A Vibestrate flow is a reusable recipe for supervised AI coding. It defines the phases, roles, model seats, approval gates, validators, and handoffs that turn one brief into a run you can watch and approve.
Why flows matter
A one-off prompt chain disappears when the chat ends. A flow stays. It gives the team a named process for recurring work: bugfix loops, deep refactors, quality arbitration, docs syncs, migrations, security review, and anything else that benefits from structured handoffs.
This is also how Vibestrate avoids single-model dependency. The flow defines the work. The crew can change. That means the same process can run with frontier models, cheaper models, local models, or a mixed crew.
What a flow can define
- Phases: plan, architect, implement, review, fix, verify, or custom steps.
- Roles: planner, builder, reviewer, verifier, arbiter, or custom roles.
- Approval gates: the moments where the human stays in control.
- Validators: commands and checks that decide whether work can continue.
- Handoffs: what each phase passes to the next model or reviewer.
- Metadata: tags, summaries, risk level, version, and author identity.
Pull, fork, publish
| Action | Command | What it means |
|---|---|---|
| Pull | vibe flow pull quality-arbitration | Install a verified flow and run it with your own local setup. |
| Pull community | vibe flow pull mira@deep-refactor | Install a community-authored flow under the author handle. |
| Publish | vibe flow publish my-flow | Share your workflow so others can inspect, fork, and run it. |
Shared does not mean hosted
Shared flows are recipes, not a cloud relay. Vibestrate still runs locally, uses your vendor CLIs, keeps your keys out of the orchestrator, and records the run on your machine. Pulling a flow gives you process, not a hosted black box.
The hub is the community layer for "how should this kind of AI coding work run?" You bring the models, keys, codebase, and approval.
