evolvingmachines-evolve 0.0.55.dev1355__py3-none-any.whl

evolve/__init__.py

@@ -0,0 +1,819 @@
+"""Evolve Python SDK - Pythonic wrapper around the TypeScript Evolve SDK."""
+
+from .agent import Evolve
+from .config import (
+    AgentConfig,
+    E2BProvider,
+    ManagedProvider,
+    DaytonaProvider,
+    ModalProvider,
+    SandboxCreateOptions,
+    SandboxNetworkPolicy,
+    SandboxProvider,
+    AgentType,
+    WorkspaceMode,
+    BrowserProvider,
+    BrowserConfig,
+    BrowserCredentialScopeEntry,
+    BrowserCredentialsConfig,
+    BrowserCredentialsClientConfig,
+    BrowserProfilesClientConfig,
+    ManagedSecretRef,
+    ManagedSecretsClientConfig,
+    AgentPluginConfig,
+    ReasoningEffort,
+    AgentPreset,
+    ValidationMode,
+    SchemaOptions,
+    IntegrationsConfig,
+    IntegrationsSetup,
+    IntegrationToolsFilter,
+    StorageConfig,
+    StorageCredentials,
+    SessionsConfig,
+    HostedClientConfig,
+)
+from .hosted import (
+    EFFORT_SUPPORT_VALUES,
+    ActiveDataset,
+    Agent,
+    AnalysesClient,
+    AnalysisPage,
+    AnalysisLabel,
+    AnalysisStatus,
+    Check,
+    CheckConfigInput,
+    CheckPage,
+    CheckSource,
+    CheckLabel,
+    CheckStatus,
+    ChecksClient,
+    TaskCheck,
+    JobListScope,
+    AgentArm,
+    AgentCapability,
+    AgentDatasetStats,
+    AgentInfo,
+    AgentModelOption,
+    AgentPage,
+    AgentResult,
+    AgentsClient,
+    AnalysisCheck,
+    AnalysisEvidence,
+    AnalysisFailure,
+    AnalyzeConfig,
+    AnalyzeConfigInput,
+    ApiKey,
+    AttemptPhase,
+    AuthClient,
+    AuthStatus,
+    Organization,
+    OrganizationDetail,
+    OrgQuota,
+    OrgRole,
+    OrgUsage,
+    OrgsClient,
+    CapabilityDocument,
+    CompareCell,
+    CompareCoverage,
+    CompareJobAggregate,
+    CompareResponse,
+    CompareTaskRow,
+    Dataset,
+    DatasetFailedTask,
+    DatasetImport,
+    DatasetImportFailure,
+    DatasetImportProgress,
+    DatasetManifestAuthor,
+    DatasetManifestMetadata,
+    DatasetImportPage,
+    JobImport,
+    JobImportFailure,
+    JobImportPage,
+    JobImportPhaseProgress,
+    JobImportProgress,
+    JobImportSkippedTrial,
+    JobImportSource,
+    JobTaskLink,
+    TrialTaskLink,
+    TASK_LINKED_BY,
+    TASK_LINK_REASONS,
+    DatasetPage,
+    DatasetPreflight,
+    PreflightDeferredCheck,
+    PreflightManifestVerdict,
+    PreflightTaskVerdict,
+    TaskNote,
+    DatasetRef,
+    DatasetSelector,
+    DatasetVersion,
+    DatasetVersionArchiveSource,
+    DatasetVersionArchiveUrlSource,
+    DatasetVersionGitSource,
+    DatasetVersionHubSource,
+    DatasetVersionSource,
+    DatasetsClient,
+    EvalSandboxProvider,
+    EvolveAPIError,
+    EvolveDigestMismatchError,
+    EvolveIncompleteDownloadError,
+    ExceptionInfo,
+    HostedEvolve,
+    HOSTED_ERROR_CODES,
+    HostedErrorCode,
+    ImportSettleError,
+    ImportTaskFailure,
+    ImportCodeBuildMeter,
+    ImportImageCounts,
+    ImportPhaseProgress,
+    ImportWarning,
+    Job,
+    JobAnalysisStats,
+    JobBuildExclusion,
+    JobCounts,
+    JobDeleteResult,
+    JobEvent,
+    JobEventType,
+    InfraFailureSignature,
+    JobFailure,
+    JobPage,
+    JobRetryConfig,
+    JobRetryConfigInput,
