Add npm virtual repositories (#152 )

npm clients need one registry for internal and public packages, but npm virtuals have no merger, so a local npm repo and npmjs cannot be served together. - merge packuments across members: union versions (member order wins), first-defined dist-tags minus ones naming missing versions - serve full and abbreviated (`install-v1`) packuments, scoped names included, via the shared virtual merge cache - point tarballs at `<virtual>/<member>/<path>` and 302 them to the owning member - reject an invalid Host with 400; serve virtual and local packuments `no-store` with `Vary: Host, X-Forwarded-Proto, Accept` - return 502 when a member errors; skip members without the package - add `docs/npm/virtual.md` and a stock `node:22` e2e-docker case Reviewed-on: #152 Co-authored-by: unkin-agent <unkin-agent@unkin.net> Co-committed-by: unkin-agent <unkin-agent@unkin.net>
Harden nuget per-request index rewrite (#149 )
2026-10-17 21:52:55 +11:00 · 2026-10-17 21:39:42 +11:00 · 2026-10-17 21:23:29 +11:00 · 2026-10-17 17:18:05 +11:00 · 2026-10-17 17:17:41 +11:00 · 2026-10-17 15:54:47 +11:00
147 changed files with 4158 additions and 1017 deletions
@@ -672,7 +672,7 @@ func TestMain(m *testing.M) {

 ```hcl
 provider "artifactapi" {
-  endpoint = "https://artifactapi.k8s.syd1.au.unkin.net"
+  endpoint = "https://artifactapi.example.com"
 }

 resource "artifactapi_remote" "terraform_registry" {
@@ -20,98 +20,26 @@ API: `http://localhost:8000` | Frontend: `http://localhost:5173`

 ## Package Types

-| Type | Mutable (auto-detected) | Immutable (auto-detected) |
-|---|---|---|
-| `generic` | nothing | everything |
-| `docker` | tag manifests, `/tags/list` | blobs, digest manifests |
-| `helm` | `index.yaml` | `.tgz` charts |
-| `pypi` | `simple/*` index pages | `.whl`, `.tar.gz` |
-| `npm` | package metadata | `.tgz` tarballs |
-| `rpm` | `repomd.xml`, `repodata/*` | `.rpm` |
-| `alpine` | `APKINDEX.tar.gz` | `.apk` |
-| `puppet` | `v3/modules/*`, `v3/releases*` | `.tar.gz` |
-| `terraform` | `*/versions` | `*/download/*/*` |
-| `goproxy` | `@v/list`, `@latest` | `.info`, `.mod`, `.zip` |
-| `cargo` | sparse index files (`config.json` synthesized) | `crates/*/*.crate` |
-| `nuget` | `index.json` (service index), registration and flat container indexes | `.nupkg`, `.nuspec` |
-| `github_rpm` | `repodata/*` (synthesized) | `.rpm` (redirected) |
-
 Providers classify paths automatically. Users only configure what to proxy and TTLs.

-### `github_rpm` — GitHub releases as a yum repo (metadata-only, no precache)
+Classification precedence, TTLs and eviction for every remote: [docs/common/caching.md](docs/common/caching.md).

-A `github_rpm` remote turns a GitHub repo's **releases** into a real `dnf`/`yum`
-repository without ever caching the packages. It scans releases for `.rpm`
-assets, derives each package's metadata (NEVRA, requires/provides/conflicts/
-obsoletes, files, checksum) and **synthesizes `repodata/` on the fly**. Package
-metadata comes from a **ranged GET of just the RPM header** (the header sits at
-the front of the file, so the whole package is never downloaded); the sha256
-checksum comes from the GitHub asset `digest` when present, else a one-time
-lazy stream. Derived metadata is cached (keyed by asset) so repodata generation
-is served from primed DB rows, never a cold on-demand derive.
+Each type's paths, Terraform examples and local/virtual behaviour are in `docs/<type>/`:

-Each package's `<location>` points back at the remote, which **302-redirects**
-the download to the `releases_remote` — an existing generic `github.com` remote
-that streams the actual bytes. `dnf` follows the redirect transparently.
-
-#### Background syncer
-
-A single process-wide **background syncer** keeps every `github_rpm` remote's
-derived metadata current off the client request path:
-
- **Prime on create.** Creating a `github_rpm` remote enqueues a background prime
-  scan, so its metadata is derived right away without blocking the create call.
-  The first `dnf` request is served from cache. If a request arrives before the
-  prime lands, it returns a retryable `503` (with `Retry-After`) rather than
-  serving an empty repo or blocking on a multi-minute derive.
- **Periodic re-check, driven by `mutable_ttl`.** Each remote is re-checked for
-  new or changed releases no more often than its `mutable_ttl`. New/changed
-  assets are derived incrementally; assets already cached are never re-fetched,
-  and assets that disappear upstream are pruned.
- **ETag / 304 conditional requests.** The releases-list `ETag` is stored per
-  remote and sent as `If-None-Match`; a `304 Not Modified` means nothing changed
-  and the syncer derives nothing. GitHub does not count `304` conditional
-  responses against the rate limit, so an unchanged repo is nearly free — this is
-  the main lever keeping GitHub traffic low.
- **Global rate limit.** Every GitHub call (releases list + each ranged asset
-  header GET) passes through a single token-bucket limiter **shared across all
-  remotes**, so GitHub is never hammered. Configure a token (`password`) on the
-  remote for the higher authenticated rate limit (~5000/hr vs ~60/hr
-  unauthenticated).
- **Multi-replica coordination.** State is shared through the database. Before a
-  periodic scan a replica must atomically claim a per-remote lease
-  (`github_rpm_sync_state`: `last_synced_at`, `etag`, `sync_lease_owner`,
-  `sync_lease_expires`); only the winner scans. This bounds total GitHub load to
-  ~once per `mutable_ttl` regardless of replica count, and the shared `etag`
-  lets any replica issue the conditional request.
-
-```hcl
-# Backend that serves the actual .rpm bytes from github.com.
-resource "artifactapi_remote_generic" "github" {
-  name     = "github"
-  base_url = "https://github.com"
-  patterns = [
-    "acme/tools/releases/download/.*\\.rpm$", # allowlist the repo's release assets
-  ]
-}
-
-resource "artifactapi_remote_github_rpm" "acme-tools" {
-  name            = "acme-tools"
-  base_url        = "https://api.github.com/repos/acme/tools" # the releases API root
-  releases_remote = "github"                                  # backend for downloads
-  mutable_ttl     = 3600                                      # release re-scan interval
-
-  # Optional: restrict which release assets become packages (regex on filename).
-  patterns = [".*\\.x86_64\\.rpm$", ".*\\.noarch\\.rpm$"]
-
-  # Optional: a token for private repos / higher API rate limits.
-  # password = "ghp_..."
-}
-```
-
-`dnf` config: `baseurl=https://artifactapi.example/api/v1/remote/acme-tools`.
-The repo is multi-arch (no `$basearch` needed) — `dnf` selects matching packages
-from the synthesized metadata.
+- `alpine`: [remote](docs/alpine/remote.md), [remote-github](docs/alpine/remote-github.md)
+- `cargo`: [remote](docs/cargo/remote.md)
+- `deb`: [remote](docs/deb/remote.md), [remote-github](docs/deb/remote-github.md)
+- `docker`: [remote](docs/docker/remote.md), [local](docs/docker/local.md)
+- `generic`: [remote](docs/generic/remote.md)
+- `gems`: [remote](docs/gems/remote.md)
+- `goproxy`: [remote](docs/goproxy/remote.md)
+- `helm`: [remote](docs/helm/remote.md), [local](docs/helm/local.md), [virtual](docs/helm/virtual.md)
+- `npm`: [remote](docs/npm/remote.md), [local](docs/npm/local.md), [virtual](docs/npm/virtual.md)
+- `nuget`: [remote](docs/nuget/remote.md)
+- `puppet`: [remote](docs/puppet/remote.md), [local](docs/puppet/local.md)
+- `pypi`: [remote](docs/pypi/remote.md)
+- `rpm`: [remote](docs/rpm/remote.md), [remote-github](docs/rpm/remote-github.md)
+- `terraform`: [remote](docs/terraform/remote.md), [local](docs/terraform/local.md)

 ### GitHub authentication

@@ -180,136 +108,9 @@ GITHUB_APP_PRIVATE_KEY_PATH=/etc/artifactapi/github-app.pem

 ## Terraform

-Remotes and virtuals are managed by Terraform. Each package type has its own resource:
+Remotes and virtuals are managed by Terraform. Each package type has its own resource; see `docs/<type>/` for examples.

-```hcl
-resource "artifactapi_remote_generic" "github" {
-  name     = "github"
-  base_url = "https://github.com"
-
-  immutable_ttl = 0
-  mutable_ttl   = 7200
-
-  patterns = [
-    "ducaale/xh/.*/xh-.*-x86_64-unknown-linux-musl.tar.gz$",
-    "mikefarah/yq/.*/yq_linux_amd64$",
-  ]
-
-  mutable_patterns = [
-    ".*/archive/refs/heads/.*\\.tar\\.gz$",
-  ]
-}
-
-resource "artifactapi_remote_docker" "dockerhub" {
-  name     = "dockerhub"
-  base_url = "https://registry-1.docker.io"
-
-  immutable_ttl    = 0
-  mutable_ttl      = 300
-  ban_tags_enabled = true
-  ban_tags         = ["latest"]
-
-  patterns = [
-    "^library/postgres",
-    "^library/redis",
-  ]
-}
-
-resource "artifactapi_remote_helm" "jetstack" {
-  name     = "jetstack"
-  base_url = "https://charts.jetstack.io"
-
-  immutable_ttl = 0
-  mutable_ttl   = 3600
-}
-
-resource "artifactapi_virtual" "helm" {
-  name         = "helm"
-  package_type = "helm"
-  members      = [artifactapi_remote_helm.jetstack.name]
-}
-```
-
-Provider: [terraform-provider-artifactapi](../terraform-provider-artifactapi)
-
-### Serving providers as a registry
-
-A local `terraform` repo is a real provider registry: upload
-`terraform-provider-{type}_{version}_{os}_{arch}.zip` files under
-`{namespace}/{type}/`, and Terraform installs them from a bare source address —
-no `.terraformrc` mirror config:
-
-```hcl
-terraform {
-  required_providers {
-    artifactapi = {
-      source  = "artifactapi.k8s.syd1.au.unkin.net/<repo>/<type>"
-      version = "0.1.2"
-    }
-  }
-}
-```
-
-The Terraform *namespace* segment is the artifactapi repo name; the provider is
-matched by *type*. The registry serves service discovery
-(`/.well-known/terraform.json`), the `providers.v1` version/download endpoints,
-and a GPG-signed `SHA256SUMS` per the provider registry protocol.
-
-Signing needs a GPG key. By default artifactapi generates one on first start and
-stores it in the database (`signing_keys` table), so every replica shares it and
-there's nothing to provision. To bring your own key instead, point
-`TF_SIGNING_KEY_PATH` at an armored private key (optionally
-`TF_SIGNING_KEY_PASSPHRASE`), which takes precedence over the generated one.
-`TF_PROVIDER_PROTOCOLS` (default `5.0,6.0`) sets the advertised plugin protocols.
-
-### Local helm repository
-
-A local `helm` repo is a real chart repository: `PUT` a packaged chart to
-`/api/v2/remotes/<repo>/files/<chart>-<version>.tgz` and `index.yaml` is
-generated from each chart's `Chart.yaml` (with sha256 digests) at
-`/api/v1/local/<repo>/index.yaml`; a file not named after its `Chart.yaml`
-name and version is stored but left out of the index. Add the local as a
-member of a helm virtual to serve it alongside remotes.
-
-```bash
-helm package ./mychart
-curl -fsSL --upload-file mychart-0.1.0.tgz \
-  https://artifactapi.k8s.syd1.au.unkin.net/api/v2/remotes/<repo>/files/mychart-0.1.0.tgz
-helm repo add <repo> https://artifactapi.k8s.syd1.au.unkin.net/api/v1/local/<repo>
-```
-
-### Local puppet forge
-
-A local `puppet` repo is a Forge: `PUT` a module tarball to
-`/api/v2/remotes/<repo>/files/<owner>-<module>-<version>.tar.gz` and the Forge
-v3 API (`/v3/modules`, `/v3/releases`, `/v3/files`) is served from each
-release's `metadata.json` at `/api/v1/local/<repo>`. Versions are immutable.
-
-```bash
-puppet module build ./hello
-curl -fsSL --upload-file pkg/unkin-hello-0.1.0.tar.gz \
-  https://artifactapi.k8s.syd1.au.unkin.net/api/v2/remotes/<repo>/files/unkin-hello-0.1.0.tar.gz
-puppet module install unkin-hello \
-  --module_repository https://artifactapi.k8s.syd1.au.unkin.net/api/v1/local/<repo>
-```
-
-### Local docker registry
-
-A local `docker` repo is a real container registry, not a mirror: it serves the
-Docker Registry HTTP API V2 for both push and pull, so any client (`docker`,
-`podman`, `skopeo`, `buildah`) can use it directly.
-
-```sh
-docker tag myapp:latest artifactapi.k8s.syd1.au.unkin.net/docker-internal/myapp:latest
-docker push        artifactapi.k8s.syd1.au.unkin.net/docker-internal/myapp:latest
-docker pull        artifactapi.k8s.syd1.au.unkin.net/docker-internal/myapp:latest
-```
-
-The first path segment after `/v2/` is the artifactapi repo name; the remainder
-is the image name. Blobs and manifests are stored through the shared
-content-addressable store (deduplicated by digest, reaped by GC once
-unreferenced); tags are mutable references and re-pushing a tag moves it. Blob
-uploads support both the monolithic and chunked (`POST`/`PATCH`/`PUT`) flows.
+Provider: [terraform-provider-artifactapi](https://git.unkin.net/unkin/terraform-provider-artifactapi)

 ## Access Control

@@ -329,6 +130,7 @@ No patterns + no blocklist = open proxy. Provider handles mutability classificat
 ```
 GET /api/v1/remote/{name}/{path}     Proxy/cache artifact
 GET /api/v1/virtual/{name}/{path}    Virtual repo (merged index)
+PUT /api/v1/local/{name}/{package}   npm publish (local npm repos)
 GET /v2/{name}/{path}                Docker Registry v2
 ```

@@ -0,0 +1,7 @@
+# `github_alpine` remote
+
+Shared classification, TTL and eviction rules: [caching](../common/caching.md).
+
+| Type | Mutable (auto-detected) | Immutable (auto-detected) |
+|---|---|---|
+| `github_alpine` | `APKINDEX.tar.gz` (synthesized) | `.apk` (redirected) |
@@ -0,0 +1,7 @@
+# `alpine` remote
+
+Shared classification, TTL and eviction rules: [caching](../common/caching.md).
+
+| Type | Mutable (auto-detected) | Immutable (auto-detected) |
+|---|---|---|
+| `alpine` | `APKINDEX.tar.gz` | `.apk` |
@@ -0,0 +1,7 @@
+# `cargo` remote
+
+Shared classification, TTL and eviction rules: [caching](../common/caching.md).
+
+| Type | Mutable (auto-detected) | Immutable (auto-detected) |
+|---|---|---|
+| `cargo` | sparse index files (`config.json` synthesized) | `crates/*/*.crate` |
@@ -0,0 +1,101 @@
+# Caching and classification (all remotes)
+
+Every proxied request on a remote is classified as **denied**, **immutable** or
+**mutable**. The class picks the storage path and the freshness TTL.
+
+## Fields
+
+| Field | Meaning |
+|---|---|
+| `patterns` | Allowlist of regexes. Empty allows every path; otherwise a non-matching path is denied (403). |
+| `blocklist` | Regexes; a match is denied (403). |
+| `immutable_patterns` | Regexes forcing a path immutable. |
+| `mutable_patterns` | Regexes forcing a path mutable. |
+| `immutable_ttl` | Seconds an immutable path stays fresh. `0` = forever. |
+| `mutable_ttl` | Seconds a mutable path stays fresh. `0` also means never expires. Omitted on create, it defaults to `3600`; omitted on update, the stored value is kept. |
+| `check_mutable` | When a mutable path's `mutable_ttl` expires, sends a conditional `HEAD` (`If-None-Match`) with the stored upstream ETag. A `304` re-arms the TTL and serves the stored copy; anything else refetches in full. The ETag is kept for `max(10 × mutable_ttl, 7 days)`, re-armed on each `304`, so idle paths age out. It is only stored when `check_mutable` is set and `mutable_ttl` is non-zero, and never for immutable paths. |
+| `stale_on_error` | On an upstream network error, serve the stored copy instead of failing. |
+| `quarantine_enabled` / `quarantine_days` | Stored on the remote but not enforced by the proxy yet. |
+| `mirrorlist` / `mirror_strategy` | rpm/deb/alpine only. Upstream pool is `[base_url] + mirrorlist`; `round_robin` (default) rotates the starting mirror, `least_conn` starts at the one with the fewest in-flight requests. Network errors and 5xx fail over to the next mirror; other statuses return as-is. |
+
+## Precedence
+
+The first rule that matches wins:
+
+1. `blocklist` matches: **denied**
+2. `patterns` is non-empty and nothing matches: **denied**
+3. `immutable_patterns` matches: **immutable**
+4. `mutable_patterns` matches: **mutable**
+5. the provider's built-in `Classify` says mutable: **mutable**
+6. anything else: **immutable**
+
+So `blocklist` beats the allowlist, and `immutable_patterns` beats
+`mutable_patterns`. Either user list overrides the built-in rules. `GET` and
+`HEAD` both apply this order.
+
+## Matching rules
+
+- Regexes are Go `regexp`, matched **unanchored** (`MatchString`). Add `^` or `$` yourself when you need them.
+- They match the path after `/api/v1/remote/<name>/`, without a leading slash, e.g. `9/BaseOS/x86_64/os/repodata/repomd.xml`.
+- The query string is not part of the path. It is dropped and not forwarded upstream.
+- `create` and `update` reject an invalid regex in `patterns`, `blocklist`, `mutable_patterns`, `immutable_patterns` or `ban_tags` with a 400. Only invalid regexes already stored before that check are skipped at match time.
+
+## Lifecycles
+
+| | Immutable | Mutable |
+|---|---|---|
+| Stored as | content-addressed blob (sha256), streamed | per-path index object, buffered and rewritten by the provider |
+| Fresh for | `immutable_ttl` (`0` = forever) | `mutable_ttl` |
+| After expiry | refetched from upstream | revalidated with `If-None-Match` when `check_mutable` is set and an ETag is stored, else refetched |
+| Upstream down | `stale_on_error` serves the stored copy and re-arms the TTL; an open circuit breaker always serves stale if a copy exists, else 503 | same |
+
+Concurrent misses on one path share a fetch lock. Waiters poll the store for
+up to 5s instead of all going upstream.
