kaddy v0.1.0 — serve → scrape → fire, all GitOps
kaddy v0.1.0 — local-first Website-as-a-Service platform milestone
The gridscale hiring-exercise brief, answered end-to-end and verified live:
Highlights
- Serve → scrape → fire: the
clubhousesample site is served over verified HTTPS through the
Cilium Gateway (cert-manager local CA, no-k); a blackbox probe watches it with CA-trusted
TLS;task demo:firebreaks it →ClubhouseDownfires in Alertmanager → restores → resolves
(~90 s). 5 marshal alerts, every one promtool fire/silent-tested with a coverage gate. - GitOps everything: ArgoCD app-of-apps — 14/14 Applications Synced/Healthy on
main
(spine, observability, monitoring, rollouts, policies, kyverno). Control-plane apps self-heal. - Progressive delivery (mulligan): Argo Rollouts + Gateway API plugin mutates live HTTPRoute
canary weights (100/0 → 20 → 50 → 100),task demo:chaosproves abort → auto-rollback. - Observability as code: kube-prometheus-stack + Loki + Alloy (capacity-trimmed for an 8 GB kind
node), Grafana dashboardkaddy-marshal(12 panels) provisioned via sidecar,
data-source-managed alerts verified through the Grafana API. - Enforcing security baseline: Kyverno v1.18.2 with 5 Enforce policies (ADR-0301 labels,
data-classification vocabulary, pod-security trio) + verifyImages (Audit, keyless attestor),
default-deny NetworkPolicies (incl. the Cilium Gateway hairpin-identity fix), restricted
ArgoCD AppProjects, SOPS-age secrets, gitleaks + pinned supply chain in CI. - Keyless signing:
ghcr.io/platformrelay/kaddy-showcase(the Slidev deck served by Caddy,
multi-stage, non-root) is built, pushed, cosign-signed via GitHub OIDC and verified in-workflow. - Interview deck: recording-ready Slidev deck — 1358-word speaker script, narrative beats with a
~10 min time budget, live iframe surfaces, AI cover-art prompt suite + branding prompts — gated in CI. - Test pyramid in CI: tofu test (17) · conftest/Rego · kyverno CLI (28) · promtool (13 cases) ·
chainsaw e2e on kind · deck gates — 5 workflows, all green on the tagged commit.
Known limitations (lab threat model — tracked with targets)
- No RBAC/OIDC yet (E1d); ArgoCD
--insecurein-cluster behind the Gateway; Grafana admin Secret is
imperative until KSOPS (E1d). verify-signed-imagesstays Audit until kaddy-signed images are the only
ghcr.io/platformrelay/*images deployed.- 4 Helm grandchild apps still
project: default(documented);mulligannamespace not yet under
default-deny; Loki-ruler log alerts + external Alertmanager receiver deferred (E5 follow-ups). - Cilium/Rollouts-dependent chainsaw suites are live-verified but CI-skipped (no Cilium substrate in
CI); Let's Encrypt prod path is cloud-only. - Deferred epics: E6 Crossplane
WebsiteXRD, E8 evidence, E9 operator, E10 portal, E13 marketplace,
e-caddy-mvptenant variants.
Three replayable audits (baseline → mid-session → final) with 0 regressions across runs:
verdict RELEASE-READY. Full history in CHANGELOG.md.