+    JobSecretRef,
+    JobSecretInline,
+    JobStats,
+    JobStatus,
+    JobTaskRollup,
+    JobTaskRollupPage,
+    JobsClient,
+    JudgeResult,
+    ModelInfo,
+    NoActiveVersionError,
+    ManagedProviderCapability,
+    PassAtKGroup,
+    PassAtKPoint,
+    ProviderCapability,
+    ReportedAgentResult,
+    ReportedTotals,
+    SkillLock,
+    SkillUpload,
+    SkillUploadPage,
+    SkillsClient,
+    SourceJob,
+    SpendSource,
+    StatusVocabulary,
+    StopResponse,
+    Task,
+    TaskBuild,
+    TaskBuildFailure,
+    TaskPage,
+    TaskProviderVerdict,
+    TimingInfo,
+    TraceEvent,
+    TraceEventPage,
+    GATEWAY_TRACE_SEQ_BASE,
+    gateway_usage_of,
+    JobGrepGroup,
+    JobGrepPage,
+    Rubric,
+    RubricCriterion,
+    Trial,
+    TrialAnalysis,
+    TrialFile,
+    TrialFilePage,
+    TrialPage,
+    TrialRetry,
+    TrialStatus,
+    TrialTally,
+    TrialUploadProvenance,
+    TrialsClient,
+    UploadProvenance,
+    UpstreamStatus,
+    VerifierEnvironmentMode,
+    VerifierResult,
+    is_hosted_error_code,
+    meta,
+    pass_at_k,
+)
+from .results import (
+    AgentResponse,
+    CheckpointInfo,
+    ExecuteResult,
+    OutputResult,
+    RunCost,
+    SessionCost,
+    SessionStatus,
+    SessionInfo,
+    SessionPage,
+    SessionEvent,
+    SessionTranscript,
+    BrowserReplay,
+    UsageReading,
+)
+from .storage_client import StorageClient
+from .sessions_client import SessionsClient
+from .browser_credentials import BrowserCredentialsClient, BrowserCredentialMetadata, BrowserCredentialsPage
+from .browser_profiles import BrowserProfilesClient, BrowserProfileMetadata, BrowserProfilesPage
+from .managed_secrets import (
+    ManagedSecretsClient,
+    ManagedSecretMetadata,
+    ManagedSecretWriteResult,
+    ManagedSecretDeleteResult,
+)
+from .utils import read_local_dir, save_local_dir
+from .bridge import (
+    SandboxNotFoundError,
+    BridgeConnectionError,
+    BridgeBuildError,
+)
+from .retry import RetryConfig, OnItemRetryCallback, execute_with_retry
+from .swarm import (
+    Swarm,
+    SwarmConfig,
+    BestOfConfig,
+    VerifyConfig,
+    SwarmResult,
+    SwarmResultList,
+    ReduceResult,
+    BestOfResult,
+    BestOfInfo,
+    VerifyInfo,
+    IndexedMeta,
+    ReduceMeta,
+    JudgeMeta,
+    VerifyMeta,
+    VerifyDecision,
+    is_swarm_result,
+    # Callback types
+    OnCandidateCompleteCallback,
+    OnJudgeCompleteCallback,
+    OnWorkerCompleteCallback,
+    OnVerifierCompleteCallback,
+)
+from .pipeline import (
+    Pipeline,
+    TerminalPipeline,
+    MapConfig,
+    FilterConfig,
+    ReduceConfig,
+    StepResult,
+    PipelineResult,
+    PipelineEvents,
+    StepStartEvent,
+    StepCompleteEvent,
+    StepErrorEvent,
+    ItemRetryEvent,
+    WorkerCompleteEvent,
+    VerifierCompleteEvent,
+    CandidateCompleteEvent,
+    JudgeCompleteEvent,
+    EmitOption,
+)
+
+from typing import List, Optional
+
+
+def storage(config: Optional[StorageConfig] = None) -> StorageClient:
+    """Create a standalone storage client for checkpoint browsing and download.
+
+    Returns a ``StorageClient`` that manages its own bridge subprocess.
+    Use as an async context manager to ensure cleanup.
