Execution is getting cheap.
Coordination isn't.
PlanSpec is a declarative specification that makes planning first-class, enabling coordination of agentic and human work via explicit, inspectable, durable, and composable plans.
Designed for environments where execution outpaces coordination.
Why PlanSpec exists
Today, planning is mostly implicit:
- Encoded in prompts and markdown
- Scattered across chat logs
- Reconstructed after the fact
This breaks down at scale.
When you introduce fast, parallel execution (LLMs, agents, automation):
- Coordination becomes the bottleneck
- Intent fragments
- Plans drift
- Review, pause, and reuse break down
PlanSpec makes plans first-class, durable artifacts.
What PlanSpec is (and is not)
What it is
- —A declarative specification for plans
- —A graph (DAG) of intent, structure, and constraints
- —Schema-validated and execution-agnostic
- —Designed for both humans and machines
- —Goal-oriented, with explicit task decomposition
- —A familiar, Kubernetes-inspired schema
What it is not
- —Not an agent framework
- —Not a workflow engine
- —Not an execution runtime
- —Not a replacement for your tools
PlanSpec describes what should happen, not how to make it happen.
Playground
Validate and visualize PlanSpec documents. The playground validates against the published v1alpha1 schemas and renders the plan graph.
Loading graph...
Click a node in the graph to view details, or validation errors will appear here.
The playground validates and visualizes structure only — it does not execute plans.
Status and scope
PlanSpec is early.
- Schemas are published as v1alpha1
- The core concepts are stable
- Tooling will evolve
The goal of this release is clarity, not completeness.
Open Source
PlanSpec is open source, dual-licensed under MIT or Apache-2.0. The specification, schemas, and examples are available on GitHub.