+
+## Eviction
+
+```sh
+curl -X DELETE https://artifactapi.example.com/api/v2/remotes/<name>/objects/<path>
+curl -X DELETE https://artifactapi.example.com/api/v2/remotes/<name>/objects/<dir>/*
+```
+
+- Evict drops the artifact row, the index object and the Redis TTL and ETag keys, so the next request goes upstream whatever the class or TTL.
+- Deleting a remote drops its Redis TTL and ETag keys, so a re-created remote of the same name starts cold.
+- A single-path evict takes the fetch lock first, so an in-flight fetch can't re-cache the path. It returns 503 if the lock stays busy.
+- A wildcard must end in `<dir>/*` and takes no lock.
+- Returns 204, or 404 for an unknown remote.
+
+## Examples
+
+Bodies for `POST https://artifactapi.example.com/api/v2/remotes`.
+
+GitHub release assets through a `generic` remote, allowlisted:
+
+```json
+{"name": "github", "package_type": "generic", "base_url": "https://github.com",
+ "immutable_ttl": 0, "mutable_ttl": 7200,
+ "patterns": ["^mikefarah/yq/releases/download/.*/yq_linux_amd64$"]}
+```
+
+rpm remote: `repodata/*` and `repomd.xml` are mutable by default, so only the TTLs are needed:
+
+```json
+{"name": "almalinux", "package_type": "rpm", "base_url": "https://repo.almalinux.org/almalinux",
+ "immutable_ttl": 0, "mutable_ttl": 3600, "check_mutable": true, "stale_on_error": true}
+```
+
+Force a moving path mutable, e.g. branch tarballs on `generic`:
+
+```json
+{"mutable_patterns": ["/archive/refs/heads/.*\\.tar\\.gz$"], "mutable_ttl": 600}
+```
+
+## Built-in classification per type
+
+Each type's page lists its auto-detected mutable and immutable paths:
+[alpine](../alpine/remote.md), [alpine github](../alpine/remote-github.md),
+[cargo](../cargo/remote.md), [deb](../deb/remote.md), [deb github](../deb/remote-github.md),
+[docker](../docker/remote.md), [generic](../generic/remote.md), [gems](../gems/remote.md), [goproxy](../goproxy/remote.md),
+[helm](../helm/remote.md), [npm](../npm/remote.md), [nuget](../nuget/remote.md),
+[puppet](../puppet/remote.md), [pypi](../pypi/remote.md), [rpm](../rpm/remote.md),
+[rpm github](../rpm/remote-github.md), [terraform](../terraform/remote.md).
@@ -0,0 +1,7 @@
+# `github_deb` remote
+
+Shared classification, TTL and eviction rules: [caching](../common/caching.md).
+
+| Type | Mutable (auto-detected) | Immutable (auto-detected) |
+|---|---|---|
+| `github_deb` | `Packages*`, `Release*` (synthesized) | `.deb` (redirected) |
@@ -0,0 +1,7 @@
+# `deb` remote
+
+Shared classification, TTL and eviction rules: [caching](../common/caching.md).
+
+| Type | Mutable (auto-detected) | Immutable (auto-detected) |
+|---|---|---|
+| `deb` | `Packages*`, `Sources*`, `Release`, `InRelease`, `Release.gpg`, `Contents-*` | `.deb` |
@@ -0,0 +1,17 @@
+# Local docker registry
+
+A local `docker` repo is a real container registry, not a mirror: it serves the
+Docker Registry HTTP API V2 for both push and pull, so any client (`docker`,
+`podman`, `skopeo`, `buildah`) can use it directly.
+
+```sh
+docker tag myapp:latest artifactapi.example.com/docker-internal/myapp:latest
+docker push        artifactapi.example.com/docker-internal/myapp:latest
+docker pull        artifactapi.example.com/docker-internal/myapp:latest
+```
+
+The first path segment after `/v2/` is the artifactapi repo name; the remainder
+is the image name. Blobs and manifests are stored through the shared
+content-addressable store (deduplicated by digest, reaped by GC once
+unreferenced); tags are mutable references and re-pushing a tag moves it. Blob
+uploads support both the monolithic and chunked (`POST`/`PATCH`/`PUT`) flows.
@@ -0,0 +1,26 @@
+# `docker` remote
+
+Shared classification, TTL and eviction rules: [caching](../common/caching.md).
+
+| Type | Mutable (auto-detected) | Immutable (auto-detected) |
+|---|---|---|
+| `docker` | tag manifests, `/tags/list` | blobs, digest manifests |
+
+## Terraform
+
+```hcl
+resource "artifactapi_remote_docker" "dockerhub" {
+  name     = "dockerhub"
+  base_url = "https://registry-1.docker.io"
+
+  immutable_ttl    = 0
+  mutable_ttl      = 300
+  ban_tags_enabled = true
+  ban_tags         = ["latest"]
+
+  patterns = [
+    "^library/postgres",
+    "^library/redis",
+  ]
+}
+```
@@ -0,0 +1,7 @@
+# `gems` remote
+
+Shared classification, TTL and eviction rules: [caching](../common/caching.md).
+
+| Type | Mutable (auto-detected) | Immutable (auto-detected) |
+|---|---|---|
+| `gems` | compact index (`versions`, `info/*`, `names`), `*specs.4.8.gz` | `gems/*.gem`, `quick/Marshal.4.8/*.gemspec.rz` |
@@ -0,0 +1,28 @@
+# `generic` remote
+
+Shared classification, TTL and eviction rules: [caching](../common/caching.md).
+
+| Type | Mutable (auto-detected) | Immutable (auto-detected) |
+|---|---|---|
+| `generic` | nothing | everything |
+
+## Terraform
+
+```hcl
+resource "artifactapi_remote_generic" "github" {
+  name     = "github"
+  base_url = "https://github.com"
+
+  immutable_ttl = 0
+  mutable_ttl   = 7200
+
+  patterns = [
+    "ducaale/xh/.*/xh-.*-x86_64-unknown-linux-musl.tar.gz$",
+    "mikefarah/yq/.*/yq_linux_amd64$",
+  ]
+
+  mutable_patterns = [
+    ".*/archive/refs/heads/.*\\.tar\\.gz$",
+  ]
+}
+```
@@ -0,0 +1,7 @@
+# `goproxy` remote
+
+Shared classification, TTL and eviction rules: [caching](../common/caching.md).
+
+| Type | Mutable (auto-detected) | Immutable (auto-detected) |
+|---|---|---|
+| `goproxy` | `@v/list`, `@latest` | `.info`, `.mod`, `.zip` |
@@ -0,0 +1,15 @@
+# Local helm repository
+
+A local `helm` repo is a real chart repository: `PUT` a packaged chart to
+`/api/v2/remotes/<repo>/files/<chart>-<version>.tgz` and `index.yaml` is
+generated from each chart's `Chart.yaml` (with sha256 digests) at
+`/api/v1/local/<repo>/index.yaml`; a file not named after its `Chart.yaml`
+name and version is stored but left out of the index. Add the local as a
+member of a helm virtual to serve it alongside remotes.
+
+```bash
+helm package ./mychart
+curl -fsSL --upload-file mychart-0.1.0.tgz \
+  https://artifactapi.example.com/api/v2/remotes/<repo>/files/mychart-0.1.0.tgz
+helm repo add <repo> https://artifactapi.example.com/api/v1/local/<repo>
+```
@@ -0,0 +1,19 @@
+# `helm` remote
+
+Shared classification, TTL and eviction rules: [caching](../common/caching.md).
+
+| Type | Mutable (auto-detected) | Immutable (auto-detected) |
+|---|---|---|
+| `helm` | `index.yaml` | `.tgz` charts |
+
+## Terraform
+
+```hcl
+resource "artifactapi_remote_helm" "jetstack" {
+  name     = "jetstack"
+  base_url = "https://charts.jetstack.io"
+
+  immutable_ttl = 0
+  mutable_ttl   = 3600
+}
+```
@@ -0,0 +1,9 @@
+# `helm` virtual
+
+```hcl
+resource "artifactapi_virtual" "helm" {
+  name         = "helm"
+  package_type = "helm"
+  members      = [artifactapi_remote_helm.jetstack.name]
+}
+```
@@ -0,0 +1,14 @@
+# Local npm registry
+
+A local `npm` repo is a real npm registry: `npm publish` (scoped and unscoped)
+and `npm install` work against `/api/v1/local/<repo>/`. npm requires an
+`_authToken` for the registry before it will publish; any value works.
+
+```ini
+# .npmrc
+registry=https://artifactapi.example.com/api/v1/local/<repo>/
+//artifactapi.example.com/api/v1/local/<repo>/:_authToken=unused
+```
+
+Versions are immutable: republishing an existing version is rejected. The
+dist-tags sent with a publish (`latest`, or `--tag`) move to that version.
@@ -0,0 +1,7 @@
+# `npm` remote
+
+Shared classification, TTL and eviction rules: [caching](../common/caching.md).
+
+| Type | Mutable (auto-detected) | Immutable (auto-detected) |
+|---|---|---|
+| `npm` | package metadata | `.tgz` tarballs |
@@ -0,0 +1,30 @@
+# `npm` virtual
+
+An npm virtual serves one registry over local and remote npm members:
+`npm install` and `npm view` work against `/api/v1/virtual/<name>/`.
+
+- Versions are the union across members; on a duplicate version the earlier
+  member wins.
+- Each dist-tag comes from the first member defining it, and is dropped if it
+  names a version missing from the merged set.
+- Tarball URLs point at the virtual (`<member>/<path>`), which 302s to the
+  owning member.
+- Full and abbreviated (`application/vnd.npm.install-v1+json`) packuments are
+  both served.
+- A member error fails the request with 502; a member without the package is
+  skipped. Merges are cached for 60s per package.
+
+```hcl
+resource "artifactapi_virtual" "npm" {
+  name         = "npm"
+  package_type = "npm"
+  members      = [artifactapi_local_npm.internal.name, artifactapi_remote_npm.npmjs.name]
+}
+```
+
+```ini
+# .npmrc
+registry=https://artifactapi.example.com/api/v1/virtual/npm/
+```
+
+Publish to the local member directly; the virtual is read-only.
@@ -0,0 +1,7 @@
+# `nuget` remote
+
+Shared classification, TTL and eviction rules: [caching](../common/caching.md).
+
+| Type | Mutable (auto-detected) | Immutable (auto-detected) |
+|---|---|---|
+| `nuget` | `index.json` (service index), registration and flat container indexes | `.nupkg`, `.snupkg`, `.nuspec`, `readme`, `icon` |
@@ -0,0 +1,14 @@
+# Local puppet forge
+
+A local `puppet` repo is a Forge: `PUT` a module tarball to
+`/api/v2/remotes/<repo>/files/<owner>-<module>-<version>.tar.gz` and the Forge
+v3 API (`/v3/modules`, `/v3/releases`, `/v3/files`) is served from each
+release's `metadata.json` at `/api/v1/local/<repo>`. Versions are immutable.
+
+```bash
+puppet module build ./hello
+curl -fsSL --upload-file pkg/unkin-hello-0.1.0.tar.gz \
+  https://artifactapi.example.com/api/v2/remotes/<repo>/files/unkin-hello-0.1.0.tar.gz
+puppet module install unkin-hello \
+  --module_repository https://artifactapi.example.com/api/v1/local/<repo>
+```
@@ -0,0 +1,7 @@
+# `puppet` remote
+
+Shared classification, TTL and eviction rules: [caching](../common/caching.md).
+
+| Type | Mutable (auto-detected) | Immutable (auto-detected) |
+|---|---|---|
+| `puppet` | `v3/modules/*`, `v3/releases*` | `.tar.gz` |
@@ -0,0 +1,7 @@
+# `pypi` remote
+
+Shared classification, TTL and eviction rules: [caching](../common/caching.md).
+
+| Type | Mutable (auto-detected) | Immutable (auto-detected) |
+|---|---|---|
+| `pypi` | `simple/*` index pages | `.whl`, `.tar.gz` |
@@ -0,0 +1,82 @@
+# `github_rpm` remote
+
+Shared classification, TTL and eviction rules: [caching](../common/caching.md).
+
+| Type | Mutable (auto-detected) | Immutable (auto-detected) |
+|---|---|---|
+| `github_rpm` | `repodata/*` (synthesized) | `.rpm` (redirected) |
+
+## GitHub releases as a yum repo (metadata-only, no precache)
+
+A `github_rpm` remote turns a GitHub repo's **releases** into a real `dnf`/`yum`
+repository without ever caching the packages. It scans releases for `.rpm`
+assets, derives each package's metadata (NEVRA, requires/provides/conflicts/
+obsoletes, files, checksum) and **synthesizes `repodata/` on the fly**. Package
+metadata comes from a **ranged GET of just the RPM header** (the header sits at
+the front of the file, so the whole package is never downloaded); the sha256
+checksum comes from the GitHub asset `digest` when present, else a one-time
+lazy stream. Derived metadata is cached (keyed by asset) so repodata generation
+is served from primed DB rows, never a cold on-demand derive.
+
+Each package's `<location>` points back at the remote, which **302-redirects**
+the download to the `releases_remote` — an existing generic `github.com` remote
+that streams the actual bytes. `dnf` follows the redirect transparently.
+
+### Background syncer
+
+A single process-wide **background syncer** keeps every `github_rpm` remote's
+derived metadata current off the client request path:
+
+- **Prime on create.** Creating a `github_rpm` remote enqueues a background prime
+  scan, so its metadata is derived right away without blocking the create call.
+  The first `dnf` request is served from cache. If a request arrives before the
+  prime lands, it returns a retryable `503` (with `Retry-After`) rather than
+  serving an empty repo or blocking on a multi-minute derive.
+- **Periodic re-check, driven by `mutable_ttl`.** Each remote is re-checked for
+  new or changed releases no more often than its `mutable_ttl`. New/changed
+  assets are derived incrementally; assets already cached are never re-fetched,
+  and assets that disappear upstream are pruned.
+- **ETag / 304 conditional requests.** The releases-list `ETag` is stored per
+  remote and sent as `If-None-Match`; a `304 Not Modified` means nothing changed
+  and the syncer derives nothing. GitHub does not count `304` conditional
+  responses against the rate limit, so an unchanged repo is nearly free — this is
+  the main lever keeping GitHub traffic low.
+- **Global rate limit.** Every GitHub call (releases list + each ranged asset
+  header GET) passes through a single token-bucket limiter **shared across all
+  remotes**, so GitHub is never hammered. Configure a token (`password`) on the
+  remote for the higher authenticated rate limit (~5000/hr vs ~60/hr
+  unauthenticated).
+- **Multi-replica coordination.** State is shared through the database. Before a
+  periodic scan a replica must atomically claim a per-remote lease
+  (`github_rpm_sync_state`: `last_synced_at`, `etag`, `sync_lease_owner`,
+  `sync_lease_expires`); only the winner scans. This bounds total GitHub load to
+  ~once per `mutable_ttl` regardless of replica count, and the shared `etag`
+  lets any replica issue the conditional request.
+
+```hcl
+# Backend that serves the actual .rpm bytes from github.com.
+resource "artifactapi_remote_generic" "github" {
+  name     = "github"
+  base_url = "https://github.com"
+  patterns = [
+    "acme/tools/releases/download/.*\\.rpm$", # allowlist the repo's release assets
+  ]
+}
+
+resource "artifactapi_remote_github_rpm" "acme-tools" {
+  name            = "acme-tools"
+  base_url        = "https://api.github.com/repos/acme/tools" # the releases API root
+  releases_remote = "github"                                  # backend for downloads
+  mutable_ttl     = 3600                                      # release re-scan interval
+
+  # Optional: restrict which release assets become packages (regex on filename).
+  patterns = [".*\\.x86_64\\.rpm$", ".*\\.noarch\\.rpm$"]
+
+  # Optional: a token for private repos / higher API rate limits.
+  # password = "ghp_..."
+}
+```
+
+`dnf` config: `baseurl=https://artifactapi.example/api/v1/remote/acme-tools`.
+The repo is multi-arch (no `$basearch` needed) — `dnf` selects matching packages
+from the synthesized metadata.
@@ -0,0 +1,7 @@
+# `rpm` remote
+
+Shared classification, TTL and eviction rules: [caching](../common/caching.md).
+
+| Type | Mutable (auto-detected) | Immutable (auto-detected) |
+|---|---|---|
+| `rpm` | `repomd.xml`, `repodata/*` | `.rpm` |
@@ -0,0 +1,29 @@
+# Serving providers as a registry
+
+A local `terraform` repo is a real provider registry: upload
+`terraform-provider-{type}_{version}_{os}_{arch}.zip` files under
+`{namespace}/{type}/`, and Terraform installs them from a bare source address —
+no `.terraformrc` mirror config:
+
+```hcl
+terraform {
+  required_providers {
+    artifactapi = {
+      source  = "artifactapi.example.com/<repo>/<type>"
+      version = "0.1.2"
+    }
+  }
+}
+```
+
+The Terraform *namespace* segment is the artifactapi repo name; the provider is
+matched by *type*. The registry serves service discovery
+(`/.well-known/terraform.json`), the `providers.v1` version/download endpoints,
+and a GPG-signed `SHA256SUMS` per the provider registry protocol.
+
+Signing needs a GPG key. By default artifactapi generates one on first start and
+stores it in the database (`signing_keys` table), so every replica shares it and
+there's nothing to provision. To bring your own key instead, point
+`TF_SIGNING_KEY_PATH` at an armored private key (optionally
+`TF_SIGNING_KEY_PASSPHRASE`), which takes precedence over the generated one.
+`TF_PROVIDER_PROTOCOLS` (default `5.0,6.0`) sets the advertised plugin protocols.
@@ -0,0 +1,7 @@
+# `terraform` remote
+
+Shared classification, TTL and eviction rules: [caching](../common/caching.md).
+
+| Type | Mutable (auto-detected) | Immutable (auto-detected) |
+|---|---|---|
+| `terraform` | `*/versions` | `*/download/*/*` |
@@ -22,15 +22,17 @@ already-running stack.
 ## Coverage

 - **Repository lifecycle** — add / change / delete for remote, local and virtual repos.
- **Caching** — one immutable artifact per remote package type (generic, docker,
-  helm, pypi, npm, rpm, alpine, puppet, terraform, goproxy, cargo, nuget) proxied through the
-  mock upstream: first fetch `X-Artifact-Source: remote`, second `cache`, bytes
+- **Caching** — one immutable artifact per remote package type (one
+  `caching_<type>_test.go` case each) proxied through the mock upstream: first fetch `X-Artifact-Source: remote`, second `cache`, bytes
  verified against the origin fixture.
 - **Local uploads** — puppet (Forge v3 API via real `r10k` and `puppet module install`),
  generic (upload/download), pypi (wheel + generated `simple/`
-  index), rpm (real package + **automatic repodata** generation).
+  index), rpm (real package + **automatic repodata** generation), npm (stock
+  `node:22` client: `npm publish` unscoped + scoped, `npm install` + `require()`,
+  second version, republish rejected).
