attractor

DOT pipeline orchestration is managed through a spec-driven layer that parses DOT DSL, normalizes graph structures, and executes deterministic traversals. This replaces ad-hoc glue code with a centralized flow control system for reproducible task completion.

Human-in-the-loop gating in graph execution pauses deterministic traversal at specific nodes, requiring manual approval before proceeding. It uses a node-handler registry with wait.human handlers to gate flow, ensuring human oversight for critical pipeline decisions.

Checkpoint persistence and resume state are added by implementing state management within the graph execution engine. The orchestration layer saves traversal progress at checkpoints, allowing workflows to resume deterministically from their last saved state after interruptions.

Yes, you can register custom node handlers for conditional and tool nodes using a modular node-handler registry. The registry supports built-in handlers like start, exit, wait.human, and conditional, while allowing clear interfaces for extending with custom handler implementations.

Attractor requires no external dependencies to run DOT DSL parsing and graph normalization. It operates independently to validate parsers, normalize graphs, and manage execution state, though it supports integrating backends like coding-agent-loop and unified-llm.

The spec-driven orchestration layer for DOT pipelines is limited by its strict deterministic traversal model, which may not suit highly dynamic or ad-hoc workflow modifications. It requires predefined node-handler interfaces and properly validated DOT syntax to function correctly.