Skip to main content Crate deepstrike_sdk Copy item path Source pub use governance::Governance ;pub use governance::GovernanceVerdict ;pub use harness::CriterionResult ;pub use harness::Harness ;pub use harness::HarnessEvent ;pub use harness::HarnessOutcome ;pub use harness::HarnessRequest ;pub use harness::QualityGate ;pub use harness_loop::EvalLoopHarness ;Deprecated pub use harness_loop::HarnessLoop ;pub use harness_loop::SinglePassHarness ;pub use harness_loop::VerdictCtx ;pub use harness_loop::VerdictFn ;pub use knowledge::KnowledgeSource ;pub use memory::DreamResult ;pub use memory::DreamStore ;pub use memory::InMemoryDreamStore ;pub use memory::WorkingMemory ;pub use providers::RuntimePolicy ;pub use providers::anthropic::AnthropicProvider ;pub use providers::openai::OpenAIProvider ;pub use providers::openai::deepseek ;pub use providers::openai::kimi ;pub use providers::openai::minimax ;pub use providers::openai::ollama ;pub use providers::openai::qwen ;pub use providers::LLMProvider ;pub use providers::ProviderRunState ;pub use providers::ProviderToolSpec ;pub use providers::StreamEvent ;pub use providers::TokenUsage ;pub use run_event::RunEvent ;pub use runtime::ChainedCredentialVault ;pub use runtime::CredentialVault ;pub use runtime::EnvCredentialVault ;pub use runtime::InMemoryCredentialVault ;pub use runtime::ExecutionPlane ;pub use runtime::LocalExecutionPlane ;pub use runtime::FileSessionLog ;pub use runtime::InMemorySessionLog ;pub use runtime::SessionEntry ;pub use runtime::SessionLog ;pub use runtime::replay_provider::ReplayProvider ;pub use runtime::replay_provider::ReplayProviderOpts ;pub use runtime::eval::judge ;pub use runtime::eval::build_eval_messages ;pub use runtime::eval::parse_verdict ;pub use runtime::eval::verdict_output_schema ;pub use runtime::eval::Verdict ;pub use runtime::McpProxyPlane ;pub use runtime::McpServerConfig ;pub use runtime::AttentionPolicy ;pub use runtime::GovernancePolicy ;pub use runtime::MemoryWriteRateLimit ;pub use runtime::NativeOsProfile ;pub use runtime::OsProfile ;pub use runtime::SchedulerBudget ;pub use runtime::assert_native_profile ;pub use runtime::default_native_governance_policy ;pub use runtime::os_profile ;pub use runtime::os_profile ;pub use runtime::DEFAULT_NATIVE_ATTENTION_POLICY ;pub use runtime::MilestoneEvaluationContext ;pub use runtime::MilestoneEvaluationHandler ;pub use runtime::MilestonePolicy ;pub use runtime::RuntimeOptions ;pub use runtime::RuntimeRunner ;pub use runtime::collect_text ;pub use runtime::PermissionRequest ;pub use runtime::PermissionRequestHandler ;pub use runtime::PermissionResponse ;pub use runtime::RunContext ;pub use runtime::ToolSuspendHandler ;pub use runtime::ToolSuspendRequest ;pub use runtime::ProcessSandboxPlane ;pub use runtime::SandboxOptions ;pub use runtime::RemoteVpcOptions ;pub use runtime::RemoteVpcPlane ;pub use safety::Permission ;pub use safety::PermissionDecision ;pub use safety::PermissionManager ;pub use safety::PermissionMode ;pub use signals::GatewayReceiver ;pub use signals::RuntimeSignal ;pub use signals::ScheduledPrompt ;pub use signals::SignalGateway ;pub use signals::SignalSource ;pub use tools::RegisteredTool ;pub use tools::SafeToolResult ;pub use tools::TextToolSession ;pub use tools::ToolChunk ;pub use tools::ToolEnvelope ;pub use tools::ToolEnvelopeFail ;pub use tools::ToolEnvelopeOk ;pub use tools::ToolSession ;pub use tools::ToolStep ;pub use tools::execute_tools ;pub use tools::fail ;pub use tools::ok ;pub use tools::read_file_tool ;pub use tools::safe_tool ;pub use tools::tool_fail ;pub use tools::validate_tool_arguments ;governance harness harness_loop knowledge memory providers run_event runtime Runtime v1 — session event log, execution planes, credential vault, and runner. safety signals tools Criterion Re-export the kernel’s Criterion so callers don’t need to import from deepstrike_core.
A single evaluation criterion with optional weight and required flag. JudgeMatch A pairwise judge assignment carried to the SDK on a tournament judge’s WorkflowSpawnInfo:
the two entrant agent ids whose produced outputs are to be compared. The SDK maps each id back
to that entrant’s candidate and asks the judge which is better. MemoryMetadata Lightweight memory metadata (kernel stores, SDK provides full content). MemoryPolicy Declarative configuration for the kernel’s long-term memory subsystem. MemoryQuery Memory query request (kernel → SDK). MemoryRetrieval Memory retrieval response (SDK → kernel). MemoryWriteRequest Memory write request (SDK → kernel). RenderedContext Structured render output aligned with LLM API slots. ResourceQuota Opt-in resource limits. An unset field imposes no limit; an unset ResourceQuota (the default,
when [crate::scheduler::state_machine::LoopStateMachine::set_resource_quota] is never called)
preserves the pre-M2 behavior of unconditional Allow for spawn / memory syscalls. WorkflowNode One node in a workflow DAG: a task plus the contract its agent runs under. WorkflowRun The state of one in-flight workflow execution. WorkflowSpawnInfo Enough to run one spawned workflow node, carried to the SDK in the WorkflowBatchSpawned
observation. Role/isolation/inheritance are canonical snake_case strings (serde names) so the
host SDK can rebuild an agent run spec — the kernel generates these specs internally, so this
is how the goal reaches the SDK that actually executes the node. WorkflowSpec A declarative workflow DAG. Error MemoryKind Memory kind (4 types, mirroring Claude Code’s taxonomy). PermissionAction Permission action for a tool. fanout_synthesize N parallel workers feeding a single synthesize barrier that depends on all of them. format_tool_error Error-aware serialization for tool-execution error paths. Replaces e.to_string() at the
sites that hand the model a failure message: gen_eval The generate→evaluate quality gate as a workflow: a Loop worker node (the task, re-run up
to max_iters, stopping early on a loop_continue=false self-signal) followed by a Verify
eval node that scores the worker’s output against the goal/criteria and emits a structured
verdict (crate::harness::verdict_output_schemaoutput_schema). generate_and_filter N parallel generators feeding a single filter/dedupe step that depends on all of them. verify_rules One fresh-context verifier per rule/claim, optionally followed by a skeptic that re-checks
every flag to suppress false positives. Result