Module runner 

Runner module - thin execution shell

The runner is a thin orchestration layer that:

• Accepts invocations
• Passes to Kernel
• Wires callbacks
• Returns streams

Core execution logic lives in the kernel module. Context types are in the context module.

CRITICAL INVARIANT: Runner Never Mutates State

Runner is a thin orchestration shell. All execution semantics live in kernel/.

Runner MUST NOT:

• Mutate execution state directly
• Retry steps (kernel handles retries)
• Handle backoff (kernel handles backoff)
• Skip nodes (kernel handles node execution)

If Runner starts doing these things → duplicated kernel logic (architectural leak).

What Runner Does:

• Wires services (session, memory, artifacts)
• Handles callbacks (before/after hooks)
• Delegates execution to ExecutionKernel
• Streams events to clients

What Runner Does NOT Do:

• Mutate execution state directly
• Implement execution semantics
• Make policy decisions

@see docs/TECHNICAL/01-EXECUTION-TELEMETRY.md @see docs/TECHNICAL/25-STREAM-PROCESSORS.md

Re-exports

pub use approval_policy::AlwaysApprovePolicy;

pub use approval_policy::AlwaysRequireApprovalPolicy;

pub use approval_policy::ApprovalPolicy;

pub use approval_policy::ApprovalPolicyConfig;

pub use approval_policy::CompositeApprovalPolicy;

pub use approval_policy::PatternApprovalPolicy;

pub use approval_policy::ThresholdApprovalPolicy;

pub use crate::kernel::run_parallel;

pub use crate::kernel::InterruptDecision;

pub use crate::kernel::InterruptReason;

pub use crate::kernel::InterruptableRunner;

pub use crate::kernel::ParallelResult;

pub use crate::context::InvocationContext;

pub use crate::context::InvocationServices;

pub use crate::context::ResourceLimits;

pub use crate::context::RuntimeContext;

pub use crate::context::RuntimeContextBuilder;

pub use crate::context::SessionContext;

pub use crate::context::TraceContext;

pub use callbacks::AfterAgentCallback;

pub use callbacks::AfterModelCallback;

pub use callbacks::AfterToolCallback;

pub use callbacks::AgentCallbackContext;

pub use callbacks::AgentCallbackResult;

pub use callbacks::BeforeAgentCallback;

pub use callbacks::BeforeModelCallback;

pub use callbacks::BeforeToolCallback;

pub use callbacks::CallbackRegistry;

pub use callbacks::ExecutionCallbacks;

pub use callbacks::ModelCallbackContext;

pub use callbacks::ModelCallbackResult;

pub use callbacks::NoOpCallbacks;

pub use callbacks::ToolCallbackContext;

pub use callbacks::ToolCallbackResult;

Modules

approval_policy

Approval policy for human-in-the-loop plan approval

callbacks

Runner Callbacks - Before/after hooks for agents, models, and tools

Structs

AgenticLoop

The agentic loop driver

AgenticLoopResult

Result of agentic loop execution

DiscoveredStep

A discovered step parsed from callable output

DiscoveryOutput

Result of parsing callable output for discovery signals

InputBlockedError

Error returned when input is blocked by processors

ProtectedRunner

Protected Runner

Runner

Runner - executes agents/graphs with abort, pause, and event streaming

Type Aliases

DefaultProtectedRunner

Protected Runner with in-memory checkpoint store (default)

DefaultRunner

Runner with in-memory checkpoint store (default)