+
+    Args:
+        config: Storage configuration (BYOK S3 or None for gateway mode)
+
+    Returns:
+        StorageClient with list_checkpoints, get_checkpoint,
+        download_checkpoint, download_files methods
+
+    Example:
+        >>> from evolve import storage, StorageConfig
+        >>>
+        >>> # BYOK mode
+        >>> async with storage(StorageConfig(url='s3://my-bucket/')) as store:
+        ...     checkpoints = await store.list_checkpoints(limit=5)
+        ...     files = await store.download_files(checkpoints[0].id)
+        >>>
+        >>> # Gateway mode (uses EVOLVE_API_KEY)
+        >>> async with storage() as store:
+        ...     checkpoints = await store.list_checkpoints()
+    """
+    from .bridge import BridgeManager
+    bridge = BridgeManager()
+    return StorageClient(bridge, config or StorageConfig(), _owns_bridge=True)
+
+
+def sessions(config: Optional[SessionsConfig] = None) -> SessionsClient:
+    """Create a standalone sessions client for historical traces and past sessions.
+
+    Returns a ``SessionsClient`` that manages its own bridge subprocess.
+    Use as an async context manager to ensure cleanup.
+
+    Gateway-only: requires ``EVOLVE_API_KEY`` unless ``SessionsConfig(api_key=...)``
+    is provided.
+
+    Args:
+        config: Optional API key / dashboard URL overrides
+
+    Returns:
+        SessionsClient with list, get, events, download, browser_replay methods
+
+    Example:
+        >>> from evolve import sessions
+        >>>
+        >>> async with sessions() as client:
+        ...     page = await client.list(limit=10, state='ended')
+        ...     trace = await client.get(page.items[0].id)
+        ...     await client.download(trace.id, to='./traces')
+        ...     await client.browser_replay(trace.id)
+    """
+    from .bridge import BridgeManager
+    bridge = BridgeManager()
+    return SessionsClient(bridge, config or SessionsConfig(), _owns_bridge=True)
+
+
+def browser_credentials(config: Optional[BrowserCredentialsClientConfig] = None) -> BrowserCredentialsClient:
+    """Create a standalone browser credentials client.
+
+    Uses EVOLVE_API_KEY unless BrowserCredentialsClientConfig(api_key=...) is provided.
+    Passwords are encrypted locally before they are sent to the dashboard.
+    """
+    return BrowserCredentialsClient(config or BrowserCredentialsClientConfig())
+
+
+def browser_profiles(config: Optional[BrowserProfilesClientConfig] = None) -> BrowserProfilesClient:
+    """Create a standalone browser profiles client.
+
+    Uses EVOLVE_API_KEY unless BrowserProfilesClientConfig(api_key=...) is provided.
+    """
+    return BrowserProfilesClient(config or BrowserProfilesClientConfig())
+
+
+def datasets(config: Optional[HostedClientConfig] = None) -> DatasetsClient:
+    """Create a standalone hosted-evals datasets client (shared catalog).
+
+    Uses EVOLVE_API_KEY unless HostedClientConfig(api_key=...) is provided.
+    """
+    return DatasetsClient(config)
+
+
+def agents(config: Optional[HostedClientConfig] = None) -> AgentsClient:
+    """Create a standalone hosted-evals registered-agents client.
+
+    Register a private agent once, then name it in job agents[].name
+    exactly like a built-in. Uses EVOLVE_API_KEY unless
+    HostedClientConfig(api_key=...) is provided.
+    """
+    return AgentsClient(config)
+
+
+def hosted(config: Optional[HostedClientConfig] = None) -> HostedEvolve:
+    """Open the hosted surface with ONE configuration.
+
+    Named ``hosted()`` rather than ``evolve()`` deliberately: ``Evolve`` is
+    already the local-sandbox class in this same package, and two exports one
+    shift key apart that do completely different things is a trap. ``hosted()``
+    says which half of the SDK you are reaching for.
+
+    The four clients underneath are built lazily, so ``await hosted().meta()``
+    works with no API key configured::
+
+        from evolve import hosted
+
+        client = hosted()
+        doc = await client.meta()            # public, no key needed
+        job = await client.jobs.start(...)   # needs EVOLVE_API_KEY
+    """
+    return HostedEvolve(config)
+
+
+def jobs(config: Optional[HostedClientConfig] = None) -> JobsClient:
+    """Create a standalone hosted-evals jobs client.
+
+    Uses EVOLVE_API_KEY unless HostedClientConfig(api_key=...) is provided.
+    watch() consumes the job's server-sent event stream until the job is
+    terminal, mirroring the TypeScript SDK.
+    """
+    return JobsClient(config)
+
+
+def trials(config: Optional[HostedClientConfig] = None) -> TrialsClient:
+    """Create a standalone hosted-evals trials client.
+
+    A trial id is globally addressable — no method takes a job id. Uses
+    EVOLVE_API_KEY unless HostedClientConfig(api_key=...) is provided.
