A deterministic development harness for mixed human–AI teams, paired with a production-ready SaaS foundation. Use either independently or combine them, without rebuilding the stack and practices behind reliable software.
The team preset follows seven visible stages. Ticket nature activates explicit, guarded shortcuts for internal work, bundled changes and Epics. Humans approve intent and behavior; automation verifies the code; the connected board keeps the audit trail. GitHub Projects provides the complete v1 delivery contract; Jira and Linear board adapters are experimental.
The gates say what humans actually do. Human testing validates the feature in a real runtime; In review reviews the ready pull request and its code. The agent can prepare evidence, but it cannot approve its own work.
Pay for rigor, not theatre. Ceremony scales with risk, so a typo does not consume the same context and review budget as an authorization change.
Bring the harness to any SaaS codebase to make proven engineering practices deterministic. Use the prebuilt architecture on its own when you need the common product foundation. Combine both to start with robust delivery and robust software—without reimplementing either.
Exemplary delivery practices become an inspectable project system. Developers and AI agents work as one team through shared SRS, tickets, skills, tests, human gates and pull requests.
Explore the harness →02 / FOUNDATIONAuthentication, tenancy, scoped RBAC, typed APIs, React, PostgreSQL and the operational baseline already work together before your first business feature.
See what is built in →SaaSFoundryAI is not tied to Claude Code. The harness shares project rules with six coding-agent profiles. Separately, provider-neutral execution contracts let a host integration classify a subtask and qualify provider, runtime, model and reasoning-effort candidates that it explicitly supplies. Workflow complexity drives operational delegation and independent review today.
The v1 boundary: these two layers are not automatically wired together. The sf CLI does not route each subtask to a model. A host or integration must supply execution candidates, invoke the planning contracts and dispatch the selected agent or model; SaaSFoundry never installs provider accounts, copies credentials or invents unavailable candidates.
See how agent hosts coexist → · Inspect adaptive execution planning →
Start with a coherent product rather than a pile of disconnected packages. Each capability below is generated, wired and documented as part of the same architecture.
AI is an interface to SaaSFoundryAI, not a replacement for its CLI. Both paths use the same manifest, installers, validation and generated project.
“Create my SaaS workspace”Describe the outcome. Your coding agent reads the project, asks only for missing decisions, then drives the explicit SaaSFoundry commands.
Install the agent skill →sf new --project-name my-productUse the interactive flow or scripted flags. The CLI remains deterministic, inspectable and suitable for automation without a model in the loop.
Open the CLI quick start →SaaSFoundryAI dogfoods its own harness. The generated product is exercised from browser to API to PostgreSQL across both supported topologies.
Live lifecycle tests cover authentication, invitations, tenant boundaries, scoped permissions, module control and account reactivation without external providers.
Validated manifests, numbered migrations and conflict-aware updates let the foundation evolve without treating generated projects as disposable demos.
Adopt the harness in an existing SaaS, start from the production foundation, or combine both. The common engineering decisions stay explicit and reproducible.