasattr(executor_decision, 'model_dump')
+                else executor_decision
+            )
+            executor_reason = raw.get('selection_reason') or raw.get('reason')
+            if executor_reason:
+                value += f'; executor: {executor_reason}'
+        return value
+
+    @staticmethod
+    def _payload_hash(request):
+        payload = request.model_dump(mode='json', exclude={'idempotency_key'})
+        return hashlib.sha256(
+            json.dumps(payload, sort_keys=True, separators=(',', ':')).encode()
+        ).hexdigest()
+
+    def _route_result(self, task, recruitment, *, assigned_agent=None):
+        if recruitment and recruitment.status == 'fulfilled':
+            decision = 'recruited'
+        elif recruitment and recruitment.status == 'awaiting_confirmation':
+            decision = 'awaiting_confirmation'
+        elif task.routing_outcome == 'assigned_existing':
+            decision = 'reused'
+        elif task.routing_outcome == 'unassigned':
+            decision = 'unassigned'
+        else:
+            decision = 'blocked'
+        notice = None
+        must_disclose = bool(
+            recruitment
+            and recruitment.confirmation_mode == 'automatic'
+            and recruitment.status == 'fulfilled'
+        )
+        if must_disclose:
+            notice = (
+                f'HR {recruitment.hr_agent_id} recruited '
+                f'{recruitment.proposed_profile.name} as a '
+                f'{recruitment.position_kind} Agent for '
+                f'{recruitment.proposed_profile.responsibility}. '
+                f'Reason: {recruitment.reason}. '
+                f'Trigger: {recruitment.trigger_source_application or "unspecified"}; '
+                f'time: {recruitment.fulfilled_at.isoformat() if recruitment.fulfilled_at else "pending"}.'
+            )
+        return ProjectAgentTaskRouteResult(
+            task=task,
+            assigned_agent=assigned_agent,
+            recruitment=recruitment,
+            decision=decision,
+            must_disclose_recruitment=must_disclose,
+            client_notice=notice,
+        )