fuli-context 0.7.4-beta.3 → 0.7.6

package/graph-provider/fuli_graph/project_agent_executor_models.py

@@ -0,0 +1,635 @@
+"""Durable executor-directory and routing contracts for Project Agents.
+
+The executor layer deliberately stores capabilities and observed availability,
+not a hard-coded mapping from a work kind to a tool or vendor.  A coordinator
+can only select an executor after the directory, workspace permission,
+preflight, capability, and (when required) health checks all pass.
+"""
+
+import hashlib
+import json
+from datetime import datetime
+from typing import Literal
+
+from pydantic import Field, field_validator, model_validator
+
+from .model_validation import reject_credentials
+from .models import SourceApplication, StrictModel
+from .project_agent_models import (
+    ProjectAgentExecutorPolicy,
+    ProjectAgentModelStrategy,
+)
+
+
+ProjectAgentExecutorRegistrationStatus = Literal[
+    'registered',
+    'disabled',
+    'revoked',
+]
+ProjectAgentExecutorPermissionStatus = Literal[
+    'pending',
+    'authorized',
+    'denied',
+    'revoked',
+]
+ProjectAgentExecutorPreflightStatus = Literal[
+    'not_run',
+    'passed',
+    'failed',
+    'expired',
+]
+ProjectAgentExecutorHealthStatus = Literal[
+    'unknown',
+    'healthy',
+    'degraded',
+    'unhealthy',
+]
+ProjectAgentExecutorScope = Literal['global', 'space', 'project', 'task']
+ProjectAgentExecutorRuleStatus = Literal['active', 'disabled', 'ended']
+ProjectAgentExecutorSelectionStatus = Literal[
+    'selected',
+    'fallback',
+    'blocked',
+    'unavailable',
+]
+ProjectAgentExecutorFallbackOutcome = Literal[
+    'not_needed',
+    'same_rule_candidate',
+    'global_priority',
+    'blocked_locked',
+    'blocked_no_candidate',
+]
+ProjectAgentExecutorEvidenceKind = Literal[
+    'terminal_outcome',
+    'rework_requested',
+    'repeated_negative_feedback',
+    'explicit_praise',
+    'test_passed',
+    'test_failed',
+    'acceptance_passed',
+    'acceptance_failed',
+    'explicit_rating',
+]
+ProjectAgentExecutorEvidenceSource = Literal[
+    'system_terminal',
+    'user_explicit',
+    'test_fact',
+]
+ProjectAgentExecutorTerminalOutcome = Literal[
+    'completed',
+    'failed',
+    'cancelled',
+]
+
+
+def project_agent_model_strategy_key(
+    strategy: ProjectAgentModelStrategy,
+) -> str:
+    """Return the canonical, provider-neutral key for a model strategy."""
+
+    payload = strategy.model_dump(mode='json')
+    # Capability hints are a set of requirements, not an ordered preference.
+    # Canonicalizing them prevents equivalent provider-neutral strategies from
+    # creating separate evidence/aggregate buckets.
+    payload['capability_hints'] = sorted(
+        (value.casefold() for value in payload.get('capability_hints', [])),
+    )
+    encoded = json.dumps(
+        payload,
+        ensure_ascii=False,
+        sort_keys=True,
+        separators=(',', ':'),
+    ).encode('utf-8')
+    return hashlib.sha256(encoded).hexdigest()
+
+
+def _unique_text(value: list[str], label: str, *, maximum: int = 512) -> list[str]:
+    normalized = [item.strip() for item in value]
+    if any(not item or len(item) > maximum for item in normalized):
+        raise ValueError(f'{label} must contain 1 to {maximum} characters')
+    if len({item.casefold() for item in normalized}) != len(normalized):
+        raise ValueError(f'{label} must be unique')
+    return normalized
+
+
+def _unique_ids(value: list[str], label: str = 'IDs') -> list[str]:
+    return _unique_text(value, label, maximum=128)
+
+
+class ProjectAgentExecutorModelRecord(StrictModel):
+    """A model actually advertised by an executor preflight.
+
+    ``provider`` and ``model`` are observations from the executor.  Routing
+    continues to use the provider-neutral ``ProjectAgentModelStrategy`` and
+    these fields are only selected after a preflight reports them available.
