mesh-flow (version 2) for general ai agent (openclaw, hermes agent)

Artifact-Driven DAG Orchestration

Complexity in AI agents often stems from implicit prompt-chaining where flow logic is buried inside instructions. mesh-flow solves this by introducing a strict, compile-then-run DAG (Directed Acyclic Graph) architecture. It decouples the flow topology from individual node reasoning, ensuring your agents follow a predictable, reproducible, and verifiable path.

What it does

This skill provides a robust framework for building complex agentic workflows using artifact-driven modeling. Instead of telling an agent to "then do X," you define what artifacts a node consumes and what it produces. The system uses a dedicated CLI to validate dependencies, detect cycles, and compile your YAML definitions into a normalized execution plan.

• Explicit Topology: Uses project.yaml as the single source of truth for your flow logic.
• Hard Gates: Enforces runtime logic (like human approval or upstream success) that prompts cannot hallucinate their way through.
• State Machine Execution: Manages node states (failed, blocked, rejected) with explicit recovery paths.
• Standardized Tracing: Generates detailed execution traces for every node, including prompt templates and tool calls.

Why use this skill

Unlike standard prompting, mesh-flow provides a structural "spine" for your agents. It prevents flow drift, enables shadow-mode testing for complex migrations, and provides a CLI for local validation and Mermaid diagram generation. It is ideal for developers building multi-step pipelines where reliability and auditability are non-negotiable.