 - **Virtual repositories** — pypi simple-index merge and helm `index.yaml` merge
-  across two members.
+  across two members; npm packument merge over two locals via a stock `node:22`
+  client (merged versions, install through the member redirect).
 - **Mirrorlist** — an rpm remote with a `mirrorlist` of extra upstream mirrors
  (pool = `base_url` + `mirrorlist`): round-robin distribution across both mirrors
  (constant-body `mockupstreama` / `mockupstreamb`), failover past a dead primary,
@@ -0,0 +1,7 @@
+//go:build dockere2e
+
+package e2edocker
+
+func init() {
+	cachingCases = append(cachingCases, cachingCase{"alpine", "alpine/x86_64/testpkg-1.0-r0.apk", "alpine/x86_64/testpkg-1.0-r0.apk"})
+}
@@ -0,0 +1,7 @@
+//go:build dockere2e
+
+package e2edocker
+
+func init() {
+	cachingCases = append(cachingCases, cachingCase{"cargo", "crates/mycrate/mycrate-1.0.0.crate", "crates/mycrate/mycrate-1.0.0.crate"})
+}
@@ -0,0 +1,7 @@
+//go:build dockere2e
+
+package e2edocker
+
+func init() {
+	cachingCases = append(cachingCases, cachingCase{"docker", "library/testimg/blobs/blobdata", "v2/library/testimg/blobs/blobdata"})
+}
@@ -0,0 +1,7 @@
+//go:build dockere2e
+
+package e2edocker
+
+func init() {
+	cachingCases = append(cachingCases, cachingCase{"gems", "gems/mygem-1.0.0.gem", "gems/mygem-1.0.0.gem"})
+}
@@ -0,0 +1,7 @@
+//go:build dockere2e
+
+package e2edocker
+
+func init() {
+	cachingCases = append(cachingCases, cachingCase{"generic", "blobs/hello.bin", "blobs/hello.bin"})
+}
@@ -0,0 +1,7 @@
+//go:build dockere2e
+
+package e2edocker
+
+func init() {
+	cachingCases = append(cachingCases, cachingCase{"goproxy", "goproxy/example.com/mod/@v/v1.0.0.zip", "goproxy/example.com/mod/@v/v1.0.0.zip"})
+}
@@ -0,0 +1,7 @@
+//go:build dockere2e
+
+package e2edocker
+
+func init() {
+	cachingCases = append(cachingCases, cachingCase{"helm", "charts/mychart-1.0.0.tgz", "charts/mychart-1.0.0.tgz"})
+}
@@ -0,0 +1,7 @@
+//go:build dockere2e
+
+package e2edocker
+
+func init() {
+	cachingCases = append(cachingCases, cachingCase{"npm", "mypkg/-/mypkg-1.0.0.tgz", "mypkg/-/mypkg-1.0.0.tgz"})
+}
@@ -0,0 +1,7 @@
+//go:build dockere2e
+
+package e2edocker
+
+func init() {
+	cachingCases = append(cachingCases, cachingCase{"nuget", "nuget/e2epkg/1.0.0/e2epkg.1.0.0.nupkg", "nuget/e2epkg/1.0.0/e2epkg.1.0.0.nupkg"})
+}
@@ -0,0 +1,7 @@
+//go:build dockere2e
+
+package e2edocker
+
+func init() {
+	cachingCases = append(cachingCases, cachingCase{"puppet", "puppet-releases/author-mod-1.0.0.tar.gz", "puppet-releases/author-mod-1.0.0.tar.gz"})
+}
@@ -0,0 +1,7 @@
+//go:build dockere2e
+
+package e2edocker
+
+func init() {
+	cachingCases = append(cachingCases, cachingCase{"pypi", "packages/foo-1.0-py3-none-any.whl", "packages/foo-1.0-py3-none-any.whl"})
+}
@@ -0,0 +1,7 @@
+//go:build dockere2e
+
+package e2edocker
+
+func init() {
+	cachingCases = append(cachingCases, cachingCase{"rpm", "rpmrepo/Packages/e2e-testpkg-1.0-1.noarch.rpm", "rpmrepo/Packages/e2e-testpkg-1.0-1.noarch.rpm"})
+}
@@ -0,0 +1,7 @@
+//go:build dockere2e
+
+package e2edocker
+
+func init() {
+	cachingCases = append(cachingCases, cachingCase{"terraform", "hashicorp/aws/download/pkg.zip", "v1/providers/hashicorp/aws/download/pkg.zip"})
+}
@@ -9,33 +9,23 @@ import (
 	"testing"
 )

+type cachingCase struct {
+	pkgType string
+	// path is the request path under /api/v1/remote/<name>/. The provider
+	// derives the upstream URL from it (docker prepends /v2/, terraform
+	// prepends /v1/providers/), and the fixture lives at that resolved path.
+	path    string
+	fixture string
+}
+
+// cachingCases is filled by one caching_<type>_test.go file per package type.
+var cachingCases []cachingCase
+
 // TestCachingPerProvider proxies one immutable artifact for every remote
 // package type through the mock upstream and asserts: first fetch is served
 // from the remote, the second from cache, and the bytes match the origin.
 func TestCachingPerProvider(t *testing.T) {
-	cases := []struct {
-		pkgType string
-		// path is the request path under /api/v1/remote/<name>/. The provider
-		// derives the upstream URL from it (docker prepends /v2/, terraform
-		// prepends /v1/providers/), and the fixture lives at that resolved path.
-		path    string
-		fixture string
-	}{
-		{"generic", "blobs/hello.bin", "blobs/hello.bin"},
-		{"npm", "mypkg/-/mypkg-1.0.0.tgz", "mypkg/-/mypkg-1.0.0.tgz"},
-		{"helm", "charts/mychart-1.0.0.tgz", "charts/mychart-1.0.0.tgz"},
-		{"pypi", "packages/foo-1.0-py3-none-any.whl", "packages/foo-1.0-py3-none-any.whl"},
-		{"rpm", "rpmrepo/Packages/e2e-testpkg-1.0-1.noarch.rpm", "rpmrepo/Packages/e2e-testpkg-1.0-1.noarch.rpm"},
-		{"alpine", "alpine/x86_64/testpkg-1.0-r0.apk", "alpine/x86_64/testpkg-1.0-r0.apk"},
-		{"puppet", "puppet-releases/author-mod-1.0.0.tar.gz", "puppet-releases/author-mod-1.0.0.tar.gz"},
-		{"goproxy", "goproxy/example.com/mod/@v/v1.0.0.zip", "goproxy/example.com/mod/@v/v1.0.0.zip"},
-		{"cargo", "crates/mycrate/mycrate-1.0.0.crate", "crates/mycrate/mycrate-1.0.0.crate"},
-		{"nuget", "nuget/e2epkg/1.0.0/e2epkg.1.0.0.nupkg", "nuget/e2epkg/1.0.0/e2epkg.1.0.0.nupkg"},
-		{"terraform", "hashicorp/aws/download/pkg.zip", "v1/providers/hashicorp/aws/download/pkg.zip"},
-		{"docker", "library/testimg/blobs/blobdata", "v2/library/testimg/blobs/blobdata"},
-	}
-
-	for _, tc := range cases {
+	for _, tc := range cachingCases {
 		t.Run(tc.pkgType, func(t *testing.T) {
 			name := "cache-" + tc.pkgType
 			createRepo(t, fmt.Sprintf(`{
@@ -0,0 +1 @@
+mygem-1.0.0 e2e fixture
@@ -0,0 +1,62 @@
+//go:build dockere2e
+
+package e2edocker
+
+import (
+	"fmt"
+	"os"
+	"os/exec"
+	"strings"
+	"testing"
+	"time"
+)
+
+// TestLocalNPMPublishInstall drives a stock npm client against a local npm
+// repo: publish unscoped and scoped packages, install and require() them in a
+// fresh project, publish a second version, and reject a republish.
+func TestLocalNPMPublishInstall(t *testing.T) {
+	network := os.Getenv("COMPOSE_NETWORK")
+	internal := os.Getenv("ARTIFACTAPI_INTERNAL")
+	if network == "" || internal == "" {
+		t.Skip("COMPOSE_NETWORK/ARTIFACTAPI_INTERNAL not set; run via scripts/docker-e2e.sh")
+	}
+	if _, err := exec.LookPath("docker"); err != nil {
+		t.Skip("docker not available on the test host")
+	}
+
+	name := fmt.Sprintf("e2e-npm-local-%d", time.Now().UnixNano())
+	createRepo(t, fmt.Sprintf(`{"name":%q,"package_type":"npm","repo_type":"local"}`, name))
+	defer deleteRepo(t, name)
+
+	reg := strings.TrimRight(internal, "/") + "/api/v1/local/" + name + "/"
+	script := `set -euo pipefail
+printf 'registry=%s\n%s:_authToken=unused\n' "$REG" "${REG#http:}" > ~/.npmrc
+mk() { d="/w/src/$1-$2"; mkdir -p "$d"; cd "$d"
+  printf '{"name":"%s","version":"%s","main":"index.js"}' "$1" "$2" > package.json
+  echo "module.exports = '$1@$2';" > index.js; }
+mk e2e-npm-plain 1.0.0 && npm publish
+mk @e2e/npm-scoped 1.0.0 && npm publish
+mk e2e-npm-plain 1.0.0
+if npm publish 2>/dev/null; then echo "republish accepted"; exit 1; fi
+mkdir -p /w/app && cd /w/app && npm init -y >/dev/null
+npm install --no-audit --cache /tmp/fresh-cache e2e-npm-plain @e2e/npm-scoped
+node -e "console.log('REQ', require('e2e-npm-plain'), require('@e2e/npm-scoped'))"
+mk e2e-npm-plain 1.1.0 && npm publish
+echo "VERSIONS $(npm view e2e-npm-plain versions --json | tr -d ' \n')"
+echo "LATEST $(npm view e2e-npm-plain dist-tags.latest)"`
+
+	out, err := exec.Command("docker", "run", "--rm", "--network", network,
+		"-e", "REG="+reg, "node:22", "bash", "-c", script).CombinedOutput()
+	if err != nil {
+		t.Fatalf("npm client run failed: %v\n%s", err, out)
+	}
+	for _, want := range []string{
+		"REQ e2e-npm-plain@1.0.0 @e2e/npm-scoped@1.0.0",
+		`VERSIONS ["1.0.0","1.1.0"]`,
+		"LATEST 1.1.0",
+	} {
+		if !strings.Contains(string(out), want) {
+			t.Fatalf("missing %q in npm output:\n%s", want, out)
+		}
+	}
+}
@@ -0,0 +1,68 @@
+//go:build dockere2e
+
+package e2edocker
+
+import (
+	"fmt"
+	"os"
+	"os/exec"
+	"strings"
+	"testing"
+	"time"
+)
+
+// TestVirtualNPMInstall publishes to two local npm repos and drives a stock
+// npm client through a virtual over both: merged versions and dist-tags, and
+// an install whose tarballs resolve via the virtual's member redirect.
+func TestVirtualNPMInstall(t *testing.T) {
+	network := os.Getenv("COMPOSE_NETWORK")
+	internal := os.Getenv("ARTIFACTAPI_INTERNAL")
+	if network == "" || internal == "" {
+		t.Skip("COMPOSE_NETWORK/ARTIFACTAPI_INTERNAL not set; run via scripts/docker-e2e.sh")
+	}
+	if _, err := exec.LookPath("docker"); err != nil {
+		t.Skip("docker not available on the test host")
+	}
+
+	id := time.Now().UnixNano()
+	a, b, virt := fmt.Sprintf("e2e-npm-a-%d", id), fmt.Sprintf("e2e-npm-b-%d", id), fmt.Sprintf("e2e-npm-v-%d", id)
+	for _, name := range []string{a, b} {
+		createRepo(t, fmt.Sprintf(`{"name":%q,"package_type":"npm","repo_type":"local"}`, name))
+		defer deleteRepo(t, name)
+	}
+	createVirtual(t, fmt.Sprintf(`{"name":%q,"package_type":"npm","members":[%q,%q]}`, virt, a, b))
+	defer deleteVirtual(t, virt)
+
+	base := strings.TrimRight(internal, "/")
+	script := `set -euo pipefail
+pub() { d="/w/src/$2-$3-$1"; mkdir -p "$d"; cd "$d"
+  printf '{"name":"%s","version":"%s","main":"index.js"}' "$2" "$3" > package.json
+  echo "module.exports = '$1:$2@$3';" > index.js
+  printf 'registry=%s\n%s:_authToken=unused\n' "$BASE/api/v1/local/$1/" "${BASE#http:}/api/v1/local/$1/" > .npmrc
+  npm publish; }
+pub "$A" @e2e/virt 1.0.0
+pub "$B" @e2e/virt 1.0.0
+pub "$B" @e2e/virt 1.1.0
+mkdir -p /w/app && cd /w/app && npm init -y >/dev/null
+echo "registry=$BASE/api/v1/virtual/$V/" > .npmrc
+echo "VERSIONS $(npm view @e2e/virt versions --json | tr -d ' \n')"
+npm install --no-audit --cache /tmp/fresh-cache @e2e/virt@1.0.0
+node -e "console.log('REQ', require('@e2e/virt'))"
+echo "RESOLVED $(node -p "require('./package-lock.json').packages['node_modules/@e2e/virt'].resolved")"`
+
+	out, err := exec.Command("docker", "run", "--rm", "--network", network,
+		"-e", "BASE="+base, "-e", "A="+a, "-e", "B="+b, "-e", "V="+virt,
+		"node:22", "bash", "-c", script).CombinedOutput()
+	if err != nil {
+		t.Fatalf("npm client run failed: %v\n%s", err, out)
+	}
+	for _, want := range []string{
+		`VERSIONS ["1.0.0","1.1.0"]`,
+		"REQ " + a + ":@e2e/virt@1.0.0",
+		"RESOLVED " + base + "/api/v1/virtual/" + virt + "/" + a + "/@e2e/virt/-/virt-1.0.0.tgz",
+	} {
+		if !strings.Contains(string(out), want) {
+			t.Fatalf("missing %q in npm output:\n%s", want, out)
+		}
+	}
+}
@@ -0,0 +1,182 @@
+//go:build e2e
+
+package e2e
+
+import (
+	"encoding/json"
+	"io"
+	"net/http"
+	"net/http/httptest"
+	"strconv"
+	"strings"
+	"testing"
+	"time"
+
+	"git.unkin.net/unkin/artifactapi/internal/testsupport"
+)
+
+// assertHeadMatchesGet checks HEAD returns GET's status, type and length with no body.