+    """
+
+    provider: str = Field(min_length=1, max_length=128)
+    model: str = Field(min_length=1, max_length=256)
+    capabilities: list[str] = Field(default_factory=list, max_length=32)
+    strategy_modes: list[Literal['adaptive', 'fast', 'balanced', 'deep']] = Field(
+        default_factory=list,
+        max_length=4,
+    )
+    reasoning_efforts: list[Literal['default', 'low', 'medium', 'high']] = Field(
+        default_factory=list,
+        max_length=4,
+    )
+    available: bool = True
+    observed_at: datetime | None = None
+    source_application: SourceApplication | None = None
+    unavailable_reason: str | None = Field(default=None, min_length=1, max_length=1024)
+
+    @field_validator('capabilities')
+    @classmethod
+    def validate_capabilities(cls, value: list[str]) -> list[str]:
+        return _unique_text(value, 'executor model capabilities')
+
+    @model_validator(mode='after')
+    def validate_availability(self):
+        if self.available and self.unavailable_reason:
+            raise ValueError('available model cannot have an unavailable reason')
+        if not self.available and not self.unavailable_reason:
+            raise ValueError('unavailable model requires a reason')
+        reject_credentials(self, 'executor model report')
+        return self
+
+
+class ProjectAgentExecutorRegistration(StrictModel):
+    personal_space_id: str = Field(min_length=1, max_length=128)
+    executor_id: str = Field(min_length=1, max_length=128)
+    display_name: str = Field(min_length=1, max_length=160)
+    executor_kind: str = Field(default='external', min_length=1, max_length=128)
+    capabilities: list[str] = Field(default_factory=list, max_length=64)
+    advertised_models: list[ProjectAgentExecutorModelRecord] = Field(
+        default_factory=list,
+        max_length=64,
+    )
+    global_priority: int = Field(default=100, ge=1, le=1_000_000)
+    health_required: bool = False
+    idempotency_key: str = Field(min_length=8, max_length=256)
+    source_application: SourceApplication | None = None
+    source_session_id: str | None = Field(default=None, min_length=1, max_length=256)
+    test_source: str | None = Field(default=None, min_length=1, max_length=256)
+    cleanup_eligible: bool = False
+    expected_revision: int | None = Field(default=None, ge=0)
+
+    @field_validator('capabilities')
+    @classmethod
+    def validate_capabilities(cls, value: list[str]) -> list[str]:
+        return _unique_text(value, 'executor capabilities')
+
+    @model_validator(mode='after')
+    def reject_sensitive_values(self):
+        reject_credentials(self, 'executor registration')
+        return self
+
+
+class ProjectAgentExecutorAuthorization(StrictModel):
+    personal_space_id: str = Field(min_length=1, max_length=128)
+    executor_id: str = Field(min_length=1, max_length=128)
+    status: ProjectAgentExecutorPermissionStatus
+    reason: str = Field(min_length=1, max_length=2048)
+    expected_revision: int | None = Field(default=None, ge=0)
+    idempotency_key: str = Field(min_length=8, max_length=256)
+
+    @model_validator(mode='after')
+    def reject_sensitive_values(self):
+        reject_credentials(self, 'executor workspace authorization')
+        return self
+
+
+class ProjectAgentExecutorPreflightReport(StrictModel):
+    personal_space_id: str = Field(min_length=1, max_length=128)
+    executor_id: str = Field(min_length=1, max_length=128)
+    status: Literal['passed', 'failed']
+    workspace_permission: bool
+    capabilities: list[str] = Field(default_factory=list, max_length=64)
+    available_models: list[ProjectAgentExecutorModelRecord] = Field(
+        default_factory=list,
+        max_length=64,
+    )
+    reason: str | None = Field(default=None, min_length=1, max_length=2048)
+    checked_at: datetime
+    idempotency_key: str = Field(min_length=8, max_length=256)
+    source_application: SourceApplication | None = None
+    source_session_id: str | None = Field(default=None, min_length=1, max_length=256)
+
+    @field_validator('capabilities')
+    @classmethod
+    def validate_capabilities(cls, value: list[str]) -> list[str]:
+        return _unique_text(value, 'preflight capabilities')
+
+    @model_validator(mode='after')
+    def validate_report(self):
+        if self.status == 'passed':
+            if not self.workspace_permission:
+                raise ValueError('passed preflight requires workspace permission')
+            if not any(model.available for model in self.available_models):
