OriginAIProduct specs for Claude Code, Cursor, and Codex

OriginAI vs GitHub Spec Kit

Spec Kit is a markdown specify → plan → tasks workflow in your repo. OriginAI is a visual product spec (RPML) you publish as an immutable release for Claude Code, Cursor, and Codex.

GitHub Spec Kit is the tool most people find when they search for spec-driven development. It is a repo-local toolkit: you write markdown specs, turn them into a plan and a task list, then let a coding agent implement those tasks. OriginAI is a different cut of the same problem — the product contract (screens, states, rules), hosted, versioned, and read by the agent as a published release.

They can sit in the same workflow. They are not substitutes.

Side by side

GitHub Spec KitOriginAI
ArtifactMarkdown in the repo (spec.md, plan.md, tasks.md)Structured RPML screens in OriginAI, published as a release
What it specifies wellIntent, scope, implementation plan, task breakdownUI: routes, states, roles, empty / loading / error, business rules on the screen
Source of truth after day oneFiles you keep editing in gitAn immutable release snapshot; later writes become change requests
How the agent reads itOpens files in the workspaceoriginai CLI or MCP (get-diff, list-documents, …) against the release
Visual previewNo (prose / checklists)Yes — workbench canvas, same language as @21stware/rpui
HostingYour git repogetoriginai.com + optional public release HTML
Best next search termspec kit, specify plan tasksRPML, product spec for Claude Code

Spec Kit owns the "how we break work down" conversation. OriginAI owns "what the product must look like when each rule fires."

When to use which

Use Spec Kit when the bottleneck is planning and tasking — a greenfield feature, an architecture decision, a long implement loop inside one repo. The agent already has the codebase; it needs a constitution and a checklist.

Use OriginAI when the bottleneck is product context that chat keeps dropping: which states exist, who sees which control, what the empty table says, which confirm dialog runs. That is the layer screenshots and tasks.md skip. You write it once in RPML, publish a release, and every session of Claude Code, Cursor, or Codex reads the same snapshot.

Use both when a feature has a UI surface. Spec Kit can plan the work; the OriginAI release is the contract the implementation is judged against (including the GitHub App spec-review check).

What OriginAI is not

OriginAI is not Cursor Origin (an AI-native git host), not getorigin.io (AI code attribution), and not a no-code app builder. The product name is always OriginAI. The CLI / npm package is originai.

On this page