+func assertHeadMatchesGet(t *testing.T, url string) {
+	t.Helper()
+	get, err := http.Get(url)
+	if err != nil {
+		t.Fatalf("GET: %v", err)
+	}
+	body, _ := io.ReadAll(get.Body)
+	get.Body.Close()
+	if get.StatusCode != http.StatusOK {
+		t.Fatalf("GET %s: %d %s", url, get.StatusCode, body)
+	}
+
+	head, err := http.Head(url)
+	if err != nil {
+		t.Fatalf("HEAD: %v", err)
+	}
+	hb, _ := io.ReadAll(head.Body)
+	head.Body.Close()
+	if head.StatusCode != http.StatusOK {
+		t.Fatalf("HEAD %s: got %d, want 200", url, head.StatusCode)
+	}
+	if len(hb) != 0 {
+		t.Errorf("HEAD returned a %d-byte body", len(hb))
+	}
+	for _, h := range []string{"Content-Type", "Cache-Control", "Vary"} {
+		if head.Header.Get(h) != get.Header.Get(h) {
+			t.Errorf("HEAD %s %q, GET %q", h, head.Header.Get(h), get.Header.Get(h))
+		}
+	}
+	if head.ContentLength != int64(len(body)) {
+		t.Errorf("HEAD Content-Length %d, GET body %d bytes", head.ContentLength, len(body))
+	}
+}
+
+func TestProxyHeadNuGetRewritten(t *testing.T) {
+	var upstream *httptest.Server
+	upstream = httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		w.Header().Set("Content-Type", "application/json")
+		json.NewEncoder(w).Encode(map[string]any{"version": "3.0.0", "resources": []map[string]string{
+			{"@id": upstream.URL + "/v3-flatcontainer/", "@type": "PackageBaseAddress/3.0.0"},
+		}})
+	}))
+	defer upstream.Close()
+
+	createRemote(t, mustFormat(`{"name": "head-nuget", "package_type": "nuget", "base_url": %q}`, upstream.URL+"/v3/index.json"))
+	defer deleteRemote(t, "head-nuget")
+
+	url := apiURL("/api/v1/remote/head-nuget/index.json")
+	assertHeadMatchesGet(t, url)
+
+	req, _ := http.NewRequest(http.MethodHead, url, nil)
+	req.Host = "evil.example:"
+	resp, err := http.DefaultClient.Do(req)
+	if err != nil {
+		t.Fatal(err)
+	}
+	resp.Body.Close()
+	if resp.StatusCode != http.StatusBadRequest {
+		t.Errorf("HEAD with invalid Host: got %d, want 400", resp.StatusCode)
+	}
+}
+
+func TestProxyHeadGitHubRPMRepodata(t *testing.T) {
+	rpm := testsupport.MinimalRPM("demo", "1.2", "3", "x86_64")
+	var gh *httptest.Server
+	mux := http.NewServeMux()
+	mux.HandleFunc("/repos/acme/tools/releases", func(w http.ResponseWriter, r *http.Request) {
+		if p := r.URL.Query().Get("page"); p != "" && p != "1" {
+			w.Write([]byte("[]"))
+			return
+		}
+		json.NewEncoder(w).Encode([]map[string]any{{"tag_name": "v1.2-3", "assets": []map[string]any{{
+			"name":                 "demo-1.2-3.x86_64.rpm",
+			"size":                 len(rpm),
+			"browser_download_url": gh.URL + "/acme/tools/releases/download/v1.2-3/demo-1.2-3.x86_64.rpm",
+		}}}})
+	})
+	mux.HandleFunc("/acme/tools/releases/download/", func(w http.ResponseWriter, r *http.Request) {
+		w.Header().Set("Content-Length", strconv.Itoa(len(rpm)))
+		w.Write(rpm)
+	})
+	gh = httptest.NewServer(mux)
+	defer gh.Close()
+
+	createRemote(t, mustFormat(`{"name": "head-ghrpm", "package_type": "github_rpm", "base_url": %q, "releases_remote": "github"}`, gh.URL+"/repos/acme/tools"))
+	defer deleteRemote(t, "head-ghrpm")
+
+	url := apiURL("/api/v1/remote/head-ghrpm/repodata/repomd.xml")
+	deadline := time.Now().Add(30 * time.Second)
+	for {
+		if _, status := getBody(t, url); status == http.StatusOK || time.Now().After(deadline) {
+			break
+		}
+		time.Sleep(500 * time.Millisecond)
+	}
+	assertHeadMatchesGet(t, url)
+}
+
+func TestProxyHeadNpmUncachedPackument(t *testing.T) {
+	const abbreviated = "application/vnd.npm.install-v1+json"
+	upstream := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		w.Header().Set("Content-Type", "application/json")
+		doc := map[string]any{"name": "left-pad", "versions": map[string]any{"1.3.0": map[string]any{}}}
+		if r.Header.Get("Accept") != abbreviated {
+			doc["readme"] = strings.Repeat("full packument ", 200)
+		}
+		json.NewEncoder(w).Encode(doc)
+	}))
+	defer upstream.Close()
+
+	createRemote(t, mustFormat(`{"name": "head-npm", "package_type": "npm", "base_url": %q}`, upstream.URL))
+	defer deleteRemote(t, "head-npm")
+
+	url := apiURL("/api/v1/remote/head-npm/left-pad")
+	do := func(method string) (*http.Response, []byte) {
+		req, _ := http.NewRequest(method, url, nil)
+		req.Header.Set("Accept", abbreviated)
+		resp, err := http.DefaultClient.Do(req)
+		if err != nil {
+			t.Fatalf("%s: %v", method, err)
+		}
+		defer resp.Body.Close()
+		b, _ := io.ReadAll(resp.Body)
+		return resp, b
+	}
+	head, _ := do(http.MethodHead)
+	get, body := do(http.MethodGet)
+	if head.StatusCode != http.StatusOK || get.StatusCode != http.StatusOK {
+		t.Fatalf("HEAD %d, GET %d", head.StatusCode, get.StatusCode)
+	}
+	if head.ContentLength != int64(len(body)) {
+		t.Errorf("uncached HEAD Content-Length %d, GET body %d bytes", head.ContentLength, len(body))
+	}
+}
+
+func TestProxyHeadNonRewritingUsesUpstreamHead(t *testing.T) {
+	methods := make(chan string, 4)
+	upstream := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		methods <- r.Method
+		w.Header().Set("Content-Type", "text/xml")
+		w.Write([]byte("<repomd/>"))
+	}))
+	defer upstream.Close()
+
+	createRemote(t, mustFormat(`{"name": "head-rpm", "package_type": "rpm", "base_url": %q}`, upstream.URL))
+	defer deleteRemote(t, "head-rpm")
+
+	resp, err := http.Head(apiURL("/api/v1/remote/head-rpm/repodata/repomd.xml"))
+	if err != nil {
+		t.Fatalf("HEAD: %v", err)
+	}
+	resp.Body.Close()
+	if resp.StatusCode != http.StatusOK || resp.Header.Get("X-Artifact-Source") != "remote" {
+		t.Fatalf("HEAD %d source %q", resp.StatusCode, resp.Header.Get("X-Artifact-Source"))
+	}
+	close(methods)
+	var seen []string
+	for m := range methods {
+		seen = append(seen, m)
+	}
+	if len(seen) != 1 || seen[0] != http.MethodHead {
+		t.Errorf("upstream saw %v, want a single HEAD", seen)
+	}
+}
@@ -0,0 +1,59 @@
+//go:build e2e
+
+package e2e
+
+import (
+	"encoding/json"
+	"net/http"
+	"strings"
+	"testing"
+)
+
+func getForwardedHTTPS(t *testing.T, url string) *http.Response {
+	t.Helper()
+	req, _ := http.NewRequest(http.MethodGet, url, nil)
+	req.Header.Set("X-Forwarded-Proto", "https")
+	client := &http.Client{CheckRedirect: func(*http.Request, []*http.Request) error { return http.ErrUseLastResponse }}
+	resp, err := client.Do(req)
+	if err != nil {
+		t.Fatalf("GET %s: %v", url, err)
+	}
+	return resp
+}
+
+func TestForwardedProtoCargoConfig(t *testing.T) {
+	createRemote(t, `{"name": "scheme-cargo", "package_type": "cargo", "base_url": "https://index.crates.io"}`)
+	defer deleteRemote(t, "scheme-cargo")
+
+	resp := getForwardedHTTPS(t, apiURL("/api/v1/remote/scheme-cargo/config.json"))
+	defer resp.Body.Close()
+	var cfg map[string]string
+	if err := json.NewDecoder(resp.Body).Decode(&cfg); err != nil {
+		t.Fatal(err)
+	}
+	if want := "https://" + strings.TrimPrefix(baseURL, "http://") + "/api/v1/remote/scheme-cargo/"; !strings.HasPrefix(cfg["dl"], want) {
+		t.Errorf("dl = %q, want prefix %q", cfg["dl"], want)
+	}
+}
+
+func TestForwardedProtoVirtualRedirect(t *testing.T) {
+	createRemote(t, `{"name": "scheme-rpm", "package_type": "rpm", "base_url": "https://rpm.example.com"}`)
+	defer deleteRemote(t, "scheme-rpm")
+	resp, err := http.Post(apiURL("/api/v2/virtuals"), "application/json",
+		strings.NewReader(`{"name": "scheme-virt", "package_type": "rpm", "members": ["scheme-rpm"]}`))
+	if err != nil {
+		t.Fatal(err)
+	}
+	resp.Body.Close()
+	if resp.StatusCode != http.StatusCreated {
+		t.Fatalf("create virtual: status %d", resp.StatusCode)
+	}
+	defer deleteRequest(t, apiURL("/api/v2/virtuals/scheme-virt"))
+
+	resp = getForwardedHTTPS(t, apiURL("/api/v1/virtual/scheme-virt/scheme-rpm/Packages/a.rpm"))
+	resp.Body.Close()
+	loc := resp.Header.Get("Location")
+	if want := "https://" + strings.TrimPrefix(baseURL, "http://") + "/api/v1/remote/scheme-rpm/Packages/a.rpm"; resp.StatusCode != http.StatusFound || loc != want {
+		t.Errorf("redirect = %d %q, want 302 %q", resp.StatusCode, loc, want)
+	}
+}
@@ -13,6 +13,7 @@ import (

 	"github.com/go-chi/chi/v5"

+	v1 "git.unkin.net/unkin/artifactapi/internal/api/v1"
 	"git.unkin.net/unkin/artifactapi/internal/database"
 	tfprov "git.unkin.net/unkin/artifactapi/internal/provider/terraform"
 	"git.unkin.net/unkin/artifactapi/internal/tfsign"
@@ -290,12 +291,5 @@ func writeJSON(w http.ResponseWriter, v any) {
 }

 func baseURL(r *http.Request) string {
-	scheme := "http"
-	if r.TLS != nil {
-		scheme = "https"
-	}
-	if fwd := r.Header.Get("X-Forwarded-Proto"); fwd != "" {
-		scheme = fwd
-	}
-	return scheme + "://" + r.Host
+	return v1.Scheme(r) + "://" + r.Host
 }
@@ -0,0 +1,334 @@
+package v1
+
+import (
+	"bytes"
+	"context"
+	"crypto/sha1"
+	"crypto/sha512"
+	"encoding/base64"
+	"encoding/hex"
+	"encoding/json"
+	"errors"
+	"fmt"
+	"io"
+	"log/slog"
+	"net/http"
+	"net/url"
+	"regexp"
+	"strings"
+
+	"github.com/go-chi/chi/v5"
+
+	"git.unkin.net/unkin/artifactapi/internal/database"
+	"git.unkin.net/unkin/artifactapi/internal/storage"
+	"git.unkin.net/unkin/artifactapi/pkg/models"
+)
+
+// Local npm repos speak the npm registry publish/read protocol. Per package:
+//   <pkg>/-/<base>-<version>.tgz   tarball (immutable)
+//   <pkg>/-/<base>-<version>.json  version manifest (immutable)
+//   <pkg>/-/dist-tags/<tag>        version string (mutable)
+
+var (
+	npmNameRe   = regexp.MustCompile(`^(@[a-z0-9][a-z0-9._~-]*/)?[a-z0-9][a-z0-9._~-]*$`)
+	npmRefRe    = regexp.MustCompile(`^[0-9A-Za-z][0-9A-Za-z.+_-]*$`)
+	npmSemverRe = regexp.MustCompile(`^(0|[1-9]\d*)\.(0|[1-9]\d*)\.(0|[1-9]\d*)(?:-((?:0|[1-9]\d*|\d*[a-zA-Z-][0-9a-zA-Z-]*)(?:\.(?:0|[1-9]\d*|\d*[a-zA-Z-][0-9a-zA-Z-]*))*))?(?:\+([0-9a-zA-Z-]+(?:\.[0-9a-zA-Z-]+)*))?$`)
+
+	npmMaxPublishBytes int64 = 256 << 20
+)
+
+func (h *ProxyHandler) localNPMRemote(r *http.Request, name string) (*models.Remote, bool) {
+	remote, err := h.db.GetRemote(r.Context(), name)
+	if err != nil {
+		return nil, false
+	}
+	return remote, remote.RepoType == models.RepoTypeLocal && remote.PackageType == models.PackageNPM
+}
+
+// npmPackageName returns the package a packument/publish path addresses, or ""
+// when the path is something else (e.g. a tarball under <pkg>/-/).
+func npmPackageName(raw string) string {
+	p, err := url.PathUnescape(raw)
+	if err != nil {
+		return ""
+	}
+	p = strings.TrimSuffix(p, "/")
+	if !npmNameRe.MatchString(p) {
+		return ""
+	}
+	return p
+}
+
+type npmRelease struct {
+	version  string
+	manifest []byte
+	tarball  []byte
+	tags     []string
+}
+
+func parseNPMPublish(body io.Reader, pkg string) (*npmRelease, error) {
+	var doc struct {
+		Name        string                     `json:"name"`
+		DistTags    map[string]string          `json:"dist-tags"`
+		Versions    map[string]json.RawMessage `json:"versions"`
+		Attachments map[string]struct {
+			Data string `json:"data"`
+		} `json:"_attachments"`
+	}
+	if err := json.NewDecoder(body).Decode(&doc); err != nil {
+		return nil, fmt.Errorf("invalid publish body: %w", err)
+	}
+	if doc.Name != pkg {
+		return nil, fmt.Errorf("body name %q does not match %q", doc.Name, pkg)
+	}
+	if len(doc.Versions) != 1 || len(doc.Attachments) != 1 {
+		return nil, errors.New("publish must carry exactly one version and one tarball")
+	}
+
+	rel := &npmRelease{}
+	for v, m := range doc.Versions {
+		rel.version, rel.manifest = v, m
+	}
+	if !npmSemverRe.MatchString(rel.version) {
+		return nil, fmt.Errorf("invalid version %q", rel.version)
+	}
+	var meta struct {
+		Name    string `json:"name"`
+		Version string `json:"version"`
+		Dist    struct {
+			Integrity string `json:"integrity"`
+			Shasum    string `json:"shasum"`
+		} `json:"dist"`
+	}
+	if err := json.Unmarshal(rel.manifest, &meta); err != nil {
+		return nil, fmt.Errorf("invalid version manifest: %w", err)
+	}
+	if meta.Name != pkg || meta.Version != rel.version {
+		return nil, errors.New("version manifest name/version mismatch")
+	}
+
+	for _, a := range doc.Attachments {
+		tgz, err := base64.StdEncoding.DecodeString(a.Data)
+		if err != nil {
+			return nil, fmt.Errorf("invalid tarball encoding: %w", err)
+		}
+		rel.tarball = tgz
+	}
+	if err := verifyNPMDist(rel.tarball, meta.Dist.Integrity, meta.Dist.Shasum); err != nil {
+		return nil, err
+	}
+
+	for tag, v := range doc.DistTags {
+		if !npmRefRe.MatchString(tag) || v != rel.version {
+			return nil, fmt.Errorf("dist-tag %q must point at the published version", tag)
+		}
+		rel.tags = append(rel.tags, tag)
+	}
+	return rel, nil
+}
+
+func verifyNPMDist(tgz []byte, integrity, shasum string) error {
+	if b64, ok := strings.CutPrefix(integrity, "sha512-"); ok {
+		sum := sha512.Sum512(tgz)
+		if base64.StdEncoding.EncodeToString(sum[:]) != b64 {
+			return errors.New("tarball does not match dist.integrity")
+		}
+		return nil
+	}
+	if shasum != "" {
+		sum := sha1.Sum(tgz)
+		if hex.EncodeToString(sum[:]) != shasum {
+			return errors.New("tarball does not match dist.shasum")
+		}
+		return nil
+	}
+	return errors.New("version manifest carries no sha512 integrity or shasum")
+}
+
+func npmError(w http.ResponseWriter, status int, msg string) {
+	w.Header().Set("Content-Type", "application/json")
+	w.WriteHeader(status)
+	_ = json.NewEncoder(w).Encode(map[string]string{"error": msg})
+}
+
+func (h *ProxyHandler) handleLocalPut(w http.ResponseWriter, r *http.Request) {
+	remote, ok := h.localNPMRemote(r, chi.URLParam(r, "localName"))
+	if !ok {
+		npmError(w, http.StatusMethodNotAllowed, "publish is only supported on local npm repositories")
+		return
+	}
+	pkg := npmPackageName(chi.URLParam(r, "*"))
+	if pkg == "" {
+		npmError(w, http.StatusBadRequest, "invalid package name")
+		return
+	}
+	rel, err := parseNPMPublish(http.MaxBytesReader(w, r.Body, npmMaxPublishBytes), pkg)
+	if err != nil {
+		var tooLarge *http.MaxBytesError
+		if errors.As(err, &tooLarge) {
+			npmError(w, http.StatusRequestEntityTooLarge, fmt.Sprintf("publish body exceeds %d bytes", npmMaxPublishBytes))
+			return
+		}
+		npmError(w, http.StatusBadRequest, err.Error())
+		return
+	}
+
+	base := pkg[strings.LastIndex(pkg, "/")+1:]
+	stem := pkg + "/-/" + base + "-" + rel.version
+	// The manifest gates visibility and tags may only name visible versions, so
+	// the order is tarball, manifest, tags. A tarball left by a failed publish is
+	// accepted again only with identical bytes; an existing manifest always conflicts.
+	conflict := func() {
+		npmError(w, http.StatusConflict, fmt.Sprintf("%s@%s already exists; overwrites are not allowed", pkg, rel.version))
+	}
+	ok, err = h.npmStoreImmutable(r.Context(), remote.Name, stem+".tgz", rel.tarball, "application/octet-stream")
+	if err != nil {
+		npmError(w, http.StatusInternalServerError, err.Error())
+		return
+	}
+	if !ok {
+		conflict()
+		return
+	}
+	if _, err := h.npmStore(r.Context(), remote.Name, stem+".json", rel.manifest, "application/json", false); err != nil {
+		if errors.Is(err, database.ErrAlreadyExists) {
+			conflict()
+			return
+		}
+		npmError(w, http.StatusInternalServerError, err.Error())
+		return
+	}
+	for _, tag := range rel.tags {
+		if _, err := h.npmStore(r.Context(), remote.Name, pkg+"/-/dist-tags/"+tag, []byte(rel.version), "text/plain", true); err != nil {
+			npmError(w, http.StatusInternalServerError, err.Error())
+			return
+		}
+	}
+
+	slog.Info("local npm package published", "repo", remote.Name, "package", pkg, "version", rel.version)
+	w.Header().Set("Content-Type", "application/json")
+	w.WriteHeader(http.StatusCreated)
+	_ = json.NewEncoder(w).Encode(map[string]any{"ok": true, "id": pkg})
+}
+
+// npmStoreImmutable reports false when path already holds different bytes.
+func (h *ProxyHandler) npmStoreImmutable(ctx context.Context, repo, path string, data []byte, contentType string) (bool, error) {
+	hash, err := h.npmStore(ctx, repo, path, data, contentType, false)
+	if !errors.Is(err, database.ErrAlreadyExists) {
+		return err == nil, err
+	}
+	existing, err := h.db.GetLocalFile(ctx, repo, path)
+	if err != nil || existing == nil {
+		return false, fmt.Errorf("read existing %s: %v", path, err)
+	}
+	return existing.ContentHash == hash, nil
+}
+
+func (h *ProxyHandler) npmStore(ctx context.Context, repo, path string, data []byte, contentType string, mutable bool) (string, error) {
+	result, err := h.cas.Store(ctx, bytes.NewReader(data), contentType)
+	if err != nil {
+		return "", err
+	}
+	if err := h.db.UpsertBlob(ctx, result.ContentHash, result.S3Key, result.SizeBytes, contentType); err != nil {
+		return "", err
+	}
+	if mutable {
+		return result.ContentHash, h.db.UpsertLocalFile(ctx, repo, path, result.ContentHash)
+	}
+	return result.ContentHash, h.db.CreateLocalFile(ctx, repo, path, result.ContentHash)
+}
+
+func (h *ProxyHandler) readLocalBlob(ctx context.Context, contentHash string) ([]byte, error) {
+	reader, _, err := h.store.Download(ctx, storage.BlobKey(strings.TrimPrefix(contentHash, "sha256:")))
+	if err != nil {
+		return nil, err
+	}
+	defer reader.Close()
+	return io.ReadAll(reader)
+}
+
+func (h *ProxyHandler) serveNPMPackument(w http.ResponseWriter, r *http.Request, remote *models.Remote, pkg string) {
+	files, err := h.db.ListLocalFilesByPrefix(r.Context(), remote.Name, pkg+"/-/")
+	if err != nil {
+		npmError(w, http.StatusInternalServerError, err.Error())
+		return
+	}
+	base, ok := requestBaseURL(w, r)
+	if !ok {
+		return
+	}
+	read := func(hash string) ([]byte, error) { return h.readLocalBlob(r.Context(), hash) }
+	tarballBase := base + "/api/v1/local/" + remote.Name + "/"
+	doc, err := buildNPMPackument(pkg, files, read, tarballBase)
+	if err != nil {
+		npmError(w, http.StatusInternalServerError, err.Error())
+		return
+	}
+	if doc == nil {
+		npmError(w, http.StatusNotFound, "not found")
+		return
+	}
+	setNPMPerHostHeaders(w)
+	w.Header().Set("Content-Type", "application/json")
+	w.Header().Set("X-Artifact-Source", "local")
+	w.WriteHeader(http.StatusOK)
+	_ = json.NewEncoder(w).Encode(doc)
+}
+
+// setNPMPerHostHeaders keeps shared caches off packuments whose tarball URLs
+// are built from the request host and Accept (abbreviated vs full).
+func setNPMPerHostHeaders(w http.ResponseWriter) {
+	w.Header().Set("Cache-Control", "no-store")
+	w.Header().Set("Vary", "Host, X-Forwarded-Proto, Accept")
+}
+
+// buildNPMPackument returns nil when pkg has no published versions.
+func buildNPMPackument(pkg string, files []database.LocalFile, read func(hash string) ([]byte, error), tarballBase string) (map[string]any, error) {
+	prefix := pkg + "/-/"
+	versions := map[string]any{}
+	tags := map[string]string{}
+	for _, f := range files {
+		// LIKE treats '_' as a wildcard, so the prefix match can over-select.
