diff --git a/.release.conf b/.release.conf
new file mode 100644
index 0000000..6401c12
--- /dev/null
+++ b/.release.conf
@@ -0,0 +1,17 @@
+# Read by the `release` skill (~/.claude/skills/release).
+#
+# Only what the Makefile cannot already say. IMAGE, HELM_CHART and HELM_REPO come from
+# `make release-vars`, so they have one definition and cannot drift from what is built.
+#
+# Defaults, set here only where this repo differs:
+#   CHARTS_REPO=$HOME/git/charts   CHARTS_DIR=<image basename>
+#   GITEA_LOGIN=Ryuvia            APPVERSION_PREFIX=
+
+# Same as the image basename, so this is only stated to be read rather than derived.
+CHARTS_DIR=terdut-server
+
+# riksdata writes appVersion: "v0.3.1", rd-web writes a bare 0.5.0; this repo writes the
+# v, like riksdata. Nothing reads the field -- .gitea/workflows/release.yaml stamps both
+# version and appVersion from the tag when it publishes -- but people read it, and until
+# 2026-09-01 it said "latest" while the tree headed for a numbered release.
+APPVERSION_PREFIX=v
diff --git a/Makefile b/Makefile
new file mode 100644
index 0000000..95a7328
--- /dev/null
+++ b/Makefile
@@ -0,0 +1,84 @@
+REGISTRY  := git.ryuvia.com
+# The personal namespace, not ryuvia — deliberately, and for one reason: Gitea
+# scopes package visibility to the owner with no per-package override, so
+# ryuvia/* is private because the org is. Publishing here keeps the image and
+# chart anonymously pullable, so no pull secret is needed in the cluster and
+# Flux needs no registry credentials. Same choice riksdata and rd-web made.
+OWNER     := niklas
+
+IMAGE      := $(REGISTRY)/$(OWNER)/terdut-server
+HELM_CHART := charts/terdut-server
+HELM_REPO  := oci://$(REGISTRY)/$(OWNER)
+
+# go.mod pins an exact patch release so nobody builds the shipped binary with a
+# toolchain carrying known stdlib CVEs. Fedora's Go package overrides the
+# upstream GOTOOLCHAIN default to `local`, which turns that pin into a hard
+# failure on a dev box one patch behind, so restore the upstream default here.
+export GOTOOLCHAIN ?= auto
+
+.PHONY: help
+help: ## Show this help
+	@grep -hE '^[a-zA-Z_-]+:.*?## ' $(MAKEFILE_LIST) | \
+		awk 'BEGIN {FS = ":.*?## "}; {printf "  \033[36m%-22s\033[0m %s\n", $$1, $$2}'
+
+## --- checks ---
+#
+# These three mirror .gitea/workflows/ci.yaml step for step, so a green `make fmt
+# lint test` here means the same thing CI means. The one deliberate difference is
+# -race below.
+
+.PHONY: test
+test: ## Run the test suite
+	go test -race ./...
+
+# CI runs a bare `go test ./...`. This is stricter on purpose: the sweeper, the
+# notifier goroutine and the deadman sweep all touch the same single-connection
+# database, and a race there would surface as a flaky production incident rather
+# than a failed build. It passes today; if it ever costs more than it catches,
+# the honest fix is to teach CI -race too, not to quietly drop it here.
+.PHONY: lint
+lint: ## go vet
+	go vet ./...
+
+# Copied from ci.yaml rather than simplified, because both of gofmt's failure
+# modes need handling and they are not alike. A file that is merely misformatted
+# is listed on stdout with exit 0 — so the failure has to be raised by hand. A
+# file that does not parse is the opposite: nothing on stdout and exit 2, which a
+# naive `[ -n "$$out" ]` reads as success. See 9046f6e.
+.PHONY: fmt
+fmt: ## Report unformatted files
+	@if ! unformatted=$$(gofmt -l .); then \
+		echo "gofmt could not parse the tree:"; gofmt -l .; exit 1; \
+	fi; \
+	if [ -n "$$unformatted" ]; then \
+		echo "gofmt needed:"; echo "$$unformatted"; gofmt -d .; exit 1; \
+	fi
+
+.PHONY: helm-lint
+helm-lint: ## Lint and render the chart
+	helm lint $(HELM_CHART) --set image.tag=v0.0.0
+	helm template terdut-server $(HELM_CHART) --namespace terdut-server \
+		--set image.tag=v0.0.0 >/dev/null
+	@# networking.listener defaults to "", which attaches the route to every
+	@# matching listener including plaintext HTTP. Production sets it, so the
+	@# default render proves nothing about the path that actually ships.
+	helm template terdut-server $(HELM_CHART) --namespace terdut-server \
+		--set image.tag=v0.0.0 --set networking.listener=https-terdut >/dev/null
+
+## --- release ---
+
+# The release process (~/.claude/skills/release) reads these rather than restating them.
+# One definition, so the version that gets tagged, the image that gets pushed and the chart
+# the wrapper pins cannot drift apart in a second copy.
+.PHONY: release-vars
+release-vars: ## Print the variables the release process reads
+	@printf 'IMAGE=%s\nHELM_CHART=%s\nHELM_REPO=%s\n' '$(IMAGE)' '$(HELM_CHART)' '$(HELM_REPO)'
+
+# There is deliberately no build/push/helm-package/helm-push/release here, unlike
+# riksdata and rd-web. .gitea/workflows/release.yaml owns publishing for this repo,
+# and it does two things a local make cannot: it builds linux/amd64 and linux/arm64
+# through buildx, and it stamps the chart's version and appVersion from the tag. A
+# `docker build && docker push` target would push a single-architecture image over
+# the multi-arch tag, which is both easy to do by accident and invisible afterwards
+# — the tag would still resolve, just not on arm64. Publishing happens by pushing a
+# tag; nothing else.