Create your first project
A project is the long-lived container for a body of work. It owns three things:
- its enclave — the security boundary it lives in, fixed at creation;
- a human-readable name and optional description;
- a per-project library — design docs, guidelines, a goal doc, and an implementation journal that travel with every agent run.
Projects deliberately do not carry a node, workspace path, or AI account — that per-run configuration lives on each session, so one project can run many concurrent sessions across different machines and accounts.
1. Create the project
Section titled “1. Create the project”Choose New project and provide:
- Name (and an optional description).
- Enclave. Pick the boundary this project belongs to. It is fixed at creation and scopes every credential, node, and AI account the project can ever reach — so choose the one that matches the code’s sensitivity (e.g. a corporate-private enclave vs. an open-source one). See Enclaves & security.
That’s it — a project is just a name + enclave + its library. The code isn’t attached to the project; each session points at a workspace path on a node where the working tree lives (today, host-directory mode means the repo is already checked out there). Automatic repository linking through a Git provider is coming soon.
2. Write a goal document
Section titled “2. Write a goal document”Every project gets a single goal document — the high-level why. At dispatch it is placed first in the context an agent reads, before designs and guidelines, so the agent understands intent before implementation. You can add it now or after your first session.
The goal doc lives in the project library alongside two other categories you can fill in over time:
- design docs — architecture and how things fit together;
- guideline docs — conventions the agent should follow.
These are materialized into every session’s workspace automatically (as
CLAUDE.md plus files under .halyard/library/), so guidance you write once is
applied on every run. The library also keeps a journal — a running log of
what changed, with auto-generated entries after each session that you can edit.
3. Next steps
Section titled “3. Next steps”- Register a node so agents have somewhere to run.
- Dispatch an agent against your first piece of work.
- Break product intent into traceable work with PRDs & requirements.