Author your first artifacts
With your project created, you're ready to author its intent. The things you author are artifacts (sometimes called waypoints, after the nodes in the workflow DAG). They are small, schema-validated documents — a persona, a requirement, a bounded context, a feature — not free-form prose.
"Artifact" and "waypoint" mean the same thing here. The word token in Waypoint refers to an authentication credential (the bootstrap token / API key from earlier steps), not to the documents you author. Don't conflate the two.
How authoring actually works
You author artifacts by working with your coding agent, not by hand-editing files one at a time. waypoint init already installed the Waypoint MCP server into your agent client and added a mode-specific agent skill plus an AGENTS.md working block. So the loop is:
-
Open your agent in the repository. For example, run
claude(Claude Code),cursor ., orzed .in the project directory. -
Let the agent walk the DAG. Guided by the installed skill, the agent authors artifacts top-down in dependency order, stopping to iterate with you at each level:
idea→ personas / pain points / user goals / product goals→ requirements (EARS)→ architecture (components, data flows, integrations, bounded contexts…)→ interfaces (UI apps, CLI commands, MCP tools, API endpoints…)→ features (and their flows) -
The agent writes each artifact through the Waypoint MCP tools. The MCP server exposes four CRUD tools —
artifact.put,artifact.get,artifact.list, andartifact.delete. Aputwrites the artifact's content and atomically updateswaypoint.jsonin the same step.
The metagraph is the source of truth for the type catalog and dependency order. You can enumerate the types for the software workflow at any time:
waypoint workflow types software
See Artifacts & the DAG and The Workflow for the deeper model.
Prove coherence with waypoint validate
As you author, run validation to confirm your artifacts are coherent:
waypoint validate
This checks schema conformance, cross-reference integrity, orphan detection, structural integrity, and requirement coverage, and emits a proof certificate (waypoint.proof.json) recording why the verdict holds. Run it before you commit. See Validation for the model and the CLI validation reference for every flag.
Cache, upload, and render
A few CLI commands keep your local cache and the hosted service in sync:
waypoint pullwarms your local cache (~/.waypoint/cache/) by downloading only the artifact blobs thatwaypoint.jsonnames and that you don't already have. Run it aftergit pullbrought in a teammate's artifacts, so the nextvalidatedoesn't make per-artifact round trips.waypoint pushuploads every artifact blob thatwaypoint.jsonnames and the service lacks. You rarely run this by hand — the pre-push git hook installed byinitruns it automatically, so other users never see a hash without its content.
Commit and push
Finish by committing and pushing. The git hooks validate your artifacts and upload their content; once the push lands, the web app renders your graph:
git add . && git commit -m "feat: initial waypoint artifacts" && git push
That's the full loop: author with your agent, waypoint validate to prove coherence, commit and push to render. From here:
- Artifacts & the DAG and The Workflow — the deeper model behind what you just authored.
- Validation — how the proof certificate makes a verdict trustworthy.
- CLI Reference — full docs for every command, including projects & artifacts and integrations.