The OpenPrompts Manifesto
The way we talk to machines should be open.
OpenPrompts is an open, community-tended catalog of the system prompts and skills that shape today's AI. We gather them, attribute them, and put them where anyone can read, compare and learn from them.
- 01
Prompts are infrastructure.
A system prompt is not a trick. It is the operating manual of a model: it decides what an assistant is, what it refuses, how it reasons. As models run more of the world, these instructions become shared infrastructure. Infrastructure should be inspectable.
- 02
Open by default.
Anthropic and OpenAI publish the prompts behind their assistants. Open tools like Crush ship theirs in the open. We believe that is the right default: the people using a system deserve to read the instructions that shape it. OpenPrompts exists to gather that openness in one place.
- 03
Attribution, always.
Every entry keeps its source and its license, and links back to where it came from. We collect and organize; we do not launder. Prompts and skills remain the work and property of the people and companies who wrote them.
- 04
Transparency over secrecy.
A prompt kept secret cannot be reviewed, criticized, or improved. Hidden instructions are where bias, manipulation, and quiet failure live. Reading prompts in the open is how we hold systems, and ourselves, accountable.
- 05
Vendor-neutral.
Frontier labs, local and offline models, terminal agents, community experiments: they all belong here. Good prompting is a craft that crosses providers. We organize by what a prompt does, not by who owns the model.
- 06
A living catalog.
Prompts change with every model release. This is not an archive; it is a working reference that anyone can extend. Add a source, run the scraper, open a pull request. The catalog is only as good as the community that tends it.
Read the prompts. Add your own.
Every entry is open, sourced and free to learn from. Browse the catalog, or contribute a prompt or skill you think the community should see.