+                raise ValueError('passed preflight requires an available model')
+        if self.status == 'failed' and not self.reason:
+            raise ValueError('failed preflight requires a reason')
+        reject_credentials(self, 'executor preflight report')
+        return self
+
+
+class ProjectAgentExecutorHealthReport(StrictModel):
+    personal_space_id: str = Field(min_length=1, max_length=128)
+    executor_id: str = Field(min_length=1, max_length=128)
+    status: ProjectAgentExecutorHealthStatus
+    reason: str | None = Field(default=None, min_length=1, max_length=2048)
+    checked_at: datetime
+    idempotency_key: str = Field(min_length=8, max_length=256)
+    source_application: SourceApplication | None = None
+    source_session_id: str | None = Field(default=None, min_length=1, max_length=256)
+
+    @model_validator(mode='after')
+    def validate_report(self):
+        if self.status in {'degraded', 'unhealthy'} and not self.reason:
+            raise ValueError('degraded or unhealthy executor requires a reason')
+        reject_credentials(self, 'executor health report')
+        return self
+
+
+class ProjectAgentExecutorRecord(StrictModel):
+    executor_id: str
+    display_name: str
+    executor_kind: str
+    registration_status: ProjectAgentExecutorRegistrationStatus
+    permission_status: ProjectAgentExecutorPermissionStatus
+    preflight_status: ProjectAgentExecutorPreflightStatus
+    health_status: ProjectAgentExecutorHealthStatus
+    health_required: bool = False
+    workspace_permission: bool = False
+    capabilities: list[str] = Field(default_factory=list)
+    available_models: list[ProjectAgentExecutorModelRecord] = Field(
+        default_factory=list,
+    )
+    global_priority: int = Field(ge=1)
+    revision: int = Field(default=0, ge=0)
+    permission_revision: int = Field(default=0, ge=0)
+    preflight_at: datetime | None = None
+    health_checked_at: datetime | None = None
+    registered_at: datetime
+    updated_at: datetime
+    test_source: str | None = None
+    cleanup_eligible: bool = False
+
+
+class ProjectAgentExecutorPriorityUpdate(StrictModel):
+    executor_id: str = Field(min_length=1, max_length=128)
+    global_priority: int = Field(ge=1, le=1_000_000)
+    expected_revision: int = Field(ge=0)
+    idempotency_key: str = Field(min_length=8, max_length=256)
+    reason: str = Field(min_length=1, max_length=2048)
+
+    @model_validator(mode='after')
+    def reject_sensitive_values(self):
+        reject_credentials(self, 'executor priority update')
+        return self
+
+
+class ProjectAgentExecutorRoutingRuleCreate(StrictModel):
+    scope: ProjectAgentExecutorScope
+    personal_space_id: str | None = Field(default=None, min_length=1, max_length=128)
+    personal_project_id: str | None = Field(default=None, min_length=1, max_length=128)
+    task_id: str | None = Field(default=None, min_length=1, max_length=128)
+    work_kind: str = Field(min_length=1, max_length=128)
+    required_capabilities: list[str] = Field(default_factory=list, max_length=32)
+    executor_ids: list[str] = Field(min_length=1, max_length=32)
+    model_strategy: ProjectAgentModelStrategy | None = None
+    priority: int = Field(default=100, ge=1, le=1_000_000)
+    reason: str = Field(min_length=1, max_length=2048)
+    idempotency_key: str = Field(min_length=8, max_length=256)
+
+    @field_validator('required_capabilities')
+    @classmethod
+    def validate_capabilities(cls, value: list[str]) -> list[str]:
+        return _unique_text(value, 'routing rule capabilities')
+
+    @field_validator('executor_ids')
+    @classmethod
+    def validate_executor_ids(cls, value: list[str]) -> list[str]:
+        return _unique_ids(value, 'routing rule executor IDs')
+
+    @model_validator(mode='after')
+    def validate_scope(self):
+        expected = {
+            'global': (False, False, False),
+            'space': (True, False, False),
+            'project': (True, True, False),
+            'task': (True, True, True),
+        }[self.scope]
+        actual = (
+            bool(self.personal_space_id),
+            bool(self.personal_project_id),
+            bool(self.task_id),
+        )
+        if actual != expected:
+            raise ValueError(
+                f'{self.scope} routing rule requires exactly the matching scope IDs'
+            )
+        reject_credentials(self, 'executor routing rule')
+        return self
+
+
+class ProjectAgentExecutorRoutingRuleRecord(
+    ProjectAgentExecutorRoutingRuleCreate
+):
+    rule_id: str