+		rest, ok := strings.CutPrefix(f.FilePath, prefix)
+		if !ok {
+			continue
+		}
+		if tag, isTag := strings.CutPrefix(rest, "dist-tags/"); isTag {
+			v, err := read(f.ContentHash)
+			if err != nil {
+				return nil, err
+			}
+			tags[tag] = string(v)
+			continue
+		}
+		stem, isManifest := strings.CutSuffix(rest, ".json")
+		if !isManifest {
+			continue
+		}
+		raw, err := read(f.ContentHash)
+		if err != nil {
+			return nil, err
+		}
+		var m map[string]any
+		if err := json.Unmarshal(raw, &m); err != nil {
+			return nil, fmt.Errorf("manifest %s: %w", f.FilePath, err)
+		}
+		dist, _ := m["dist"].(map[string]any)
+		if dist == nil {
+			dist = map[string]any{}
+			m["dist"] = dist
+		}
+		dist["tarball"] = tarballBase + prefix + stem + ".tgz"
+		v, _ := m["version"].(string)
+		versions[v] = m
+	}
+	if len(versions) == 0 {
+		return nil, nil
+	}
+	for tag, v := range tags {
+		if _, ok := versions[v]; !ok {
+			delete(tags, tag)
+		}
+	}
+	return map[string]any{"_id": pkg, "name": pkg, "dist-tags": tags, "versions": versions}, nil
+}
@@ -0,0 +1,333 @@
+package v1
+
+import (
+	"context"
+	"encoding/json"
+	"fmt"
+	"io"
+	"net/http"
+	"net/http/httptest"
+	"os"
+	"strings"
+	"sync"
+	"testing"
+	"time"
+
+	"github.com/go-chi/chi/v5"
+
+	"git.unkin.net/unkin/artifactapi/internal/database"
+	"git.unkin.net/unkin/artifactapi/internal/storage"
+	"git.unkin.net/unkin/artifactapi/internal/testsupport"
+	"git.unkin.net/unkin/artifactapi/internal/virtual"
+	"git.unkin.net/unkin/artifactapi/pkg/models"
+)
+
+var (
+	testDSN   string
+	testMinio testsupport.MinioConn
+)
+
+func TestMain(m *testing.M) {
+	ctx := context.Background()
+	dsn, termPG, err := testsupport.StartPostgres(ctx)
+	if err != nil {
+		os.Exit(m.Run())
+	}
+	conn, termMinio, err := testsupport.StartMinio(ctx)
+	if err != nil {
+		termPG()
+		os.Exit(m.Run())
+	}
+	testDSN, testMinio = dsn, conn
+	code := m.Run()
+	termMinio()
+	termPG()
+	os.Exit(code)
+}
+
+func npmTestServer(t *testing.T) (*ProxyHandler, *httptest.Server) {
+	t.Helper()
+	if testDSN == "" {
+		t.Skip("Docker unavailable")
+	}
+	db, err := database.New(testDSN)
+	if err != nil {
+		t.Fatal(err)
+	}
+	t.Cleanup(db.Close)
+	var store *storage.S3
+	for i := 0; i < 20; i++ {
+		if store, err = storage.NewS3(testMinio.Endpoint, testMinio.AccessKey, testMinio.SecretKey, "npm", false, ""); err == nil {
+			break
+		}
+		time.Sleep(500 * time.Millisecond)
+	}
+	if err != nil {
+		t.Fatal(err)
+	}
+	h := NewProxyHandler(nil, nil, db, store, nil)
+	r := chi.NewRouter()
+	r.Mount("/api/v1", h.Routes())
+	srv := httptest.NewServer(r)
+	t.Cleanup(srv.Close)
+	return h, srv
+}
+
+func npmRepo(t *testing.T, h *ProxyHandler, pt models.PackageType) string {
+	t.Helper()
+	name := fmt.Sprintf("npm-%s-%d", pt, time.Now().UnixNano())
+	if err := h.db.CreateRemote(context.Background(), &models.Remote{Name: name, PackageType: pt, RepoType: models.RepoTypeLocal}); err != nil {
+		t.Fatal(err)
+	}
+	return name
+}
+
+func npmDo(t *testing.T, method, url, body string) (int, string) {
+	t.Helper()
+	req, _ := http.NewRequest(method, url, strings.NewReader(body))
+	resp, err := http.DefaultClient.Do(req)
+	if err != nil {
+		t.Fatal(err)
+	}
+	defer func() { _ = resp.Body.Close() }()
+	b, _ := io.ReadAll(resp.Body)
+	return resp.StatusCode, string(b)
+}
+
+func packumentVersions(t *testing.T, url string) map[string]any {
+	t.Helper()
+	code, body := npmDo(t, http.MethodGet, url, "")
+	if code != http.StatusOK {
+		t.Fatalf("GET %s = %d: %s", url, code, body)
+	}
+	var doc struct {
+		Versions map[string]any `json:"versions"`
+	}
+	if err := json.Unmarshal([]byte(body), &doc); err != nil {
+		t.Fatal(err)
+	}
+	return doc.Versions
+}
+
+func packumentTags(t *testing.T, url string) map[string]string {
+	t.Helper()
+	_, body := npmDo(t, http.MethodGet, url, "")
+	var doc struct {
+		Tags map[string]string `json:"dist-tags"`
+	}
+	if err := json.Unmarshal([]byte(body), &doc); err != nil {
+		t.Fatal(err)
+	}
+	return doc.Tags
+}
+
+func concurrentPuts(t *testing.T, url string, bodies ...string) []int {
+	t.Helper()
+	var wg sync.WaitGroup
+	codes := make([]int, len(bodies))
+	for i, b := range bodies {
+		wg.Add(1)
+		go func() {
+			defer wg.Done()
+			codes[i], _ = npmDo(t, http.MethodPut, url, b)
+		}()
+	}
+	wg.Wait()
+	return codes
+}
+
+func TestNPMPublishHandler(t *testing.T) {
+	h, srv := npmTestServer(t)
+	repo := npmRepo(t, h, models.PackageNPM)
+	base := srv.URL + "/api/v1/local/" + repo + "/"
+
+	t.Run("scoped name via %2f", func(t *testing.T) {
+		body := publishBody(t, "@acme/widget", []byte("scoped tgz"), publishOpts{})
+		if code, b := npmDo(t, http.MethodPut, base+"@acme%2fwidget", body); code != http.StatusCreated {
+			t.Fatalf("publish = %d: %s", code, b)
+		}
+		v := packumentVersions(t, base+"@acme%2fwidget")
+		dist := v["1.0.0"].(map[string]any)["dist"].(map[string]any)
+		tarball := dist["tarball"].(string)
+		if !strings.HasPrefix(tarball, base+"@acme/widget/-/widget-1.0.0.tgz") {
+			t.Fatalf("tarball = %s", tarball)
+		}
+		if code, b := npmDo(t, http.MethodGet, tarball, ""); code != http.StatusOK || b != "scoped tgz" {
+			t.Fatalf("tarball fetch = %d %q", code, b)
+		}
+	})
+
+	t.Run("republish", func(t *testing.T) {
+		if code, b := npmDo(t, http.MethodPut, base+"repub", publishBody(t, "repub", []byte("v1"), publishOpts{})); code != http.StatusCreated {
+			t.Fatalf("first publish = %d: %s", code, b)
+		}
+		if code, _ := npmDo(t, http.MethodPut, base+"repub", publishBody(t, "repub", []byte("different"), publishOpts{})); code != http.StatusConflict {
+			t.Errorf("different bytes = %d, want 409", code)
+		}
+		if code, b := npmDo(t, http.MethodPut, base+"repub", publishBody(t, "repub", []byte("v2"), publishOpts{version: "1.1.0"})); code != http.StatusCreated {
+			t.Fatalf("publish 1.1.0 = %d: %s", code, b)
+		}
+		if code, _ := npmDo(t, http.MethodPut, base+"repub", publishBody(t, "repub", []byte("v1"), publishOpts{})); code != http.StatusConflict {
+			t.Errorf("identical republish = %d, want 409", code)
+		}
+		if tags := packumentTags(t, base+"repub"); tags["latest"] != "1.1.0" {
+			t.Errorf("latest = %q, want 1.1.0", tags["latest"])
+		}
+	})
+
+	t.Run("rejected republish leaves tags", func(t *testing.T) {
+		for _, v := range []string{"1.0.0", "1.1.0"} {
+			if code, b := npmDo(t, http.MethodPut, base+"rej", publishBody(t, "rej", []byte("tgz "+v), publishOpts{version: v})); code != http.StatusCreated {
+				t.Fatalf("publish %s = %d: %s", v, code, b)
+			}
+		}
+		if code, _ := npmDo(t, http.MethodPut, base+"rej", publishBody(t, "rej", []byte("tgz 1.0.0"), publishOpts{shasum: "x"})); code != http.StatusConflict {
+			t.Fatalf("re-PUT 1.0.0 = %d, want 409", code)
+		}
+		if tags := packumentTags(t, base+"rej"); tags["latest"] != "1.1.0" {
+			t.Errorf("latest = %q, want 1.1.0", tags["latest"])
+		}
+	})
+
+	t.Run("mid-publish window", func(t *testing.T) {
+		if code, b := npmDo(t, http.MethodPut, base+"mid", publishBody(t, "mid", []byte("v1"), publishOpts{})); code != http.StatusCreated {
+			t.Fatalf("publish = %d: %s", code, b)
+		}
+		ctx := context.Background()
+		if _, err := h.npmStore(ctx, repo, "mid/-/mid-1.0.1.tgz", []byte("v2"), "application/octet-stream", false); err != nil {
+			t.Fatal(err)
+		}
+		if tags := packumentTags(t, base+"mid"); tags["latest"] != "1.0.0" {
+			t.Errorf("tarball-only 1.0.1: latest = %q", tags["latest"])
+		}
+		if _, err := h.npmStore(ctx, repo, "mid/-/dist-tags/latest", []byte("1.0.1"), "text/plain", true); err != nil {
+			t.Fatal(err)
+		}
+		if tags := packumentTags(t, base+"mid"); tags["latest"] == "1.0.1" {
+			t.Error("latest points at invisible 1.0.1")
+		}
+	})
+
+	t.Run("concurrent identical", func(t *testing.T) {
+		body := publishBody(t, "same", []byte("same tgz"), publishOpts{})
+		codes := concurrentPuts(t, base+"same", body, body)
+		if codes[0]+codes[1] != http.StatusCreated+http.StatusConflict || codes[0] == codes[1] {
+			t.Fatalf("codes = %v, want one 201 and one 409", codes)
+		}
+		if tags := packumentTags(t, base+"same"); len(tags) != 1 || tags["latest"] != "1.0.0" {
+			t.Errorf("tags = %v", tags)
+		}
+	})
+
+	t.Run("concurrent different manifests", func(t *testing.T) {
+		tgz := []byte("race tgz")
+		codes := concurrentPuts(t, base+"race",
+			publishBody(t, "race", tgz, publishOpts{shasum: "a", tags: map[string]string{"a": "1.0.0"}}),
+			publishBody(t, "race", tgz, publishOpts{shasum: "b", tags: map[string]string{"b": "1.0.0"}}))
+		won := map[int]string{}
+		for i, c := range codes {
+			won[c] = []string{"a", "b"}[i]
+		}
+		if len(won) != 2 || won[http.StatusCreated] == "" || won[http.StatusConflict] == "" {
+			t.Fatalf("codes = %v, want one 201 and one 409", codes)
+		}
+		v := packumentVersions(t, base+"race")["1.0.0"].(map[string]any)
+		if got := v["dist"].(map[string]any)["shasum"]; got != won[http.StatusCreated] {
+			t.Errorf("manifest shasum = %v, winner %s", got, won[http.StatusCreated])
+		}
+		tags := packumentTags(t, base+"race")
+		if len(tags) != 1 || tags[won[http.StatusCreated]] != "1.0.0" {
+			t.Errorf("tags = %v, winner %s", tags, won[http.StatusCreated])
+		}
+	})
+
+	t.Run("retry after tarball-only write", func(t *testing.T) {
+		tgz := []byte("half published")
+		if _, err := h.npmStore(context.Background(), repo, "retry/-/retry-1.0.0.tgz", tgz, "application/octet-stream", false); err != nil {
+			t.Fatal(err)
+		}
+		if code, _ := npmDo(t, http.MethodGet, base+"retry", ""); code != http.StatusNotFound {
+			t.Fatalf("tarball-only version visible: %d", code)
+		}
+		if code, _ := npmDo(t, http.MethodPut, base+"retry", publishBody(t, "retry", []byte("other bytes"), publishOpts{})); code != http.StatusConflict {
+			t.Errorf("retry with different bytes = %d, want 409", code)
+		}
+		if code, b := npmDo(t, http.MethodPut, base+"retry", publishBody(t, "retry", tgz, publishOpts{})); code != http.StatusCreated {
+			t.Fatalf("retry with identical bytes = %d: %s", code, b)
+		}
+		if _, ok := packumentVersions(t, base+"retry")["1.0.0"]; !ok {
+			t.Error("retried version missing from packument")
+		}
+	})
+
+	t.Run("concurrent versions", func(t *testing.T) {
+		codes := concurrentPuts(t, base+"conc",
+			publishBody(t, "conc", []byte("tgz 1.0.0"), publishOpts{version: "1.0.0"}),
+			publishBody(t, "conc", []byte("tgz 2.0.0"), publishOpts{version: "2.0.0"}))
+		if codes[0] != http.StatusCreated || codes[1] != http.StatusCreated {
+			t.Fatalf("codes = %v", codes)
+		}
+		v := packumentVersions(t, base+"conc")
+		if _, ok := v["1.0.0"]; !ok {
+			t.Error("1.0.0 missing")
+		}
+		if _, ok := v["2.0.0"]; !ok {
+			t.Error("2.0.0 missing")
+		}
+	})
+
+	t.Run("body too large", func(t *testing.T) {
+		old := npmMaxPublishBytes
+		npmMaxPublishBytes = 64
+		defer func() { npmMaxPublishBytes = old }()
+		if code, _ := npmDo(t, http.MethodPut, base+"big", publishBody(t, "big", []byte("x"), publishOpts{})); code != http.StatusRequestEntityTooLarge {
+			t.Errorf("oversized body = %d, want 413", code)
+		}
+	})
+
+	t.Run("non-npm local", func(t *testing.T) {
+		generic := npmRepo(t, h, models.PackageGeneric)
+		if code, _ := npmDo(t, http.MethodPut, srv.URL+"/api/v1/local/"+generic+"/pkg", publishBody(t, "pkg", []byte("x"), publishOpts{})); code != http.StatusMethodNotAllowed {
+			t.Errorf("publish to generic local = %d, want 405", code)
+		}
+	})
+}
+
+func TestNPMPackumentPerHost(t *testing.T) {
+	h, srv := npmTestServer(t)
+	h.virtualEngine = virtual.NewEngine(h.db, nil)
+	h.virtualEngine.NPMLocal = h.LocalNPMPackument
+	repo := npmRepo(t, h, models.PackageNPM)
+	virt := "v" + repo
+	if err := h.db.CreateVirtual(context.Background(), &models.Virtual{Name: virt, PackageType: models.PackageNPM, Members: []string{repo}}); err != nil {
+		t.Fatal(err)
+	}
+	if code, b := npmDo(t, http.MethodPut, srv.URL+"/api/v1/local/"+repo+"/hostpkg", publishBody(t, "hostpkg", []byte("tgz"), publishOpts{})); code != http.StatusCreated {
+		t.Fatalf("publish = %d: %s", code, b)
+	}
+	routes := h.Routes()
+	get := func(path, host, proto string) *httptest.ResponseRecorder {
+		r := httptest.NewRequest(http.MethodGet, path, nil)
+		r.Host = host
+		if proto != "" {
+			r.Header.Set("X-Forwarded-Proto", proto)
+		}
+		w := httptest.NewRecorder()
+		routes.ServeHTTP(w, r)
+		return w
+	}
+	for _, path := range []string{"/local/" + repo + "/hostpkg", "/virtual/" + virt + "/hostpkg"} {
+		for _, host := range []string{"x@evil.com", "evil.example:"} {
+			if w := get(path, host, ""); w.Code != http.StatusBadRequest {
+				t.Errorf("%s Host %q = %d, want 400", path, host, w.Code)
+			}
+		}
+		w := get(path, "aa.example", "https")
+		if w.Code != http.StatusOK || !strings.Contains(w.Body.String(), `"tarball":"https://aa.example/api/v1/`) {
+			t.Fatalf("%s = %d: %s", path, w.Code, w.Body)
+		}
+		if cc, vary := w.Header().Get("Cache-Control"), w.Header().Get("Vary"); cc != "no-store" || vary != "Host, X-Forwarded-Proto, Accept" {
+			t.Errorf("%s Cache-Control %q Vary %q", path, cc, vary)
+		}
+	}
+}
@@ -0,0 +1,166 @@
+package v1
+
+import (
+	"crypto/sha1"
+	"crypto/sha512"
+	"encoding/base64"
+	"encoding/hex"
+	"encoding/json"
+	"errors"
+	"strings"
+	"testing"
+
+	"git.unkin.net/unkin/artifactapi/internal/database"
+)
+
+func TestNPMPackageName(t *testing.T) {
+	tests := map[string]string{
+		"left-pad":                "left-pad",
+		"left-pad/":               "left-pad",
+		"@acme%2fwidget":          "@acme/widget",
+		"@acme%2Fwidget":          "@acme/widget",
+		"@acme/widget":            "@acme/widget",
+		"left-pad/-/left-pad.tgz": "",
+		"@acme/widget/-/w-1.tgz":  "",
+		"a/b":                     "",
+		"../etc":                  "",
+		"Upper":                   "",
+		"":                        "",
+	}
+	for in, want := range tests {
+		if got := npmPackageName(in); got != want {
+			t.Errorf("npmPackageName(%q) = %q, want %q", in, got, want)
+		}
+	}
+}
+
+type publishOpts struct {
+	bodyName, manifestName, version, integrity, shasum string
+	tags                                               map[string]string
+	extraVersion                                       bool
+}
+
+func publishBody(t *testing.T, pkg string, tgz []byte, o publishOpts) string {
+	t.Helper()
+	sha := sha512.Sum512(tgz)
+	sh1 := sha1.Sum(tgz)
+	if o.bodyName == "" {
+		o.bodyName = pkg
+	}
+	if o.manifestName == "" {
+		o.manifestName = pkg
+	}
+	if o.version == "" {
+		o.version = "1.0.0"
+	}
+	if o.integrity == "" {
+		o.integrity = "sha512-" + base64.StdEncoding.EncodeToString(sha[:])
+	}
+	if o.shasum == "" {
+		o.shasum = hex.EncodeToString(sh1[:])
+	}
+	if o.tags == nil {
+		o.tags = map[string]string{"latest": o.version}
+	}
+	versions := map[string]any{o.version: map[string]any{
