mesh-flow (version 2) for general ai agentic (codex, claude code, opencode)

Artifact-Driven DAG Orchestration

Modern AI agents often fail because of brittle, implicit prompt-chaining. Mesh Flow solves this by providing a robust compile-then-run DAG system that moves orchestration logic out of prompts and into a structured framework. It ensures your multi-agent workflows are predictable, verifiable, and governed by hard-coded logic rather than model probabilities.

What it does

• Decouples Topology from Logic: Define your workflow in a single project.yaml identifying nodes by what artifacts they consume and produce.
• Strict Validation: The compiler checks for circular dependencies, missing artifact producers, and schema violations before execution starts.
• Enforced Hard Gates: Implement human-in-the-loop approvals, artifact presence checks, and custom verifiers that models cannot bypass.
• Standardized Adapter Contract: Provides a uniform interface for all skill nodes, ensuring consistent trace logging and error reporting.

Why use this skill

Unlike simple sequential prompting, Mesh Flow offers a professional state machine with three distinct failure states (failed, blocked, rejected) and automated recovery paths. It includes a CLI for local validation, execution, and generating Mermaid visualizations of your agent's decision graph. This is essential for developers building production-grade agentic systems where reliability and observability are non-negotiable.