Foundry turns a spec doc — or a work-in-progress repo — into its own isolated Todoist project, an MVP build loop, and eventually its own staging deploy pipeline. No shared state, ever.
Every Foundry-tracked project gets a manifest — slug, Todoist project id, repo, domains, MVP checklist status. Every foundry-* skill reads it instead of guessing. It's the single mechanism that keeps a skill from ever acting on a project it wasn't meant for.
Login, dashboard, one real feature — or whatever the spec defines as MVP — gets built before any deploy infrastructure is touched. The moment the gate passes, foundry-deploy-staging spins up automatically. MVP2 and Future tasks keep moving in the background either way.
Own Todoist project, own repo, own VPS subdomain, own secrets, own scheduled routine — namespaced by slug from the start. Two Foundry projects, or a Foundry project and the existing Aura setup, should never be able to collide.
Nothing below has started — Phase 1 is blocked on a handful of open decisions first.
Manifest location, MVP definition source, routine cadence, single-vs-concurrent-project support. Naming is resolved — "Foundry."
Prove out spec/repo → isolated Todoist project with a prioritized MVP / MVP2 / Future breakdown. No code touched yet.
foundry-init · foundry-spec-to-tasksRun manually against one MVP task to confirm the worktree/test/evidence loop works standalone.
foundry-build-taskFirst full spec → MVP → live staging site run, end to end.
foundry-mvp-gate · foundry-deploy-stagingA new project behaves like Aura's setup — recurring pickup and review, on its own schedule.
foundry-routine-setup · foundry-reviewAdded last, on purpose. Always a separate, explicitly human-confirmed step — never auto-chained.
foundry-deploy-prodStaging can trigger itself once the MVP gate passes. Going live is always a decision Darren makes on purpose.