GuardMesh exists to make bounded agent decisions explicit, portable, and inspectable.
Modern agent systems often have strong building blocks around planning, delegation, contracts, skills, and execution traces. What is still frequently missing is a small, portable layer that can answer a practical boundary question at the point of action:
"Should this action proceed?"
GuardMesh focuses on that question.
GuardMesh is:
- a lightweight policy specification for bounded actions
- a validator for policy and request artifacts
- a decision engine that returns
allow,deny,escalate, orkill - a developer-friendly CLI for local evaluation and integration prototyping
GuardMesh is not:
- a full enterprise governance platform
- a distributed runtime sidecar system
- a control-room UI
- a database-backed policy service
- a complete approval workflow product
Those directions may matter later, but they are intentionally outside the first alpha.
The alpha should be narrow, credible, and easy to understand quickly.
If someone opens the repository, they should immediately understand:
- what a GuardMesh policy pack looks like
- what kind of request it evaluates
- how decisions are produced
- where the project could expand next without feeling bloated today
GuardMesh complements other modular agent building blocks. It does not replace execution frameworks, capability tokens, contracts, or trace systems. Instead, it provides a focused decision layer that can sit at runtime boundaries wherever bounded action checks are needed.
- clean repository structure
- typed implementation with a coherent evaluation model
- strong baseline docs
- useful example packs
- local CLI flow that works end to end
The first version succeeds by being disciplined, not by being broad.