+		"name": o.manifestName, "version": o.version, "main": "index.js",
+		"dist": map[string]any{"integrity": o.integrity, "shasum": o.shasum, "tarball": "http://elsewhere/x.tgz"},
+	}}
+	if o.extraVersion {
+		versions["2.0.0"] = map[string]any{"name": pkg, "version": "2.0.0"}
+	}
+	b, _ := json.Marshal(map[string]any{
+		"_id": pkg, "name": o.bodyName, "dist-tags": o.tags, "versions": versions,
+		"_attachments": map[string]any{pkg + "-" + o.version + ".tgz": map[string]any{
+			"content_type": "application/octet-stream", "data": base64.StdEncoding.EncodeToString(tgz), "length": len(tgz),
+		}},
+	})
+	return string(b)
+}
+
+func TestParseNPMPublish(t *testing.T) {
+	tgz := []byte("fake tarball bytes")
+
+	rel, err := parseNPMPublish(strings.NewReader(publishBody(t, "@acme/widget", tgz, publishOpts{tags: map[string]string{"latest": "1.0.0", "beta": "1.0.0"}})), "@acme/widget")
+	if err != nil {
+		t.Fatalf("valid scoped publish rejected: %v", err)
+	}
+	if rel.version != "1.0.0" || string(rel.tarball) != string(tgz) || len(rel.tags) != 2 {
+		t.Fatalf("unexpected release: %+v", rel)
+	}
+
+	if _, err := parseNPMPublish(strings.NewReader(publishBody(t, "left-pad", tgz, publishOpts{integrity: "none", shasum: hex.EncodeToString(make([]byte, 20))})), "left-pad"); err == nil {
+		t.Fatal("shasum fallback should reject a mismatched tarball")
+	}
+	if _, err := parseNPMPublish(strings.NewReader(publishBody(t, "left-pad", tgz, publishOpts{integrity: "none"})), "left-pad"); err != nil {
+		t.Fatalf("matching shasum fallback rejected: %v", err)
+	}
+
+	bad := map[string]publishOpts{
+		"body name mismatch":        {bodyName: "other"},
+		"manifest name mismatch":    {manifestName: "other"},
+		"integrity mismatch":        {integrity: "sha512-AAAA"},
+		"path-like version":         {version: "../1.0.0"},
+		"non-semver version":        {version: "1..2"},
+		"leading-zero version":      {version: "01.0.0"},
+		"tag at another version":    {tags: map[string]string{"latest": "9.9.9"}},
+		"path-like tag":             {tags: map[string]string{"../x": "1.0.0"}},
+		"multiple versions at once": {extraVersion: true},
+	}
+	for name, o := range bad {
+		if _, err := parseNPMPublish(strings.NewReader(publishBody(t, "left-pad", tgz, o)), "left-pad"); err == nil {
+			t.Errorf("%s: expected rejection", name)
+		}
+	}
+	if _, err := parseNPMPublish(strings.NewReader(`{"name":"left-pad","versions":{}}`), "left-pad"); err == nil {
+		t.Error("deprecate-style body without attachments should be rejected")
+	}
+}
+
+func TestBuildNPMPackument(t *testing.T) {
+	blobs := map[string]string{
+		"h1": `{"name":"my_pkg","version":"1.0.0","dist":{"shasum":"aa","tarball":"http://elsewhere/x.tgz"}}`,
+		"h2": `{"name":"my_pkg","version":"1.1.0"}`,
+		"h3": "1.1.0",
+		"h4": "2.0.0",
+		"hx": `{"name":"myXpkg","version":"6.6.6"}`,
+	}
+	read := func(h string) ([]byte, error) {
+		b, ok := blobs[h]
+		if !ok {
+			return nil, errors.New("missing blob")
+		}
+		return []byte(b), nil
+	}
+	files := []database.LocalFile{
+		{FilePath: "my_pkg/-/dist-tags/latest", ContentHash: "h3"},
+		{FilePath: "my_pkg/-/dist-tags/next", ContentHash: "h4"},
+		{FilePath: "my_pkg/-/my_pkg-1.0.0.json", ContentHash: "h1"},
+		{FilePath: "my_pkg/-/my_pkg-1.0.0.tgz", ContentHash: "nope"},
+		{FilePath: "my_pkg/-/my_pkg-1.1.0.json", ContentHash: "h2"},
+		{FilePath: "myXpkg/-/myXpkg-6.6.6.json", ContentHash: "hx"},
+	}
+
+	doc, err := buildNPMPackument("my_pkg", files, read, "http://reg/api/v1/local/npm/")
+	if err != nil {
+		t.Fatal(err)
+	}
+	if tags := doc["dist-tags"].(map[string]string); tags["latest"] != "1.1.0" || len(tags) != 1 {
+		t.Errorf("dist-tags = %v", tags)
+	}
+	versions := doc["versions"].(map[string]any)
+	if len(versions) != 2 {
+		t.Fatalf("versions = %v (LIKE over-match must be dropped)", versions)
+	}
+	v1 := versions["1.0.0"].(map[string]any)["dist"].(map[string]any)
+	if v1["tarball"] != "http://reg/api/v1/local/npm/my_pkg/-/my_pkg-1.0.0.tgz" || v1["shasum"] != "aa" {
+		t.Errorf("1.0.0 dist = %v", v1)
+	}
+	if versions["1.1.0"].(map[string]any)["dist"].(map[string]any)["tarball"] != "http://reg/api/v1/local/npm/my_pkg/-/my_pkg-1.1.0.tgz" {
+		t.Error("1.1.0 missing rewritten tarball")
+	}
+
+	if doc, err := buildNPMPackument("absent", files, read, "x/"); err != nil || doc != nil {
+		t.Errorf("absent package: doc=%v err=%v", doc, err)
+	}
+}
@@ -0,0 +1,26 @@
+package v1
+
+import (
+	"context"
+	"encoding/json"
+
+	"git.unkin.net/unkin/artifactapi/internal/virtual"
+)
+
+// LocalNPMPackument serves a local npm repo's packument to npm virtuals, with
+// tarball paths relative to the repo.
+func (h *ProxyHandler) LocalNPMPackument(ctx context.Context, repo, pkg string) ([]byte, error) {
+	files, err := h.db.ListLocalFilesByPrefix(ctx, repo, pkg+"/-/")
+	if err != nil {
+		return nil, err
+	}
+	read := func(hash string) ([]byte, error) { return h.readLocalBlob(ctx, hash) }
+	doc, err := buildNPMPackument(pkg, files, read, "")
+	if err != nil {
+		return nil, err
+	}
+	if doc == nil {
+		return nil, virtual.ErrNotFound
+	}
+	return json.Marshal(doc)
+}
@@ -6,7 +6,10 @@ import (
 	"fmt"
 	"io"
 	"log/slog"
+	"net"
 	"net/http"
+	"strconv"
+	"strings"

 	"github.com/go-chi/chi/v5"

@@ -16,6 +19,7 @@ import (
 	"git.unkin.net/unkin/artifactapi/internal/proxy"
 	"git.unkin.net/unkin/artifactapi/internal/storage"
 	"git.unkin.net/unkin/artifactapi/internal/virtual"
+	"git.unkin.net/unkin/artifactapi/pkg/models"
 )

 type ProxyHandler struct {
@@ -41,7 +45,9 @@ func NewProxyHandler(engine *proxy.Engine, virtualEngine *virtual.Engine, db *da
 func (h *ProxyHandler) Routes() chi.Router {
 	r := chi.NewRouter()
 	r.Get("/remote/{remoteName}/*", h.handleProxy)
+	r.Head("/remote/{remoteName}/*", h.handleProxyHead)
 	r.Get("/local/{localName}/*", h.handleLocal)
+	r.Put("/local/{localName}/*", h.handleLocalPut)
 	r.Get("/virtual/{virtualName}/*", h.handleVirtual)
 	return r
 }
@@ -68,31 +74,43 @@ func (h *ProxyHandler) handleDockerPing(w http.ResponseWriter, r *http.Request)
 	w.WriteHeader(http.StatusOK)
 }

-func (h *ProxyHandler) handleProxy(w http.ResponseWriter, r *http.Request) {
+// remoteFor resolves the remote and provider named in the request, writing the
+// error response itself when either is missing.
+func (h *ProxyHandler) remoteFor(w http.ResponseWriter, r *http.Request) (*models.Remote, provider.Provider, bool) {
 	remoteName := chi.URLParam(r, "remoteName")
-	path := chi.URLParam(r, "*")
-
 	remote, err := h.db.GetRemote(r.Context(), remoteName)
 	if err != nil {
 		http.Error(w, fmt.Sprintf("remote %q not found", remoteName), http.StatusNotFound)
-		return
+		return nil, nil, false
 	}
-
 	prov, err := provider.Get(remote.PackageType)
 	if err != nil {
 		http.Error(w, fmt.Sprintf("no provider for %q", remote.PackageType), http.StatusInternalServerError)
+		return nil, nil, false
+	}
+	return remote, prov, true
+}
+
+// serveRemote lets metadata-only remotes (e.g. github_rpm) synthesize their own
+// responses and redirect package downloads instead of proxying bytes.
+func (h *ProxyHandler) serveRemote(w http.ResponseWriter, r *http.Request, remote *models.Remote, prov provider.Provider, path string) bool {
+	rs, ok := prov.(provider.RemoteServer)
+	return ok && rs.ServeRemote(w, r, *remote, path, fmt.Sprintf("%s://%s", Scheme(r), r.Host), h.db)
+}
+
+func (h *ProxyHandler) handleProxy(w http.ResponseWriter, r *http.Request) {
+	remote, prov, ok := h.remoteFor(w, r)
+	if !ok {
 		return
 	}
-
-	// Metadata-only remotes (e.g. github_rpm) synthesize their own responses and
-	// redirect package downloads to a backend remote instead of proxying bytes.
-	if rs, ok := prov.(provider.RemoteServer); ok {
-		proxyBaseURL := fmt.Sprintf("%s://%s", scheme(r), r.Host)
-		if rs.ServeRemote(w, r, *remote, path, proxyBaseURL, h.db) {
-			return
-		}
+	path := chi.URLParam(r, "*")
+	if h.serveRemote(w, r, remote, prov, path) {
+		return
 	}
+	h.serveFetch(w, r, remote, prov, path)
+}

+func (h *ProxyHandler) serveFetch(w http.ResponseWriter, r *http.Request, remote *models.Remote, prov provider.Provider, path string) {
 	result, err := h.engine.Fetch(r.Context(), *remote, path, prov, r.Header)
 	if err != nil {
 		var proxyErr *proxy.ProxyError
@@ -100,7 +118,7 @@ func (h *ProxyHandler) handleProxy(w http.ResponseWriter, r *http.Request) {
 			http.Error(w, proxyErr.Message, proxyErr.Status)
 			return
 		}
-		slog.Error("proxy fetch failed", "remote", remoteName, "path", path, "error", err)
+		slog.Error("proxy fetch failed", "remote", remote.Name, "path", path, "error", err)
 		http.Error(w, "bad gateway", http.StatusBadGateway)
 		return
 	}
@@ -108,37 +126,43 @@ func (h *ProxyHandler) handleProxy(w http.ResponseWriter, r *http.Request) {

 	var body io.Reader = result.Reader
 	if sr, ok := prov.(provider.ServeRewriter); ok && prov.Classify(path) == provider.Mutable {
-		raw, err := io.ReadAll(result.Reader)
-		if err != nil {
-			http.Error(w, "bad gateway", http.StatusBadGateway)
+		raw, ok := rewriteServed(w, r, sr, *remote, result.Reader)
+		if !ok {
 			return
 		}
-		raw = sr.RewriteServed(raw, *remote, fmt.Sprintf("%s://%s", scheme(r), r.Host))
 		body, result.Size = bytes.NewReader(raw), int64(len(raw))
 	}

 	w.Header().Set("Content-Type", result.ContentType)
 	w.Header().Set("X-Artifact-Source", result.Source)
 	if result.Size > 0 {
+		w.Header().Set("Content-Length", fmt.Sprintf("%d", result.Size))
 		w.Header().Set("X-Artifact-Size", fmt.Sprintf("%d", result.Size))
 	}
 	w.WriteHeader(http.StatusOK)
 	io.Copy(w, body)
 }

-func (h *ProxyHandler) handleProxyHead(w http.ResponseWriter, r *http.Request) {
-	remoteName := chi.URLParam(r, "remoteName")
-	path := chi.URLParam(r, "*")
+func rewritesMutable(prov provider.Provider, path string) bool {
+	_, sr := prov.(provider.ServeRewriter)
+	_, rr := prov.(provider.ResponseRewriter)
+	return (sr || rr) && prov.Classify(path) == provider.Mutable
+}

-	remote, err := h.db.GetRemote(r.Context(), remoteName)
-	if err != nil {
-		http.Error(w, fmt.Sprintf("remote %q not found", remoteName), http.StatusNotFound)
+// handleProxyHead answers HEAD like GET without a body. Responses a provider
+// synthesizes or rewrites per request go through the GET path so their headers
+// match; everything else is answered from store metadata or an upstream HEAD.
+func (h *ProxyHandler) handleProxyHead(w http.ResponseWriter, r *http.Request) {
+	remote, prov, ok := h.remoteFor(w, r)
+	if !ok {
 		return
 	}
-
-	prov, err := provider.Get(remote.PackageType)
-	if err != nil {
-		http.Error(w, fmt.Sprintf("no provider for %q", remote.PackageType), http.StatusInternalServerError)
+	path := chi.URLParam(r, "*")
+	if h.serveRemote(w, r, remote, prov, path) {
+		return
+	}
+	if rewritesMutable(prov, path) {
+		h.serveFetch(w, r, remote, prov, path)
 		return
 	}

@@ -149,7 +173,7 @@ func (h *ProxyHandler) handleProxyHead(w http.ResponseWriter, r *http.Request) {
 			http.Error(w, proxyErr.Message, proxyErr.Status)
 			return
 		}
-		slog.Error("proxy head failed", "remote", remoteName, "path", path, "error", err)
+		slog.Error("proxy head failed", "remote", remote.Name, "path", path, "error", err)
 		http.Error(w, "bad gateway", http.StatusBadGateway)
 		return
 	}
@@ -172,7 +196,10 @@ func (h *ProxyHandler) handleVirtual(w http.ResponseWriter, r *http.Request) {
 		return
 	}

-	proxyBaseURL := fmt.Sprintf("%s://%s", scheme(r), r.Host)
+	proxyBaseURL, ok := requestBaseURL(w, r)
+	if !ok {
+		return
+	}

 	loc, ok, err := h.virtualEngine.MemberRedirect(r.Context(), *virt, path, r.URL.RawQuery, proxyBaseURL)
 	if err != nil {
@@ -184,7 +211,14 @@ func (h *ProxyHandler) handleVirtual(w http.ResponseWriter, r *http.Request) {
 		return
 	}

-	body, contentType, err := h.virtualEngine.Fetch(r.Context(), *virt, path, proxyBaseURL)
+	var body []byte
+	var contentType string
+	if virt.PackageType == models.PackageNPM {
+		abbreviated := strings.Contains(r.Header.Get("Accept"), virtual.NPMAbbreviatedType)
+		body, contentType, err = h.virtualEngine.FetchNPM(r.Context(), *virt, path, proxyBaseURL, abbreviated)
+	} else {
+		body, contentType, err = h.virtualEngine.Fetch(r.Context(), *virt, path, proxyBaseURL)
+	}
 	if errors.Is(err, virtual.ErrNotFound) {
 		http.Error(w, "not found", http.StatusNotFound)
 		return
@@ -195,6 +229,9 @@ func (h *ProxyHandler) handleVirtual(w http.ResponseWriter, r *http.Request) {
 		return
 	}

+	if virt.PackageType == models.PackageNPM {
+		setNPMPerHostHeaders(w)
+	}
 	w.Header().Set("Content-Type", contentType)
 	w.Header().Set("X-Artifact-Source", "virtual")
 	w.WriteHeader(http.StatusOK)
@@ -211,6 +248,13 @@ func (h *ProxyHandler) handleLocal(w http.ResponseWriter, r *http.Request) {
 		return
 	}

+	if remote.PackageType == models.PackageNPM {
+		if pkg := npmPackageName(path); pkg != "" {
+			h.serveNPMPackument(w, r, remote, pkg)
+			return
+		}
+	}
+
 	prov, _ := provider.Get(remote.PackageType)
 	if indexer, ok := prov.(provider.LocalIndexer); ok {
 		if indexer.ServeLocalIndex(w, r, h.db, remote.Name, path) {
@@ -249,12 +293,78 @@ func (h *ProxyHandler) serveLocalFile(w http.ResponseWriter, r *http.Request, re
 	io.Copy(w, reader)
 }

-func scheme(r *http.Request) string {
+// Scheme returns the client-facing scheme: TLS, else the first
+// X-Forwarded-Proto value when it is http or https, else http.
+func Scheme(r *http.Request) string {
 	if r.TLS != nil {
 		return "https"
 	}
-	if fwd := r.Header.Get("X-Forwarded-Proto"); fwd != "" {
+	fwd, _, _ := strings.Cut(r.Header.Get("X-Forwarded-Proto"), ",")
+	if fwd = strings.ToLower(strings.TrimSpace(fwd)); fwd == "https" {
 		return fwd
 	}
 	return "http"
 }
+
+// rewriteServed rewrites a cached body for the requesting host and marks the
+// response uncacheable by shared caches, since it varies per Host.
+func rewriteServed(w http.ResponseWriter, r *http.Request, sr provider.ServeRewriter, remote models.Remote, src io.Reader) ([]byte, bool) {
+	base, ok := requestBaseURL(w, r)
+	if !ok {
+		return nil, false
+	}
+	raw, err := io.ReadAll(src)
+	if err != nil {
+		http.Error(w, "bad gateway", http.StatusBadGateway)
+		return nil, false
+	}
+	raw = sr.RewriteServed(raw, remote, base)
+	w.Header().Set("Cache-Control", "no-store")
+	w.Header().Set("Vary", "Host, X-Forwarded-Proto")
+	w.Header().Set("Content-Length", strconv.Itoa(len(raw)))
+	return raw, true
+}
+
+// requestBaseURL returns the client-facing scheme://host, answering 400 on an
+// invalid Host.
+func requestBaseURL(w http.ResponseWriter, r *http.Request) (string, bool) {
+	if !validHost(r.Host) {
+		http.Error(w, "invalid host header", http.StatusBadRequest)
+		return "", false
+	}
+	return Scheme(r) + "://" + r.Host, true
+}
+
+// validHost accepts host[:port] where host is a DNS name or IP literal.