+    """
+    return TrialsClient(config)
+
+
+def analyses(config: Optional[HostedClientConfig] = None) -> AnalysesClient:
+    """Create a standalone hosted-evals analyses client.
+
+    The catalog of trace-analysis runs (``list()``), each row carrying the
+    trial, job, and task it judged. Uses EVOLVE_API_KEY unless
+    HostedClientConfig(api_key=...) is provided.
+    """
+    return AnalysesClient(config)
+
+
+def checks(config: Optional[HostedClientConfig] = None) -> ChecksClient:
+    """Create a standalone hosted-evals checks client.
+
+    Task quality checks — Harbor's ``harbor check <PATH>``, hosted: upload a
+    task directory (or a directory of them), read the report, list your
+    checks. Uses EVOLVE_API_KEY unless HostedClientConfig(api_key=...) is
+    provided.
+    """
+    return ChecksClient(config)
+
+
+def skills(config: Optional[HostedClientConfig] = None) -> SkillsClient:
+    """Create a standalone hosted-evals skills client (platform uploads).
+
+    An uploaded skill is an immutable folder referenced as ``upload:<id>`` in
+    job ``agents[].skills``, next to skills.sh and git references. Uses
+    EVOLVE_API_KEY unless HostedClientConfig(api_key=...) is provided.
+    """
+    return SkillsClient(config)
+
+
+def auth(config: Optional[HostedClientConfig] = None) -> AuthClient:
+    """Create a standalone hosted-evals auth client (caller identity).
+
+    Wave-gated: until the server's wave lands, status() reports the route's
+    not-found as the API error it is. Uses EVOLVE_API_KEY unless
+    HostedClientConfig(api_key=...) is provided.
+    """
+    return AuthClient(config)
+
+
+def orgs(config: Optional[HostedClientConfig] = None) -> OrgsClient:
+    """Create a standalone hosted-evals organizations client (the read pair).
+
+    ``list()`` — every organization you belong to; ``get(slug)`` — one
+    organization's role, member count, quota and usage. Uses EVOLVE_API_KEY
+    unless HostedClientConfig(api_key=...) is provided.
+    """
+    return OrgsClient(config)
+
+
+def managed_secrets(config: Optional[ManagedSecretsClientConfig] = None) -> ManagedSecretsClient:
+    """Create a standalone managed secrets client.
+
+    Uses EVOLVE_API_KEY unless ManagedSecretsClientConfig(api_key=...) is provided.
+    Values are never returned.
+    """
+    return ManagedSecretsClient(config or ManagedSecretsClientConfig())
+
+
+async def list_checkpoints(
+    storage: Optional[StorageConfig] = None,
+    limit: Optional[int] = None,
+    tag: Optional[str] = None,
+) -> List[CheckpointInfo]:
+    """List checkpoints without creating a full Evolve instance.
+
+    Uses the lightweight :class:`StorageClient` path instead of a full
+    Evolve initialization (no agent/sandbox setup needed).
+
+    Args:
+        storage: Storage configuration (BYOK S3 or None for gateway mode)
+        limit: Maximum number of checkpoints to return
+        tag: Filter by session tag
+
+    Returns:
+        List of CheckpointInfo sorted by newest first
+
+    Example:
+        >>> from evolve import list_checkpoints, StorageConfig
+        >>> checkpoints = await list_checkpoints(
+        ...     storage=StorageConfig(url='s3://my-bucket/prefix/'),
+        ...     limit=10,
+        ... )
+    """
+    from .bridge import BridgeManager
+    bridge = BridgeManager()
+    store = StorageClient(bridge, storage or StorageConfig(), _owns_bridge=True)
+    try:
+        return await store.list_checkpoints(limit=limit, tag=tag)
+    finally:
+        await store.close()
+
+
+__version__ = '0.0.55.dev1355'
+
+__all__ = [
+    # Main classes
