AI-native, spec-driven enterprise .NET accelerator: composable NuGet libraries + templates + AI skills + guardrails + living documentation. Not a framework — a golden path: a road paved with enterprise opinion on top of Microsoft (Aspire, Extensions.*).
New here? The guide — getting started, the six concepts, the CorPay tour, proof stories. Conceptual grounding: Strategy/Foundation · Constitution: ADR-0001..0010
- ✔ Composable packages — added to an existing project in 5 minutes (L1) or scaffolded from scratch (L3)
- ✔ Spec-driven:
.goldpath/manifest.yaml+ OpenAPI/AsyncAPI as the single source of truth - ✔ AI in the development layer: skills generate, guardrails verify, humans approve
- ✔ Proof-driven: every claim below is a test/bench that runs in CI (nightly 10-shape golden-manifest matrix, real containers)
- ✘ Not a framework (imposes no structure), no custom DSL, no wrappers around Microsoft
dotnet new install Goldpath.Templates@0.1.0-preview.6 # preview: pin the version
dotnet tool install -g Goldpath.Cli --prerelease
dotnet tool install -g specdrift # the deterministic engine behind goldpath check/add
dotnet new goldpath-solution -n Acme.Orders --db postgresql --broker rabbitmq --features bulk
cd Acme.Orders && goldpath check # spec validate + drift + build, one verb
dotnet run --project src/Acme.Orders.AppHostGrow it feature by feature: goldpath add feature notification, goldpath add worker,
goldpath db add AddInvoices. Every admin surface (/goldpath/admin/*) is fail-closed,
audited and contract-frozen; every module ships
its Grafana board, runbooks and measured performance
on a pinned CI profile.
One line — app.MapGoldpathConsole() — and the app serves its own operations console out
of the Goldpath.Console package's embedded assets: no Node on the running system, behind
the same fail-closed ops floor as /goldpath/admin/*, and lit only by the modules the app
actually composes.
Today answers the only question an operator opens a console to ask, across every service in the registry, before opening any of them:
Runs discovers the fleets from the store — every run with its chunks, its repair queue
and its deadline verdict, and rerun for the one that failed:
Bulk intake puts the engine's own validation report under the four-eyes gate, so the person approving sees exactly which row the file got wrong:
These are captured, never mocked: scripts/console-screenshots.sh brings up Postgres and
RabbitMQ, runs a real app, uploads a real file, triggers real jobs, and photographs what
comes back. The console is proven the same way — scripts/console-smoke.sh drives three
real apps (open, auth-floored, tenant-scoped) plus the app-served console, under an axe
accessibility gate.
| Layer | Packages |
|---|---|
| Floor (Ring A) | Abstractions · ServiceDefaults · ApiDefaults · Data · Messaging |
| Cross-cutting (Ring B) | Auth · Idempotency · AuditTrail · MultiTenancy · SoftDelete · Locking (+SqlServer) · Caching · DataProtection |
| Execution ladder (L2→L4) | Jobs (clustered, checkpointed, kill-9-recoverable) · Archival · Bulk (finance-grade intake→gate→execute→repair) · Notification · Campaign (paced fan-out, live throttle) |
| Operations | Console (MapGoldpathConsole() — the app serves its own, from embedded assets) |
| Tooling | Analyzers (GP#### executable standards) · goldpath CLI · Goldpath.Templates |
| Directory | Contents |
|---|---|
docs/strategy/ · docs/adr/ · docs/rfc/ |
Strategy, the 10-ADR constitution, module RFCs + frozen contracts |
docs/guide/ · docs/stories/ |
The adopter's path (start here) · proof stories |
docs/ops/ · docs/upgrades/ |
Migrations/trace/release runbooks · per-release upgrade guides |
schemas/manifest/v1/ |
Manifest JSON Schema + valid/invalid corpus (CI corpus gate: valid must pass, invalid must fail) |
packages/ · analyzers/ |
The NuGet train (20 packages) · GP#### Roslyn rules |
templates/ · tools/ |
dotnet new pack (solution + worker) · the goldpath CLI |
ui/kit/ · ui/console/ |
The console's primitives · the console itself (embedded into Goldpath.Console) |
tests/ |
619 unit + 32 integration proofs (Testcontainers) + bench suite; 204 UI unit tests + a browser smoke against three real apps |
skills/ · rulesets/ · samples/ |
pointers — the shipped skill layer and rulesets live inside templates/ · reference app (CorPay) |
Everything in this repo — documentation, code, identifiers, commits, PRs — is English. Conversation with the team may be Turkish. In customer repos, the domain-memory language is chosen per customer (Turkish by default for Turkish customers — see domain-memory-v1 §4).
Every dependency, including first-party OSS (Mediant, Mockifyr, Spec Engine — personal GitHub), is consumed as a published, pinned NuGet package (nuget.org → internal mirror where required). Source references/submodules are forbidden. Details: foundation §10.
One version train: every Goldpath.* package, the CLI and the template pack ship the
same version. SemVer with the pre-1.0 rules spelled out — 0.x.y patches are always
safe to take blind; 0.(x+1) minors may break but never silently: every break ships
with a step-by-step upgrade guide (docs/upgrades/), and the PublicAPI.*.txt ledger
diff is the mechanical proof of what changed. Support: latest release only pre-1.0;
from 1.0, previous major gets security fixes for 6 months. The full written contract:
docs/rfc/goldpath-versioning.md.