+    status: ProjectAgentExecutorRuleStatus = 'active'
+    revision: int = Field(default=0, ge=0)
+    created_at: datetime
+    updated_at: datetime
+
+
+class ProjectAgentExecutorRoutingRuleUpdate(StrictModel):
+    personal_space_id: str = Field(min_length=1, max_length=128)
+    rule_id: str = Field(min_length=1, max_length=128)
+    expected_revision: int = Field(ge=0)
+    status: Literal['active', 'disabled', 'ended']
+    reason: str = Field(min_length=1, max_length=2048)
+    idempotency_key: str = Field(min_length=8, max_length=256)
+
+    @model_validator(mode='after')
+    def reject_sensitive_values(self):
+        reject_credentials(self, 'executor routing rule update')
+        return self
+
+
+class ProjectAgentExecutorSelectionRequest(StrictModel):
+    personal_space_id: str = Field(min_length=1, max_length=128)
+    personal_project_id: str = Field(min_length=1, max_length=128)
+    task_id: str = Field(min_length=1, max_length=128)
+    agent_id: str = Field(min_length=1, max_length=128)
+    work_kind: str = Field(min_length=1, max_length=128)
+    required_capabilities: list[str] = Field(default_factory=list, max_length=32)
+    model_strategy: ProjectAgentModelStrategy = Field(
+        default_factory=ProjectAgentModelStrategy
+    )
+    executor_policy: ProjectAgentExecutorPolicy | None = None
+    idempotency_key: str = Field(min_length=8, max_length=256)
+    source_application: SourceApplication | None = None
+    source_session_id: str | None = Field(default=None, min_length=1, max_length=256)
+
+    @field_validator('required_capabilities')
+    @classmethod
+    def validate_capabilities(cls, value: list[str]) -> list[str]:
+        return _unique_text(value, 'required executor capabilities')
+
+    @model_validator(mode='after')
+    def reject_sensitive_values(self):
+        reject_credentials(self, 'executor selection request')
+        return self
+
+
+class ProjectAgentExecutorSelection(StrictModel):
+    selection_id: str
+    task_id: str
+    personal_space_id: str
+    personal_project_id: str
+    agent_id: str
+    status: ProjectAgentExecutorSelectionStatus
+    outcome: ProjectAgentExecutorSelectionStatus
+    selected_executor_id: str | None = None
+    # ``executor_id`` is a compatibility alias consumed by the task router.
+    executor_id: str | None = None
+    selected_provider: str | None = None
+    selected_model: str | None = None
+    matched_rule_id: str | None = None
+    rule_id: str | None = None
+    matched_rule_scope: ProjectAgentExecutorScope | None = None
+    candidate_executor_ids: list[str] = Field(default_factory=list)
+    selection_reason: str
+    reason: str
+    fallback_outcome: ProjectAgentExecutorFallbackOutcome = 'not_needed'
+    fallback_from_executor_id: str | None = None
+    fallback_reason: str | None = None
+    blocked_reason: str | None = None
+    model_strategy: ProjectAgentModelStrategy
+    model_strategy_key: str = Field(min_length=64, max_length=64)
+    model_strategy_source: Literal[
+        'task',
+        'assignment',
+        'agent',
+        'routing_rule',
+        'coordinator',
+    ] = 'agent'
+    executor_policy: ProjectAgentExecutorPolicy = Field(
+        default_factory=ProjectAgentExecutorPolicy
+    )
+    idempotency_key: str | None = Field(default=None, min_length=8, max_length=256)
+    payload_fingerprint: str | None = Field(default=None, min_length=16, max_length=128)
+    created_at: datetime
+
+    @model_validator(mode='after')
+    def validate_strategy_key(self):
+        if self.model_strategy_key != project_agent_model_strategy_key(
+            self.model_strategy
+        ):
+            raise ValueError('model strategy key does not match model strategy')
+        return self
+
+
+class ProjectAgentExecutorActualReport(StrictModel):
+    personal_space_id: str = Field(min_length=1, max_length=128)
+    personal_project_id: str = Field(min_length=1, max_length=128)
+    task_id: str = Field(min_length=1, max_length=128)
+    run_id: str = Field(min_length=1, max_length=128)
+    agent_id: str = Field(min_length=1, max_length=128)
+    executor_id: str = Field(min_length=1, max_length=128)
+    provider: str = Field(min_length=1, max_length=128)
+    model: str = Field(min_length=1, max_length=256)
+    model_strategy: ProjectAgentModelStrategy = Field(
+        default_factory=ProjectAgentModelStrategy
+    )
+    model_strategy_source: Literal[
+        'task',
+        'assignment',
+        'agent',
+        'routing_rule',
+        'coordinator',