+    'Evolve',
+    'Swarm',
+    'Pipeline',
+    'TerminalPipeline',
+
+    # Evolve Configuration
+    'AgentConfig',
+    'E2BProvider',
+    'ManagedProvider',
+    'DaytonaProvider',
+    'ModalProvider',
+    'SandboxProvider',
+    'AgentType',
+    'WorkspaceMode',
+    'BrowserProvider',
+    'BrowserConfig',
+    'BrowserCredentialScopeEntry',
+    'BrowserCredentialsConfig',
+    'BrowserCredentialsClientConfig',
+    'BrowserProfilesClientConfig',
+    'ManagedSecretRef',
+    'ManagedSecretsClientConfig',
+    'AgentPluginConfig',
+    'ReasoningEffort',
+    'AgentPreset',
+    'ValidationMode',
+    'SchemaOptions',
+    'IntegrationsConfig',
+    'IntegrationsSetup',
+    'IntegrationToolsFilter',
+    'StorageConfig',
+    'StorageCredentials',
+    'SessionsConfig',
+    'BrowserCredentialsClient',
+    'BrowserCredentialMetadata',
+    'BrowserCredentialsPage',
+    'BrowserProfilesClient',
+    'BrowserProfileMetadata',
+    'BrowserProfilesPage',
+    'ManagedSecretsClient',
+    'ManagedSecretMetadata',
+    'ManagedSecretWriteResult',
+    'ManagedSecretDeleteResult',
+    'SandboxCreateOptions',
+    'SandboxNetworkPolicy',
+
+    # Evolve Results
+    'AgentResponse',
+    'CheckpointInfo',
+    'ExecuteResult',  # Backward compatibility alias for AgentResponse
+    'OutputResult',
+    'RunCost',
+    'SessionCost',
+    'SessionStatus',
+    'SessionInfo',
+    'SessionPage',
+    'SessionEvent',
+    'SessionTranscript',
+    'BrowserReplay',
+    'UsageReading',
+
+    # Standalone clients
+    'StorageClient',
+    'SessionsClient',
+    'browser_credentials',
+    'browser_profiles',
+    'managed_secrets',
+
+    # Hosted evals (datasets + agents + jobs + trials + auth)
+    'HostedClientConfig',
+    'DatasetsClient',
+    'AgentsClient',
+    'JobsClient',
+    'TrialsClient',
+    'AnalysesClient',
+    'ChecksClient',
+    'AuthClient',
+    'OrgsClient',
+    'EvolveAPIError',
+    'EvolveDigestMismatchError',
+    'EvolveIncompleteDownloadError',
+    'NoActiveVersionError',
+    'ImportSettleError',
+    'Dataset',
+    'ActiveDataset',
+    'DatasetManifestAuthor',
+    'DatasetManifestMetadata',
+    'DatasetVersion',
+    'DatasetVersionSource',
+    'DatasetVersionArchiveSource',
+    'DatasetVersionArchiveUrlSource',
+    'DatasetVersionGitSource',
+    'DatasetVersionHubSource',
+    'DatasetRef',
+    'DatasetSelector',
+    'DatasetFailedTask',
+    'TaskBuild',
+    'TaskBuildFailure',
+    'DatasetImport',
+    'DatasetImportFailure',
+    'DatasetImportProgress',
+    'ImportTaskFailure',
+    'ImportCodeBuildMeter',
+    'ImportImageCounts',
+    'ImportPhaseProgress',
+    'ImportWarning',
+    'Task',
+    'TaskProviderVerdict',
+    'DatasetPreflight',
+    'PreflightDeferredCheck',
+    'PreflightManifestVerdict',
+    'PreflightTaskVerdict',
+    'TaskNote',
+    'AgentArm',
+    'AgentDatasetStats',
+    'Job',
+    'JobAnalysisStats',
+    'JobBuildExclusion',
+    'JobCounts',
+    'JobDeleteResult',
+    'JobFailure',
+    'JobRetryConfig',
+    'JobRetryConfigInput',
+    'JobSecretRef',
+    'JobSecretInline',
+    'JobStats',
+    'JobStatus',
+    'AnalyzeConfig',
+    'AnalyzeConfigInput',
+    'AnalysisCheck',
+    'AnalysisEvidence',
+    'AnalysisFailure',
+    'Rubric',
+    'RubricCriterion',
+    'TrialAnalysis',
+    'CheckConfigInput',
+    'CheckSource',
+    'TaskCheck',
+    'Check',
+    'TrialStatus',
+    'EvalSandboxProvider',
+    'SpendSource',
+    'VerifierEnvironmentMode',
+    'AttemptPhase',
+    'JobTaskRollup',
+    'PassAtKGroup',
+    'PassAtKPoint',
+    'pass_at_k',
