A SaaS starter that tells your coding agent what to preserve.
Toolbound Stack is designed for founders who want Claude, Codex, Cursor, or Windsurf to customise a paid SaaS without breaking billing, email, database, auth, setup, or verification rails.
Codex-ready SaaS starter · Claude-ready SaaS boilerplate · SaaS starter for Cursor · SaaS starter for Windsurf
AGENTS.md and CLAUDE.md define safe edit zones and high-caution billing, usage, email, and outreach areas.
The /handoff page gives buyers a one-paste prompt for the first coding-agent session.
The /setup page shows integration readiness without printing secret values.
The repo keeps Stripe, Resend, Postgres, Clerk-ready auth, and release delivery documented as rails to preserve.
Agent handoff
The first agent prompt should not start from a blank repo.
Coding agents perform better when the repo has a memory layer. Toolbound gives them product context, architecture notes, domain language, setup docs, and verification commands before they start editing.
Use AGENTS.md for repo-level agent behaviour.
Use ARCHITECTURE.md and DOMAIN_GLOSSARY.md for shared context.
Use START_HERE.md and FIRST_RUN.md for buyer setup.
Use docs/codebase-map.md to locate product, billing, email, auth, and discovery surfaces.
Safety loop
Every serious agent edit needs a verification path.
Toolbound treats verification as part of the product, not an optional cleanup step. Agents are instructed to run tests, linting, builds, and setup checks before claiming a launch-facing change is done.
npm run setup:check for integration readiness.
npm run test for business logic and metadata contracts.
npm run lint for code quality.
npm run build for production route and metadata validation.
FAQ
Common questions
What makes Toolbound Codex-ready?
It includes repo instructions, safe edit zones, setup checks, and a one-paste handoff prompt so Codex can start with product context instead of guessing the architecture.
What makes Toolbound Claude-ready?
It includes CLAUDE.md, AGENTS.md, architecture notes, setup docs, and high-caution boundaries for billing, auth, email, database, usage, and outreach flows.
Can I still use Cursor or Windsurf?
Yes. The repo memory is Markdown and the verification commands are normal npm scripts, so the same guidance works across common coding-agent tools.