+    ] = 'agent'
+    matched_rule_id: str | None = Field(default=None, min_length=1, max_length=128)
+    fallback_reason: str | None = Field(default=None, min_length=1, max_length=2048)
+    idempotency_key: str = Field(min_length=8, max_length=256)
+    occurred_at: datetime
+    source_application: SourceApplication | None = None
+    source_session_id: str | None = Field(default=None, min_length=1, max_length=256)
+
+    @model_validator(mode='after')
+    def reject_sensitive_values(self):
+        reject_credentials(self, 'actual executor report')
+        return self
+
+
+class ProjectAgentExecutorOutcomeEvidenceCreate(StrictModel):
+    personal_space_id: str = Field(min_length=1, max_length=128)
+    personal_project_id: str = Field(min_length=1, max_length=128)
+    work_kind: str = Field(min_length=1, max_length=128)
+    agent_id: str = Field(min_length=1, max_length=128)
+    executor_id: str = Field(min_length=1, max_length=128)
+    task_id: str = Field(min_length=1, max_length=128)
+    run_id: str | None = Field(default=None, min_length=1, max_length=128)
+    model_strategy: ProjectAgentModelStrategy = Field(
+        default_factory=ProjectAgentModelStrategy
+    )
+    model_strategy_key: str | None = Field(
+        default=None,
+        min_length=64,
+        max_length=64,
+    )
+    evidence_kind: ProjectAgentExecutorEvidenceKind
+    source: ProjectAgentExecutorEvidenceSource
+    terminal_outcome: ProjectAgentExecutorTerminalOutcome | None = None
+    rating: int | None = Field(default=None, ge=1, le=5)
+    reference_ids: list[str] = Field(default_factory=list, max_length=16)
+    note: str | None = Field(default=None, min_length=1, max_length=2048)
+    idempotency_key: str = Field(min_length=8, max_length=256)
+    occurred_at: datetime
+
+    @field_validator('reference_ids')
+    @classmethod
+    def validate_reference_ids(cls, value: list[str]) -> list[str]:
+        return _unique_ids(value, 'evidence reference IDs')
+
+    @model_validator(mode='after')
+    def validate_evidence(self):
+        expected_strategy_key = project_agent_model_strategy_key(self.model_strategy)
+        if self.model_strategy_key and self.model_strategy_key != expected_strategy_key:
+            raise ValueError('model strategy key does not match model strategy')
+        system_kinds = {'terminal_outcome'}
+        test_kinds = {'test_passed', 'test_failed'}
+        explicit_kinds = {
+            'rework_requested',
+            'repeated_negative_feedback',
+            'explicit_praise',
+            'acceptance_passed',
+            'acceptance_failed',
+            'explicit_rating',
+        }
+        if self.evidence_kind in system_kinds and self.source != 'system_terminal':
+            raise ValueError('terminal outcome evidence must come from the system')
+        if self.evidence_kind in test_kinds and self.source != 'test_fact':
+            raise ValueError('test evidence must be recorded as a test fact')
+        if self.evidence_kind in explicit_kinds and self.source != 'user_explicit':
+            raise ValueError('feedback evidence must be explicitly supplied by a user')
+        if self.evidence_kind == 'terminal_outcome' and not self.terminal_outcome:
+            raise ValueError('terminal outcome evidence requires a terminal outcome')
+        if self.evidence_kind in system_kinds | test_kinds and not self.reference_ids:
+            raise ValueError('system or test evidence requires an evidence reference')
+        if self.evidence_kind != 'terminal_outcome' and self.terminal_outcome:
+            raise ValueError('terminal outcome is only valid for terminal evidence')
+        if self.evidence_kind == 'explicit_rating' and self.rating is None:
+            raise ValueError('explicit rating evidence requires a rating')
+        if self.evidence_kind != 'explicit_rating' and self.rating is not None:
+            raise ValueError('rating is only valid for explicit rating evidence')
+        if self.evidence_kind in explicit_kinds and not self.reference_ids and not self.note:
+            raise ValueError('explicit feedback requires a reference or note')
+        reject_credentials(self, 'executor outcome evidence')
+        return self
+
+
+class ProjectAgentExecutorOutcomeEvidenceRecord(
+    ProjectAgentExecutorOutcomeEvidenceCreate
+):
+    # Stored evidence always exposes the canonical bucket key.  The create
+    # contract keeps this optional for backwards-compatible callers; the store
+    # computes it before returning a durable record.
