How it Works:
Instead of shipping binaries or source code, you share instructions and specs in form of a prompt. You can take this prompt, paste it into their agent or IDE and watch it build. If it’s not a perfect fit? Fork it, tweak it, and generate your custom version.
All meta infos like version, description, test cases etc. are stored in a frontmatter block at the start of the prompt. So it's one file containing all the infos you need. (https://openprompthub.io/docs)
Features of the platform: - Versioned prompts with infos which models to best use - Forking for customization - Security Scans: prompts are scanned for security issues and prompt injections. - User can give feedback, if the prompt successfully build what was promissed (scoped on models, so you know which one to best use for execution) - Flagging mechanism
It’s an MVP, but the core features—versioning, model-specific build status, and security scanning — are live.
I'm currently looking into further features, such as: - a git-like cli for publishing prompts and downloading/piping them directly to your agent - multi stage/file prompts for more complex applications - configurable prompts for e.g. switching programming languages, features, etc. - better spec and test definition for build verification
I’d love your feedback... on the idea, the spec and the platform.