Releases move from the self-signed CN=unom cert to Azure Artifact Signing (formerly Trusted Signing): account `unomsigning`, profile `unom-io`, signed by the `punktfunk-ci-signing` service principal, which holds only the Artifact Signing Certificate Profile Signer role scoped to that one profile. Both pack scripts gain the backend ahead of the existing .pfx and ephemeral fallbacks, so canary and fork builds are unaffected. Three things that are easy to get wrong, and are handled here rather than discovered in the field: Azure mints a leaf certificate per signing request that expires in about three days. Both scripts previously retried WITHOUT a timestamp when a timestamped sign failed — under Azure that ships an artifact which verifies on the runner and goes untrusted days later, on every user's machine at once. The retry is now gated on the mode: still lenient for a .pfx whose cert outlives the release, a hard failure for Azure. The MSIX manifest Publisher must equal the signer subject byte-for-byte, because package identity is Name + Publisher. The default is now the profile's verified subject, written with `[char]0xFC` escapes rather than literal umlauts so this UTF-8-without-BOM file cannot silently mojibake the DN into one that no longer matches. pack-msix.ps1 now also reads the signature back off the packed .msix and fails on drift — asymmetric on purpose: a subject that disagrees is fatal, a subject that cannot be read is only a warning, since Get-AuthenticodeSignature's .msix support varies by Windows version and signtool has already reported success by then. NOTE this changes package identity, so existing installs need an uninstall, not an upgrade. The updater's leaf-pinning note was wrong and is corrected: update/windows.rs claimed the AUTHENTICODE_SHA256 field made Trusted Signing "a manifest edit", but a per-request leaf is exactly what a leaf pin cannot track — a pin would go stale within days and reject every release after it. Drivers are deliberately untouched: their catalogs keep the DRIVER_CERT_* cert and the installer still plants it as a machine root. The two signatures were always independent (SmartScreen/UAC vs PnP), which is why the installer could move without them. Whether a publicly-trusted catalog would let us drop that root plant is recorded as an unverified follow-up, not assumed. Verified: both scripts parse under the PowerShell 7 AST parser, both workflows are valid YAML, the evaluated Publisher default matches the subject Azure reports for the profile (86 chars, ordinal), rustfmt clean. NOT verified on Windows — the sign path itself needs an on-glass run on .133.
Punktfunk winget REST source
A self-hosted winget REST source serving the
Punktfunk Windows host, so users can install and upgrade with winget instead of downloading a
setup.exe by hand.
# one-time, from an ELEVATED prompt (winget requires admin to add a source)
winget source add -n punktfunk https://winget.punktfunk.unom.io -t Microsoft.Rest
winget install unom.PunktfunkHost # silent, wizard defaults
winget install unom.PunktfunkHost --interactive # the full wizard
winget upgrade unom.PunktfunkHost
Why self-hosted rather than the community repo
microsoft/winget-pkgs gates submissions on its Binary-Validation-Error /
Validation-Defender-Error checks, which need a publicly trusted signing cert. That blocker is
gone — the host installer is now signed with Azure Artifact Signing (see packaging/windows/README.md)
— but a self-hosted source is still the right call: it has no such gate, can carry Agreements
(verified-developers-only upstream), and can serve channels the community repo would never accept.
What it implements
The winget client consumes exactly three endpoints. The reference implementation's other twenty
(/packages/**) are its admin API for mutating a CosmosDB; a catalogue generated at release
time has nothing to mutate, so they are not implemented.
| Endpoint | Notes |
|---|---|
GET /information |
Source identity, ServerSupportedVersions, declared capabilities. |
POST /manifestSearch |
Query / Inclusions (OR) / Filters (AND) / FetchAllManifests. 204 on no match — not an empty 200. |
GET /packageManifests/{id} |
Full manifest; honours ?Version=. Identifier match is case-insensitive. |
GET /healthz |
Not part of the spec — liveness for compose. |
NormalizedPackageNameAndPublisher is deliberately declared unsupported. winget derives it
client-side from ARP entries with its own normalization (case folding plus stripping legal suffixes,
punctuation and version-like tails); claiming support would mean reimplementing that exactly, and a
near-miss silently mis-correlates an installed host. Correlation rides on ProductCode instead,
which Inno gives us exactly (<AppId>_is1).