+func validHost(hostport string) bool {
+	host, port := hostport, ""
+	if h, p, err := net.SplitHostPort(hostport); err == nil {
+		if p == "" {
+			return false
+		}
+		host, port = h, p
+	} else if strings.HasPrefix(hostport, "[") {
+		return strings.HasSuffix(hostport, "]") && net.ParseIP(hostport[1:len(hostport)-1]) != nil
+	}
+	if port != "" {
+		if n, err := strconv.Atoi(port); err != nil || n < 1 || n > 65535 {
+			return false
+		}
+	}
+	if net.ParseIP(host) != nil {
+		return true
+	}
+	if strings.HasPrefix(hostport, "[") {
+		return false
+	}
+	if host == "" || len(host) > 253 {
+		return false
+	}
+	for _, label := range strings.Split(strings.TrimSuffix(host, "."), ".") {
+		if label == "" || len(label) > 63 || label[0] == '-' || label[len(label)-1] == '-' ||
+			strings.Trim(strings.ToLower(label), "abcdefghijklmnopqrstuvwxyz0123456789-_") != "" {
+			return false
+		}
+	}
+	return true
+}
@@ -0,0 +1,74 @@
+package v1
+
+import (
+	"net/http"
+	"net/http/httptest"
+	"strings"
+	"testing"
+
+	"git.unkin.net/unkin/artifactapi/internal/provider/nuget"
+	"git.unkin.net/unkin/artifactapi/pkg/models"
+)
+
+func TestValidHost(t *testing.T) {
+	for _, h := range []string{"aa.example", "aa.example:8443", "artifactapi.k8s.syd1.au.unkin.net", "10.0.0.1:80", "[::1]:8080", "[::1]", "::1", "localhost", "my_host.local"} {
+		if !validHost(h) {
+			t.Errorf("validHost(%q) = false, want true", h)
+		}
+	}
+	for _, h := range []string{"", "evil\".example", "a b", "aa.example:0", "aa.example:99999", "aa.example:x", "aa.example:", ":8080", "-bad.example", "a..b", "[::1", "[nope]:80", "a/b", "a<script>", strings.Repeat("a", 64) + ".example"} {
+		if validHost(h) {
+			t.Errorf("validHost(%q) = true, want false", h)
+		}
+	}
+}
+
+var nugetOrg = models.Remote{Name: "nuget-org", BaseURL: "https://api.nuget.org/v3/index.json"}
+
+const serviceIndex = `{"resources":[{"@id":"https://api.nuget.org/v3-flatcontainer/"}]}`
+
+func rewrite(t *testing.T, host, proto string) (*httptest.ResponseRecorder, string, bool) {
+	t.Helper()
+	r := httptest.NewRequest(http.MethodGet, "/api/v1/remote/nuget-org/index.json", nil)
+	r.Host = host
+	if proto != "" {
+		r.Header.Set("X-Forwarded-Proto", proto)
+	}
+	w := httptest.NewRecorder()
+	raw, ok := rewriteServed(w, r, &nuget.Provider{}, nugetOrg, strings.NewReader(serviceIndex))
+	return w, string(raw), ok
+}
+
+func TestRewriteServedHeaders(t *testing.T) {
+	w, body, ok := rewrite(t, "aa.example:8443", "https")
+	if !ok {
+		t.Fatal("valid request rejected")
+	}
+	if want := `{"resources":[{"@id":"https://aa.example:8443/api/v1/remote/nuget-org/v3-flatcontainer/"}]}`; body != want {
+		t.Errorf("body = %s, want %s", body, want)
+	}
+	if got := w.Header().Get("Cache-Control"); got != "no-store" {
+		t.Errorf("Cache-Control = %q, want no-store", got)
+	}
+	if got := w.Header().Get("Vary"); got != "Host, X-Forwarded-Proto" {
+		t.Errorf("Vary = %q", got)
+	}
+	if got := w.Header().Get("Content-Length"); got != "91" || len(body) != 91 {
+		t.Errorf("Content-Length = %q, body len %d", got, len(body))
+	}
+}
+
+func TestRewriteServedIgnoresBogusProto(t *testing.T) {
+	_, plain, _ := rewrite(t, "aa.example", "")
+	_, bogus, _ := rewrite(t, "aa.example", "javascript")
+	if bogus != plain || !strings.Contains(bogus, `"http://aa.example/`) {
+		t.Errorf("bogus X-Forwarded-Proto changed the body: %s", bogus)
+	}
+}
+
+func TestRewriteServedRejectsInvalidHost(t *testing.T) {
+	w, _, ok := rewrite(t, `evil"},{"x":"`, "")
+	if ok || w.Code != http.StatusBadRequest {
+		t.Errorf("invalid host: ok=%v code=%d, want rejected with 400", ok, w.Code)
+	}
+}
@@ -0,0 +1,41 @@
+package v1
+
+import (
+	"net/http"
+	"testing"
+
+	"github.com/go-chi/chi/v5"
+
+	"git.unkin.net/unkin/artifactapi/internal/provider"
+	"git.unkin.net/unkin/artifactapi/internal/provider/npm"
+	"git.unkin.net/unkin/artifactapi/internal/provider/nuget"
+	"git.unkin.net/unkin/artifactapi/internal/provider/rpm"
+)
+
+func TestRoutesRemoteHead(t *testing.T) {
+	r := (&ProxyHandler{}).Routes()
+	for _, m := range []string{http.MethodGet, http.MethodHead} {
+		if !r.Match(chi.NewRouteContext(), m, "/remote/myremote/some/path") {
+			t.Errorf("%s /remote/myremote/some/path not routed", m)
+		}
+	}
+}
+
+func TestRewritesMutable(t *testing.T) {
+	tests := []struct {
+		name string
+		prov provider.Provider
+		path string
+		want bool
+	}{
+		{"npm packument", &npm.Provider{}, "left-pad", true},
+		{"npm tarball", &npm.Provider{}, "left-pad/-/left-pad-1.3.0.tgz", false},
+		{"nuget index", &nuget.Provider{}, "index.json", true},
+		{"rpm repodata", &rpm.Provider{}, "repodata/repomd.xml", false},
+	}
+	for _, tc := range tests {
+		if got := rewritesMutable(tc.prov, tc.path); got != tc.want {
+			t.Errorf("%s: rewritesMutable = %v, want %v", tc.name, got, tc.want)
+		}
+	}
+}
@@ -7,14 +7,28 @@ import (
 )

 func TestScheme(t *testing.T) {
-	if got := scheme(&http.Request{TLS: &tls.ConnectionState{}}); got != "https" {
+	if got := Scheme(&http.Request{TLS: &tls.ConnectionState{}}); got != "https" {
 		t.Errorf("TLS request scheme = %q, want https", got)
 	}
 	r := &http.Request{Header: http.Header{"X-Forwarded-Proto": {"https"}}}
-	if got := scheme(r); got != "https" {
+	if got := Scheme(r); got != "https" {
 		t.Errorf("X-Forwarded-Proto scheme = %q, want https", got)
 	}
-	if got := scheme(&http.Request{Header: http.Header{}}); got != "http" {
+	for fwd, want := range map[string]string{
+		"https, http": "https",
+		"http, https": "http",
+		"HTTPS":       "https",
+		" https ":     "https",
+		"javascript":  "http",
+		"ftp":         "http",
+		"https;evil":  "http",
+	} {
+		r := &http.Request{Header: http.Header{"X-Forwarded-Proto": {fwd}}}
+		if got := Scheme(r); got != want {
+			t.Errorf("X-Forwarded-Proto %q scheme = %q, want %s", fwd, got, want)
+		}
+	}
+	if got := Scheme(&http.Request{Header: http.Header{}}); got != "http" {
 		t.Errorf("default scheme = %q, want http", got)
 	}
 }
@@ -67,8 +67,12 @@ func (h *RemotesHandler) get(w http.ResponseWriter, r *http.Request) {
 	writeJSON(w, http.StatusOK, remote)
 }

+// defaultMutableTTL matches the remotes.mutable_ttl column default; a create
+// that omits mutable_ttl gets it, an explicit 0 still means never expire.
+const defaultMutableTTL = 3600
+
 func (h *RemotesHandler) create(w http.ResponseWriter, r *http.Request) {
-	var remote models.Remote
+	remote := models.Remote{MutableTTL: defaultMutableTTL}
 	if err := json.NewDecoder(r.Body).Decode(&remote); err != nil {
 		http.Error(w, "invalid json", http.StatusBadRequest)
 		return
@@ -114,7 +118,15 @@ func (h *RemotesHandler) create(w http.ResponseWriter, r *http.Request) {

 func (h *RemotesHandler) update(w http.ResponseWriter, r *http.Request) {
 	name := chi.URLParam(r, "name")
+	// Capture the current backend before the update so we can tell whether the
+	// remote's base_url (its upstream) changed. A read failure just means we
+	// skip the freshness flush; it must not block the update.
+	existing, err := h.db.GetRemote(r.Context(), name)
+	oldKnown := err == nil
 	var remote models.Remote
+	if oldKnown {
+		remote.MutableTTL = existing.MutableTTL
+	}
 	if err := json.NewDecoder(r.Body).Decode(&remote); err != nil {
 		http.Error(w, "invalid json", http.StatusBadRequest)
 		return
@@ -132,13 +144,6 @@ func (h *RemotesHandler) update(w http.ResponseWriter, r *http.Request) {
 		http.Error(w, err.Error(), http.StatusBadRequest)
 		return
 	}
-	// Capture the current backend before the update so we can tell whether the
-	// remote's base_url (its upstream) changed. A read failure just means we
-	// skip the freshness flush; it must not block the update.
-	oldBaseURL, oldKnown := "", false
-	if existing, err := h.db.GetRemote(r.Context(), name); err == nil {
-		oldBaseURL, oldKnown = existing.BaseURL, true
-	}
 	if err := h.db.UpdateRemote(r.Context(), &remote); err != nil {
 		http.Error(w, err.Error(), http.StatusInternalServerError)
 		return
@@ -147,13 +152,13 @@ func (h *RemotesHandler) update(w http.ResponseWriter, r *http.Request) {
 	// Release / APKINDEX): purge it so the next request re-fetches from the new
 	// upstream instead of serving stale data until TTL expiry. A flush failure
 	// is logged but does not fail the request — the DB update already landed.
-	if oldKnown && oldBaseURL != remote.BaseURL && h.cache != nil {
+	if oldKnown && existing.BaseURL != remote.BaseURL && h.cache != nil {
 		if err := h.cache.FlushRemote(r.Context(), name); err != nil {
 			slog.Warn("flush cached metadata after base_url change failed",
 				"remote", name, "error", err)
 		} else {
 			slog.Info("flushed cached metadata after base_url change",
-				"remote", name, "old_base_url", oldBaseURL, "new_base_url", remote.BaseURL)
+				"remote", name, "old_base_url", existing.BaseURL, "new_base_url", remote.BaseURL)
 		}
 	}
 	writeJSON(w, http.StatusOK, remote)
@@ -165,6 +170,11 @@ func (h *RemotesHandler) del(w http.ResponseWriter, r *http.Request) {
 		http.Error(w, err.Error(), http.StatusInternalServerError)
 		return
 	}
+	if h.cache != nil {
+		if err := h.cache.FlushRemote(r.Context(), name); err != nil {
+			slog.Warn("flush cached metadata after delete failed", "remote", name, "error", err)
+		}
+	}
 	w.WriteHeader(http.StatusNoContent)
 }

@@ -94,3 +94,28 @@ func TestUpdateFlushFailureStillSucceeds(t *testing.T) {
 		t.Fatalf("flush calls = %v, want exactly one attempted flush", ff.calls)
 	}
 }
+
+// Deleting a remote flushes its cached state so a re-created remote of the
+// same name starts cold; a flush error does not fail the delete.
+func TestDeleteFlushesCache(t *testing.T) {
+	if testDSN == "" {
+		t.Skip("Docker unavailable")
+	}
+	db, err := database.New(testDSN)
+	if err != nil {
+		t.Fatal(err)
+	}
+	defer db.Close()
+
+	for _, ff := range []*fakeFlusher{{}, {err: errors.New("redis down")}} {
+		const name = "rpm-flush-delete"
+		seedRemote(t, db, name, "https://example.com/repo")
+		h := NewRemotesHandler(db, ff, nil).Routes()
+		if c := do(t, h, "DELETE", "/"+name, ""); c != 204 {
+			t.Fatalf("delete = %d, want 204", c)
+		}
+		if len(ff.calls) != 1 || ff.calls[0] != name {
+			t.Fatalf("flush calls = %v, want exactly one flush of %q", ff.calls, name)
+		}
+	}
+}
@@ -0,0 +1,57 @@
+package v2
+
+import (
+	"context"
+	"testing"
+
+	"git.unkin.net/unkin/artifactapi/internal/database"
+)
+
+// An omitted mutable_ttl defaults to 3600 on create and keeps the stored value
+// on update; an explicit 0 is stored as 0 in both.
+func TestMutableTTLOmittedVsZero(t *testing.T) {
+	if testDSN == "" {
+		t.Skip("Docker unavailable")
+	}
+	db, err := database.New(testDSN)
+	if err != nil {
+		t.Fatal(err)
+	}
+	defer db.Close()
+	h := NewRemotesHandler(db, nil, nil).Routes()
+	ttl := func(name string) int {
+		t.Helper()
+		r, err := db.GetRemote(context.Background(), name)
+		if err != nil {
+			t.Fatalf("get %s: %v", name, err)
+		}
+		return r.MutableTTL
+	}
+	const body = `"package_type":"rpm","base_url":"https://example.com/repo"`
+
+	steps := []struct {
+		method, path, json string
+		want               int
+	}{
+		{"POST", "/", `{"name":"ttl-omit",` + body + `}`, 3600},
+		{"PUT", "/ttl-omit", `{` + body + `,"mutable_ttl":120}`, 120},
+		{"PUT", "/ttl-omit", `{` + body + `}`, 120},
+		{"PUT", "/ttl-omit", `{` + body + `,"mutable_ttl":0}`, 0},
+		{"PUT", "/ttl-omit", `{` + body + `}`, 0},
+	}
+	for i, s := range steps {
+		if c := do(t, h, s.method, s.path, s.json); c/100 != 2 {
+			t.Fatalf("step %d %s %s = %d", i, s.method, s.path, c)
+		}
+		if got := ttl("ttl-omit"); got != s.want {
+			t.Errorf("step %d: mutable_ttl = %d, want %d", i, got, s.want)
+		}
+	}
+
+	if c := do(t, h, "POST", "/", `{"name":"ttl-zero",`+body+`,"mutable_ttl":0}`); c != 201 {
+		t.Fatalf("create explicit 0 = %d", c)
+	}
+	if got := ttl("ttl-zero"); got != 0 {
+		t.Errorf("explicit 0 stored as %d", got)
+	}
+}
@@ -86,6 +86,12 @@ func TestETagAndToken(t *testing.T) {
 	if v, _ := testRedis.GetETag(ctx, "r", "p"); v != `"abc"` {
 		t.Errorf("etag = %q", v)
 	}
+	if err := testRedis.SetETag(ctx, "r", "p", "", 0); err != nil {
+		t.Fatal(err)
+	}
+	if v, _ := testRedis.GetETag(ctx, "r", "p"); v != "" {
+		t.Errorf("empty etag should drop the stored one, got %q", v)
+	}

 	if v, _ := testRedis.GetToken(ctx, "missing"); v != "" {
 		t.Error("missing token should be empty")
@@ -57,8 +57,12 @@ func (r *Redis) ReleaseLock(ctx context.Context, remote, path string) error {
 	return r.client.Del(ctx, key).Err()
 }

+// SetETag stores etag for path, or drops the stored one when etag is empty.
 func (r *Redis) SetETag(ctx context.Context, remote, path, etag string, ttl time.Duration) error {
 	key := fmt.Sprintf("etag:%s:%s", remote, path)
+	if etag == "" {
+		return r.client.Del(ctx, key).Err()
+	}
 	return r.client.Set(ctx, key, etag, ttl).Err()
 }

@@ -0,0 +1,3 @@
+package all
+
+import _ "git.unkin.net/unkin/artifactapi/internal/provider/alpine"
@@ -0,0 +1,3 @@
+package all
+
+import _ "git.unkin.net/unkin/artifactapi/internal/provider/cargo"
@@ -0,0 +1,3 @@
+package all
+
+import _ "git.unkin.net/unkin/artifactapi/internal/provider/deb"
@@ -0,0 +1,3 @@
+package all
+
+import _ "git.unkin.net/unkin/artifactapi/internal/provider/docker"
@@ -0,0 +1,3 @@
+package all
+
+import _ "git.unkin.net/unkin/artifactapi/internal/provider/gems"
@@ -0,0 +1,3 @@
+package all
+
+import _ "git.unkin.net/unkin/artifactapi/internal/provider/generic"
@@ -0,0 +1,3 @@
+package all
+
+import _ "git.unkin.net/unkin/artifactapi/internal/provider/goproxy"
@@ -0,0 +1,3 @@
+package all
+
+import _ "git.unkin.net/unkin/artifactapi/internal/provider/helm"
@@ -0,0 +1,3 @@
+package all
+
+import _ "git.unkin.net/unkin/artifactapi/internal/provider/npm"
@@ -0,0 +1,3 @@
+package all
+
+import _ "git.unkin.net/unkin/artifactapi/internal/provider/nuget"
@@ -0,0 +1,3 @@
+package all
+
+import _ "git.unkin.net/unkin/artifactapi/internal/provider/puppet"
@@ -0,0 +1,3 @@
+package all
+
+import _ "git.unkin.net/unkin/artifactapi/internal/provider/pypi"
@@ -0,0 +1,3 @@
+package all
+
+import _ "git.unkin.net/unkin/artifactapi/internal/provider/rpm"
@@ -0,0 +1,3 @@
+package all
+
+import _ "git.unkin.net/unkin/artifactapi/internal/provider/terraform"
@@ -0,0 +1,15 @@
+package provider
+
+import (
+	"context"
+	"io"
+)
+
+type FileStore interface {
+	ListFilesByPrefix(ctx context.Context, repoName, prefix string) ([]FileEntry, error)
+	ListPackages(ctx context.Context, repoName string) ([]string, error)
+}
+
+type BlobReader interface {
+	Download(ctx context.Context, key string) (io.ReadCloser, int64, error)
+}
@@ -0,0 +1,10 @@
+package provider
+
+type IndexMerger interface {
+	MergeIndexes(members []MemberIndex, proxyBaseURL string) ([]byte, error)
+}
+
+type MemberIndex struct {
+	RemoteName string
+	Body       []byte
+}
@@ -0,0 +1,11 @@
+package provider
+
+import (
+	"context"
+	"net/http"
+)
+
+type LocalIndexer interface {
+	ServeLocalIndex(w http.ResponseWriter, r *http.Request, files FileStore, repoName, path string) bool
+	GenerateLocalIndex(ctx context.Context, files FileStore, repoName, path string) ([]byte, error)
+}
@@ -0,0 +1,6 @@
+package provider
+
+type LocalUploader interface {
+	ValidateUpload(filePath string) (storagePath, contentType string, err error)
+	UploadResponse(storagePath, contentHash string, sizeBytes int64) map[string]any
+}
@@ -0,0 +1,11 @@
+package provider
+
+import (
+	"context"
+)
+
+// PostDeleteHook lets a provider clean up derived state (e.g. RPM metadata that
+// feeds generated repodata) after a local file is removed.
