dag-runtime

What problem does it solve?

Orchestrates and executes planned Directed Acyclic Graph (DAG) workflows to coordinate concurrent agents, enforce isolation and permissions, and provide full execution tracing and failure recovery so teams can run complex multi-node tasks reliably.

Core Features & Use Cases

• Parallel wave processing: Advance DAGs in waves with deterministic progression, allowing non-dependent nodes to run concurrently for speed.
• Agent spawning and isolation: Spawn agents per node with selectable isolation levels (container, process, context, or none) to balance security and functionality.
• Permission enforcement and inheritance: Ensure child node permissions are intersected with parent permissions to prevent permission creep.
• Failure escalation and retry policies: Classify failures (timeouts, schema errors, permissions, missing skills) and apply configurable retries, backoff, or human escalation.
• Execution tracing and salvage: Record node-level traces (model, tokens, cost, duration) and recover work from crashed agents for resumable runs.
• Use Case: Run a multi-wave security analysis DAG that isolates risky code execution in containers, retries malformed outputs with schema reminders, and composes a final report from available partial outputs.

Quick Start

Execute the DAG named "security-scan" with parallel waves, container isolation for risky nodes, and retry policies applied.