Layout
| File | Role |
|---|---|
handler.mjs |
The three endpoints. Pure (Request, catalogue) -> Response, no I/O, no deps. |
server.mjs |
node:http front. Loads the catalogue, reloads on mtime change. |
build-data.mjs |
Generates data/data.json from the published release manifests. Build-time only. |
test.mjs |
28 checks driving handler.mjs directly. |
compose.production.yml |
The unom-1 service. |
How it runs
Gitea releases (manifest trio per stable tag)
│ build-data.mjs ← walks ALL releases, no local state [CI, Linux job]
▼
data/data.json (~8 KB) ← rsynced to unom-1 on each stable tag
▼
bun + server.mjs on unom-1:3240 ← edge Caddy TLS ← winget.punktfunk.unom.io
Stock oven/bun image with the two .mjs files bind-mounted — the same shape as the flatpak server
(stock caddy:2-alpine + a bind-mounted Caddyfile). There is no image to build, publish or pull, so
a code change deploys by scp plus docker compose up -d.
The server has zero dependencies: it reads a generated JSON file. Parsing YAML and talking to
Gitea is build-data.mjs's job, and that runs in CI. Only that build step needs node_modules.
Content and config deploy separately, matching the flatpak repo:
deploy-services.yml(wingetjob) shipscompose.production.yml+ the two.mjsfiles.windows-host.yml(winget-sourcejob) builds and shipsdata/data.jsonon each stable tag.
server.mjs reloads the catalogue when its mtime changes, so publishing a release needs no restart.
A half-written file mid-rsync keeps the previous catalogue rather than taking the source down.
Why the catalogue is derived from releases
build-data.mjs walks the Gitea releases rather than reading local files. That is what makes
winget install --version and winget upgrade work: winget resolves both against the version
list, so a source that only knew the newest release could neither pin an older version nor show an
upgrade path from one.
It also means the source holds no state — re-running the build reproduces the same data.json, so a
lost deployment is one command away and nothing can drift.
Local development
npm install
npm run build:local # data/data.json from ../*.yaml (single version — fine for dev)
npm test # 28 checks, no network, no server
npm start # http://localhost:3240
npm test is the gate that matters. A malformed response does not fail loudly — winget reports "no
package found", so a shape regression looks like a missing package rather than a broken source. CI
runs it before anything reaches the box.
Note a real winget source add will not accept a local instance: it requires HTTPS with a publicly
trusted certificate, which is what the edge vhost provides in production.
First-time setup
-
DNS —
winget.punktfunk.unom.io→ the unom-1 hcloud box, in theunom.ioCloudflare zone (DNS-only, same asdocs). ✅ done -
Caddy vhost — in
unom/infra,caddy/Caddyfile, next to the existingdocs.punktfunk.unom.ioblock. ✅ donewinget.punktfunk.unom.io { import security_headers reverse_proxy localhost:3240 }⚠ Not by hand on the box.
~/caddy/Caddyfileon unom-1 looks like the config but is a copy thatdeploy-all.shrsyncs over fromunom/infra, and there is no.gitthere to warn you. A vhost added only on the box survives until the next deploy and no longer: this one was hand-added on 2026-07-26, and the 2026-07-31 hardening commit rewrote the file from the repo's own copy and deleted it — its message says "all 7 vhosts" when there were 8 live.Until the vhost exists the hostname resolves but the TLS handshake fails — Caddy has no certificate for a name it does not serve. That is the expected state on first setup, not a broken deploy, but it is also exactly how the clobber presents later: an
internal_error(alert 80) with no peer certificate, which winget reports asWINHTTP_CALLBACK_STATUS_FLAG_SECURITY_CHANNEL_ERROR/0x8a15003band which looks like a client bug. Diagnose by SNI, not by port 80 — Caddy 308s every Host to https, including names it has never heard of, so a redirect there proves nothing:openssl s_client -connect winget.punktfunk.unom.io:443 \ -servername winget.punktfunk.unom.io </dev/null 2>&1 | grep -E '^subject=|alert' -
Run
deploy-services.ymlto place the compose file and start the container. It serves 503 until a catalogue exists — expected, and visible on/healthz. -
Publish a stable tag, or build and ship the catalogue by hand:
cd packaging/winget/server && npm install && npm run build && npm test
scp data/data.json <deploy-user>@<unom-1>:~/unom-winget/data/data.json
Then verify from anywhere:
curl -s https://winget.punktfunk.unom.io/information | jq .
curl -s -X POST https://winget.punktfunk.unom.io/manifestSearch \
-H 'content-type: application/json' -d '{"FetchAllManifests":true}' | jq '.Data[].PackageIdentifier'
Releases published before winget support shipped have no manifests attached, so
npm run buildskips them. Until the first stable tag lands with them, build with--local ..to serve the checked-in manifests, or backfill the manifest trio onto an existing release.