+    'SourceJob',
+    'SkillLock',
+    'SkillUpload',
+    'SkillUploadPage',
+    'SkillsClient',
+    'TrialTally',
+    'JobEvent',
+    'JobEventType',
+    'InfraFailureSignature',
+    'CompareResponse',
+    'CompareJobAggregate',
+    'CompareCell',
+    'CompareCoverage',
+    'CompareTaskRow',
+    'Trial',
+    'TrialRetry',
+    'TimingInfo',
+    'ModelInfo',
+    'AgentInfo',
+    'AgentResult',
+    'JudgeResult',
+    'VerifierResult',
+    'ExceptionInfo',
+    'StopResponse',
+    'TraceEvent',
+    'TraceEventPage',
+    'GATEWAY_TRACE_SEQ_BASE',
+    'gateway_usage_of',
+    'JobGrepGroup',
+    'JobGrepPage',
+    'TrialFile',
+    'TrialFilePage',
+    'JobPage',
+    'TrialPage',
+    'AnalysisPage',
+    'CheckPage',
+    'JobListScope',
+    'AnalysisStatus',
+    'AnalysisLabel',
+    'CheckStatus',
+    'CheckLabel',
+    'JobTaskRollupPage',
+    'DatasetPage',
+    'AgentPage',
+    'TaskPage',
+    'Agent',
+    'DatasetImportPage',
+    'JobImport',
+    'JobImportFailure',
+    'JobImportPage',
+    'JobImportPhaseProgress',
+    'JobImportProgress',
+    'JobImportSkippedTrial',
+    'JobImportSource',
+    'JobTaskLink',
+    'TrialTaskLink',
+    'TASK_LINKED_BY',
+    'TASK_LINK_REASONS',
+    'ApiKey',
+    'AuthStatus',
+    'Organization',
+    'OrganizationDetail',
+    'OrgQuota',
+    'OrgRole',
+    'OrgUsage',
+    'datasets',
+    'agents',
+    'jobs',
+    'trials',
+    'analyses',
+    'checks',
+    'skills',
+    'orgs',
+    'auth',
+    'hosted',
+    'meta',
+    'HostedEvolve',
+    'CapabilityDocument',
+    'AgentCapability',
+    'AgentModelOption',
+    'EFFORT_SUPPORT_VALUES',
+    'ManagedProviderCapability',
+    'ProviderCapability',
+    'StatusVocabulary',
+    'ReportedAgentResult',
+    'ReportedTotals',
+    'TrialUploadProvenance',
+    'UploadProvenance',
+    'UpstreamStatus',
+    'HOSTED_ERROR_CODES',
+    'HostedErrorCode',
+    'is_hosted_error_code',
+
+    # Standalone functions
+    'storage',
+    'sessions',
+    'list_checkpoints',
+
+    # Swarm Configuration
+    'SwarmConfig',
+    'BestOfConfig',
+    'VerifyConfig',
+
+    # Swarm Results
+    'SwarmResult',
+    'SwarmResultList',
+    'ReduceResult',
+    'BestOfResult',
+    'BestOfInfo',
+    'VerifyInfo',
+    'VerifyDecision',
+
+    # Swarm Metadata
+    'IndexedMeta',
+    'ReduceMeta',
+    'JudgeMeta',
+    'VerifyMeta',
+
+    # Swarm Helpers
+    'is_swarm_result',
+
+    # Swarm Callback types
+    'OnCandidateCompleteCallback',
+    'OnJudgeCompleteCallback',
+    'OnWorkerCompleteCallback',
+    'OnVerifierCompleteCallback',
+
+    # Pipeline Configuration
+    'MapConfig',
+    'FilterConfig',
+    'ReduceConfig',
+
+    # Pipeline Results
+    'StepResult',
+    'PipelineResult',
+
+    # Pipeline Events
+    'PipelineEvents',
+    'StepStartEvent',
+    'StepCompleteEvent',
+    'StepErrorEvent',
+    'ItemRetryEvent',
+    'WorkerCompleteEvent',
+    'VerifierCompleteEvent',
+    'CandidateCompleteEvent',
+    'JudgeCompleteEvent',
+    'EmitOption',
+
+    # Retry — the SWARM client-side retry helper. The hosted job auto-retry
+    # policy is a different shape with its own names: JobRetryConfig /
+    # JobRetryConfigInput above (the spec calls that pair RetryConfig /
+    # RetryConfigInput; the Python export is prefixed so the two never shadow).
+    'RetryConfig',
+    'OnItemRetryCallback',
+    'execute_with_retry',
+
+    # Utilities
+    'read_local_dir',
+    'save_local_dir',
+
+    # Exceptions
+    'SandboxNotFoundError',
+    'BridgeConnectionError',
+    'BridgeBuildError',
+]