Status: early preview. This repository is a short statement of intent, not a specification. There is no schema here yet, and nothing here should be treated as stable or normative — it exists so people can see where SeedSpec is headed before the protocol itself is public.
SeedSpec is a protocol for packaging product intent — ideas, product specs, configuration, integration context, and related artifacts — into a portable format that a person or an implementing agent can pick up and act on, without the protocol prescribing the resulting application's architecture, language, or workflow.
The rough shape:
idea + optional artifacts
-> a SeedSpec package
-> configuration + feature discovery
-> a resolved project specification
-> user-directed agent handoff
-> a working application + verification evidence
A few intentions that are already fairly settled, even this early:
- Product intent over implementation detail. A package describes what should exist, not how to build it.
- Portability without forced generality. Packages should move between agents and toolchains without every package having to anticipate every possible consumer.
- The end user directs the implementation agent. SeedSpec hands off; it doesn't orchestrate.
- Markdown for behavior, structured data for discovery and configuration. Human-readable where it matters, structured where it needs to be validated.
Expect this list to change. The current, real work is happening in the protocol repository — this README is a placeholder for the story, not the substance.
SeedSpec is organized across three sites, each with a different job:
- seedspec.dev — the protocol. Where developers and agents go for the spec itself, raw files, and machine-readable docs.
- seedspec.org — the SeedSpec Foundation. The neutral, non-commercial home for governance and stewardship of the protocol.
- seedspec.ai — hosting. A commercial product built on top of the protocol, kept separate from it.
All three are early placeholders today.