+    model_strategy_key: str = Field(min_length=64, max_length=64)
+    evidence_id: str
+    ignored: bool = False
+    ignored_reason: str | None = None
+    created_at: datetime
+
+
+class ProjectAgentExecutorEvidenceContribution(StrictModel):
+    evidence_id: str
+    evidence_kind: ProjectAgentExecutorEvidenceKind
+    signal: Literal['success', 'failure', 'rework', 'rating', 'neutral']
+    value: float
+    decay_weight: float = Field(ge=0, le=1)
+    occurred_at: datetime
+    reference_ids: list[str] = Field(default_factory=list)
+
+
+class ProjectAgentExecutorOutcomeAggregate(StrictModel):
+    personal_space_id: str
+    personal_project_id: str
+    work_kind: str
+    agent_id: str
+    executor_id: str
+    model_strategy: ProjectAgentModelStrategy
+    model_strategy_key: str = Field(min_length=64, max_length=64)
+    sample_count: int = Field(ge=0)
+    recent_count: int = Field(ge=0)
+    success_count: int = Field(ge=0)
+    rework_count: int = Field(ge=0)
+    failure_count: int = Field(ge=0)
+    rating_count: int = Field(ge=0)
+    average_rating: float | None = Field(default=None, ge=1, le=5)
+    neutral_due_to_insufficient_evidence: bool = False
+    ignored: bool = False
+    reset_at: datetime | None = None
+    as_of: datetime
+    decay_half_life_days: float = Field(gt=0)
+    weighted_success: float = Field(ge=0)
+    weighted_failure: float = Field(ge=0)
+    evidence_refs: list[str] = Field(default_factory=list)
+    evidence_contributions: list[ProjectAgentExecutorEvidenceContribution] = Field(
+        default_factory=list
+    )
+
+    @model_validator(mode='after')
+    def validate_strategy_key(self):
+        if self.model_strategy_key != project_agent_model_strategy_key(
+            self.model_strategy
+        ):
+            raise ValueError('model strategy key does not match model strategy')
+        return self
+
+
+class ProjectAgentExecutorEvidenceIgnore(StrictModel):
+    personal_space_id: str = Field(min_length=1, max_length=128)
+    personal_project_id: str = Field(min_length=1, max_length=128)
+    agent_id: str = Field(min_length=1, max_length=128)
+    evidence_id: str = Field(min_length=1, max_length=128)
+    reason: str = Field(min_length=1, max_length=2048)
+    idempotency_key: str = Field(min_length=8, max_length=256)
+
+    @model_validator(mode='after')
+    def reject_sensitive_values(self):
+        reject_credentials(self, 'executor evidence ignore')
+        return self
+
+
+class ProjectAgentExecutorOutcomeReset(StrictModel):
+    personal_space_id: str = Field(min_length=1, max_length=128)
+    personal_project_id: str = Field(min_length=1, max_length=128)
+    work_kind: str = Field(min_length=1, max_length=128)
+    agent_id: str = Field(min_length=1, max_length=128)
+    executor_id: str = Field(min_length=1, max_length=128)
+    model_strategy: ProjectAgentModelStrategy = Field(
+        default_factory=ProjectAgentModelStrategy
+    )
+    model_strategy_key: str | None = Field(
+        default=None,
+        min_length=64,
+        max_length=64,
+    )
+    reason: str = Field(min_length=1, max_length=2048)
+    idempotency_key: str = Field(min_length=8, max_length=256)
+    reset_at: datetime
+
+    @model_validator(mode='after')
+    def reject_sensitive_values(self):
+        expected_strategy_key = project_agent_model_strategy_key(self.model_strategy)
+        if self.model_strategy_key and self.model_strategy_key != expected_strategy_key:
+            raise ValueError('model strategy key does not match model strategy')
+        reject_credentials(self, 'executor evidence reset')
+        return self
+
+
+# Names used by the task store and by older integration code.
+ProjectAgentExecutorDecision = ProjectAgentExecutorSelection
+ProjectAgentExecutorRouteResult = ProjectAgentExecutorSelection