+type PostDeleteHook interface {
+	AfterDelete(ctx context.Context, repoName, storagePath string, db MetadataDeleter) error
+}
@@ -0,0 +1,9 @@
+package provider
+
+import (
+	"context"
+)
+
+type PostUploadHook interface {
+	AfterUpload(ctx context.Context, repoName, storagePath, contentHash string, blobs BlobReader, db MetadataStore)
+}
@@ -0,0 +1,25 @@
+package provider
+
+import (
+	"net/http"
+
+	"git.unkin.net/unkin/artifactapi/pkg/models"
+)
+
+// RemoteServer lets a remote provider fully answer a request itself instead of
+// going through the byte-proxy engine. It is the remote-side analog of
+// LocalIndexer: a metadata-only remote (e.g. github_rpm) uses it to synthesize
+// repodata from derived per-asset metadata and to redirect package downloads to
+// a backend remote, without ever precaching the packages. Returning false lets
+// the normal proxy path take over.
+type RemoteServer interface {
+	ServeRemote(w http.ResponseWriter, r *http.Request, remote models.Remote, path, proxyBaseURL string, store RemoteMetadataStore) bool
+}
+
+// RemoteMetadataStore is the persistence surface a RemoteServer needs to cache
+// and read the metadata it derives per upstream asset. *database.DB satisfies it.
+type RemoteMetadataStore interface {
+	RPMMetadataReader
+	MetadataStore
+	MetadataDeleter
+}
@@ -0,0 +1,7 @@
+package provider
+
+// ResponseRewriter marks providers whose RewriteResponse changes mutable
+// bodies, so the served length is only known after a fetch.
+type ResponseRewriter interface {
+	RewritesResponse()
+}
@@ -0,0 +1,12 @@
+package provider
+
+import (
+	"git.unkin.net/unkin/artifactapi/pkg/models"
+)
+
+// ServeRewriter rewrites a cached mutable remote response per request, for
+// indexes whose URLs must name the requesting host and so cannot be rewritten
+// before caching.
+type ServeRewriter interface {
+	RewriteServed(body []byte, remote models.Remote, proxyBaseURL string) []byte
+}
@@ -0,0 +1,62 @@
+// Package gems proxies a RubyGems source: mutable compact index (versions,
+// info/*, names) and legacy specs indexes, plus immutable .gem and
+// quick/Marshal.4.8 gemspec downloads.
+package gems
+
+import (
+	"context"
+	"net/http"
+	"strings"
+
+	"git.unkin.net/unkin/artifactapi/internal/auth"
+	"git.unkin.net/unkin/artifactapi/internal/provider"
+	"git.unkin.net/unkin/artifactapi/pkg/models"
+)
+
+func init() {
+	provider.Register(&Provider{})
+}
+
+type Provider struct{}
+
+func (p *Provider) Type() models.PackageType { return models.PackageGems }
+
+func (p *Provider) Classify(path string) provider.Mutability {
+	if isGem(path) || isQuickSpec(path) {
+		return provider.Immutable
+	}
+	return provider.Mutable
+}
+
+func (p *Provider) ContentType(path string) string {
+	path = strings.TrimLeft(path, "/")
+	if path == "versions" || path == "names" || strings.HasPrefix(path, "info/") {
+		return "text/plain; charset=utf-8"
+	}
+	if strings.HasPrefix(path, "api/v1/") && strings.HasSuffix(path, ".json") {
+		return "application/json"
+	}
+	return "application/octet-stream"
+}
+
+func (p *Provider) UpstreamURL(remote models.Remote, path string) string {
+	return strings.TrimRight(remote.BaseURL, "/") + "/" + strings.TrimLeft(path, "/")
+}
+
+func (p *Provider) RewriteResponse(_ []byte, _ models.Remote, _ string) ([]byte, error) {
+	return nil, nil
+}
+
+func (p *Provider) AuthHeaders(_ context.Context, remote models.Remote) (http.Header, error) {
+	return auth.BasicHeaders(remote), nil
+}
+
+func isGem(path string) bool {
+	path = strings.TrimLeft(path, "/")
+	return strings.HasPrefix(path, "gems/") && strings.HasSuffix(path, ".gem")
+}
+
+func isQuickSpec(path string) bool {
+	path = strings.TrimLeft(path, "/")
+	return strings.HasPrefix(path, "quick/Marshal.4.8/") && strings.HasSuffix(path, ".gemspec.rz")
+}
@@ -0,0 +1,61 @@
+package gems
+
+import (
+	"testing"
+
+	"git.unkin.net/unkin/artifactapi/internal/provider"
+	"git.unkin.net/unkin/artifactapi/pkg/models"
+)
+
+func TestClassify(t *testing.T) {
+	p := &Provider{}
+	tests := []struct {
+		path string
+		want provider.Mutability
+	}{
+		{"gems/rack-3.1.8.gem", provider.Immutable},
+		{"/gems/nokogiri-1.16.7-x86_64-linux.gem", provider.Immutable},
+		{"quick/Marshal.4.8/rack-3.1.8.gemspec.rz", provider.Immutable},
+		{"versions", provider.Mutable},
+		{"names", provider.Mutable},
+		{"info/rack", provider.Mutable},
+		{"info/gems", provider.Mutable},
+		{"specs.4.8.gz", provider.Mutable},
+		{"latest_specs.4.8.gz", provider.Mutable},
+		{"prerelease_specs.4.8.gz", provider.Mutable},
+		{"api/v1/dependencies", provider.Mutable},
+		{"quick/Marshal.4.8/", provider.Mutable},
+	}
+	for _, tt := range tests {
+		if got := p.Classify(tt.path); got != tt.want {
+			t.Errorf("Classify(%q) = %v, want %v", tt.path, got, tt.want)
+		}
+	}
+}
+
+func TestContentType(t *testing.T) {
+	p := &Provider{}
+	tests := map[string]string{
+		"versions":                 "text/plain; charset=utf-8",
+		"info/rack":                "text/plain; charset=utf-8",
+		"names":                    "text/plain; charset=utf-8",
+		"gems/rack-3.1.8.gem":      "application/octet-stream",
+		"specs.4.8.gz":             "application/octet-stream",
+		"api/v1/gems/rack.json":    "application/json",
+		"api/v1/dependencies.json": "application/json",
+		"api/v1/dependencies":      "application/octet-stream",
+	}
+	for path, want := range tests {
+		if got := p.ContentType(path); got != want {
+			t.Errorf("ContentType(%q) = %q, want %q", path, got, want)
+		}
+	}
+}
+
+func TestUpstreamURL(t *testing.T) {
+	p := &Provider{}
+	r := models.Remote{BaseURL: "https://rubygems.org/"}
+	if got := p.UpstreamURL(r, "/info/rack"); got != "https://rubygems.org/info/rack" {
+		t.Errorf("UpstreamURL = %q", got)
+	}
+}
@@ -51,6 +51,8 @@ func (p *Provider) UpstreamURL(remote models.Remote, path string) string {
 	return strings.TrimRight(remote.BaseURL, "/") + "/" + strings.TrimLeft(path, "/")
 }

+func (p *Provider) RewritesResponse() {}
+
 func (p *Provider) RewriteResponse(body []byte, remote models.Remote, proxyBaseURL string) ([]byte, error) {
 	if proxyBaseURL == "" {
 		return nil, nil
@@ -0,0 +1,49 @@
+package provider
+
+import (
+	"context"
+)
+
+// AlpineMetadataStore / AlpineMetadataDeleter / AlpineMetadataReader are the
+// Alpine-specific persistence surfaces. They are kept separate from the shared
+// RPM/Deb metadata interfaces so the apk provider can type-assert the generic
+// MetadataStore/MetadataDeleter/FileStore it is handed without widening (and
+// thus perturbing the test doubles of) the rpm and deb providers. *database.DB
+// satisfies all three.
+type AlpineMetadataStore interface {
+	InsertAlpineMetadata(ctx context.Context, meta *AlpineMetadata) error
+}
+
+type AlpineMetadataDeleter interface {
+	DeleteAlpineMetadata(ctx context.Context, repoName, filePath string) error
+}
+
+type AlpineMetadataReader interface {
+	ListAlpineMetadataEntries(ctx context.Context, repoName string) ([]AlpineMetadata, error)
+}
+
+// AlpineMetadata is the derived per-package metadata for an Alpine .apk, holding
+// the fields an APKINDEX record carries plus the apk pull checksum (Q1…, the
+// sha1 of the control gzip stream) and the download/installed sizes.
+type AlpineMetadata struct {
+	RepoName         string
+	FilePath         string
+	ContentHash      string
+	Checksum         string   // C: "Q1" + base64(sha1(control gzip stream))
+	Name             string   // P:
+	Version          string   // V:
+	Arch             string   // A:
+	DownloadSize     int64    // S: on-disk .apk size
+	InstalledSize    int64    // I: unpacked size from .PKGINFO
+	Description      string   // T:
+	URL              string   // U:
+	License          string   // L:
+	Origin           string   // o:
+	Maintainer       string   // m:
+	BuildTime        int64    // t:
+	Commit           string   // c:
+	ProviderPriority string   // k:
+	Depends          []string // D:
+	Provides         []string // p:
+	InstallIf        []string // i:
+}
@@ -0,0 +1,34 @@
+package provider
+
+import (
+	"context"
+	"time"
+)
+
+// DebMetadataReader is the read surface the deb LocalIndexer needs to
+// regenerate a flat apt repository (Packages/Release) from stored rows.
+// *database.DB satisfies it; ServeLocalIndex type-asserts the FileStore to it,
+// mirroring how the rpm provider reaches its RPMMetadataReader.
+type DebMetadataReader interface {
+	ListDebMetadataEntries(ctx context.Context, repoName string) ([]DebMetadata, error)
+}
+
+// DebMetadata is the derived per-package metadata for a Debian .deb, carrying
+// the full raw control stanza so the Packages index can be regenerated
+// faithfully alongside the computed size/md5/sha256 apt requires.
+type DebMetadata struct {
+	RepoName     string
+	FilePath     string
+	ContentHash  string
+	Name         string
+	Version      string
+	Architecture string
+	Control      string
+	Size         int64
+	MD5          string
+	SHA256       string
+	// CreatedAt is the persisted insert time; the Release Date: is derived from
+	// the newest value so the index is byte-identical across replicas and
+	// regenerations (issue #117) rather than stamped from wall clock.
+	CreatedAt time.Time
+}
@@ -0,0 +1,30 @@
+package provider
+
+import (
+	"context"
+	"time"
+)
+
+type HelmMetadataStore interface {
+	InsertHelmMetadata(ctx context.Context, meta *HelmMetadata) error
+}
+
+type HelmMetadataDeleter interface {
+	DeleteHelmMetadata(ctx context.Context, repoName, filePath string) error
+}
+
+type HelmMetadataReader interface {
+	ListHelmMetadataEntries(ctx context.Context, repoName string) ([]HelmMetadata, error)
+}
+
+// HelmMetadata is a local chart's Chart.yaml, kept verbatim so index.yaml
+// entries carry every field helm reads.
+type HelmMetadata struct {
+	RepoName    string
+	FilePath    string
+	ContentHash string
+	Name        string
+	Version     string
+	ChartYAML   string
+	CreatedAt   time.Time
+}
@@ -0,0 +1,32 @@
+package provider
+
+import (
+	"context"
+	"time"
+)
+
+type PuppetMetadataStore interface {
+	InsertPuppetMetadata(ctx context.Context, meta *PuppetMetadata) error
+}
+
+type PuppetMetadataDeleter interface {
+	DeletePuppetMetadata(ctx context.Context, repoName, filePath string) error
+}
+
+type PuppetMetadataReader interface {
+	ListPuppetMetadataEntries(ctx context.Context, repoName, slug string) ([]PuppetMetadata, error)
+}
+
+// PuppetMetadata is a local module release; MetadataJSON is the tarball's
+// metadata.json verbatim so Forge release objects carry every field.
+type PuppetMetadata struct {
+	RepoName     string
+	FilePath     string
+	ContentHash  string
+	Slug         string
+	Version      string
+	MetadataJSON string
+	Size         int64
+	MD5          string
+	CreatedAt    time.Time
+}
@@ -0,0 +1,60 @@
+package provider
+
+import (
+	"context"
+	"time"
+)
+
+type RPMMetadataReader interface {
+	ListRPMMetadataEntries(ctx context.Context, repoName string) ([]RPMMetadata, error)
+}
+
+type RPMMetadata struct {
+	RepoName      string
+	FilePath      string
+	ContentHash   string
+	Name          string
+	Epoch         int
+	Version       string
+	Release       string
+	Arch          string
+	Summary       string
+	Description   string
+	RPMSize       int64
+	InstalledSize int64
+	License       string
+	Vendor        string
+	Group         string
+	BuildHost     string
+	SourceRPM     string
+	URL           string
+	Packager      string
+	Requires      []RPMDep
+	Provides      []RPMDep
+	Conflicts     []RPMDep
+	Obsoletes     []RPMDep
+	Files         []RPMFile
+	Changelogs    []RPMChangelog
+	// CreatedAt is the persisted upload timestamp; used as a stable, replica-independent
+	// value for the repodata <time>/<revision> fields so generated indexes are deterministic.
+	CreatedAt time.Time
+}
+
+type RPMDep struct {
+	Name    string `json:"name"`
+	Flags   string `json:"flags,omitempty"`
+	Epoch   string `json:"epoch,omitempty"`
+	Version string `json:"version,omitempty"`
+	Release string `json:"release,omitempty"`
+}
+
+type RPMFile struct {
+	Path string `json:"path"`
+	Type string `json:"type,omitempty"`
+}
+
+type RPMChangelog struct {
+	Author string `json:"author"`
+	Date   int64  `json:"date"`
+	Text   string `json:"text"`
+}
@@ -0,0 +1,15 @@
+package provider
+
+import (
+	"context"
+)
+
+type MetadataStore interface {
+	InsertRPMMetadata(ctx context.Context, meta *RPMMetadata) error
+	InsertDebMetadata(ctx context.Context, meta *DebMetadata) error
+}
+
+type MetadataDeleter interface {
+	DeleteRPMMetadata(ctx context.Context, repoName, filePath string) error
+	DeleteDebMetadata(ctx context.Context, repoName, filePath string) error
+}
@@ -37,6 +37,8 @@ func (p *Provider) UpstreamURL(remote models.Remote, path string) string {
 	return strings.TrimRight(remote.BaseURL, "/") + "/" + strings.TrimLeft(path, "/")
 }

+func (p *Provider) RewritesResponse() {}
+
 func (p *Provider) RewriteResponse(body []byte, remote models.Remote, proxyBaseURL string) ([]byte, error) {
 	if proxyBaseURL == "" || !json.Valid(body) {
 		return nil, nil
@@ -1,5 +1,6 @@
 // Package nuget proxies a NuGet v3 feed: the service index, registration pages
-// and flat container indexes are mutable, .nupkg/.nuspec downloads immutable.
+// and flat container indexes are mutable; package, symbol, nuspec, readme and
+// icon blobs are immutable.
 package nuget

 import (
@@ -7,6 +8,7 @@ import (
 	"encoding/json"
 	"net/http"
 	"net/url"
+	pathpkg "path"
 	"strings"

 	"git.unkin.net/unkin/artifactapi/internal/auth"
@@ -34,10 +36,12 @@ func (p *Provider) Classify(path string) provider.Mutability {

 func (p *Provider) ContentType(path string) string {
 	switch {
-	case strings.HasSuffix(path, ".nupkg"):
+	case strings.HasSuffix(path, ".nupkg"), strings.HasSuffix(path, ".snupkg"), pathpkg.Base(path) == "icon":
 		return "application/octet-stream"
 	case strings.HasSuffix(path, ".nuspec"):
 		return "application/xml"
+	case pathpkg.Base(path) == "readme":
+		return "text/markdown"
 	}
 	return "application/json"
 }
@@ -64,8 +68,8 @@ func (p *Provider) RewriteServed(body []byte, remote models.Remote, proxyBaseURL
 	if from == "" || !json.Valid(body) {
 		return body
 	}
-	to := strings.TrimRight(proxyBaseURL, "/") + "/api/v1/remote/" + remote.Name
-	return []byte(strings.ReplaceAll(string(body), from, to))
+	to, _ := json.Marshal(strings.TrimRight(proxyBaseURL, "/") + "/api/v1/remote/" + remote.Name)
+	return []byte(strings.ReplaceAll(string(body), from, string(to[1:len(to)-1])))
 }

 func (p *Provider) AuthHeaders(_ context.Context, remote models.Remote) (http.Header, error) {
@@ -80,6 +84,14 @@ func origin(baseURL string) string {
 	return u.Scheme + "://" + u.Host
 }

+// isPackage matches flat container blobs: {id}.{ver}.(s)nupkg, {id}.nuspec and
+// the {id}/{ver}/readme and {id}/{ver}/icon files.
 func isPackage(path string) bool {
-	return strings.HasSuffix(path, ".nupkg") || strings.HasSuffix(path, ".nuspec")
+	for _, ext := range []string{".nupkg", ".snupkg", ".nuspec"} {
+		if strings.HasSuffix(path, ext) {
+			return true
+		}
+	}
+	base := pathpkg.Base(path)
+	return (base == "readme" || base == "icon") && strings.Count(strings.Trim(path, "/"), "/") >= 2
 }
@@ -19,7 +19,11 @@ func TestClassify(t *testing.T) {
 	}{
 		{"v3-flatcontainer/newtonsoft.json/13.0.3/newtonsoft.json.13.0.3.nupkg", provider.Immutable},
 		{"v3-flatcontainer/newtonsoft.json/13.0.3/newtonsoft.json.nuspec", provider.Immutable},
+		{"v3-flatcontainer/newtonsoft.json/13.0.3/newtonsoft.json.13.0.3.snupkg", provider.Immutable},
+		{"v3-flatcontainer/newtonsoft.json/13.0.3/readme", provider.Immutable},
+		{"v3-flatcontainer/newtonsoft.json/13.0.3/icon", provider.Immutable},
 		{"v3-flatcontainer/newtonsoft.json/index.json", provider.Mutable},
+		{"readme", provider.Mutable},
 		{"v3/registration5-gz-semver2/newtonsoft.json/index.json", provider.Mutable},
 		{"index.json", provider.Mutable},
 	}
@@ -30,6 +34,28 @@ func TestClassify(t *testing.T) {
 	}
 }

+func TestContentType(t *testing.T) {
+	p := &Provider{}
+	for path, want := range map[string]string{
+		"v3-flatcontainer/a/1.0.0/a.1.0.0.snupkg": "application/octet-stream",
+		"v3-flatcontainer/a/1.0.0/icon":           "application/octet-stream",
+		"v3-flatcontainer/a/1.0.0/readme":         "text/markdown",
+		"v3-flatcontainer/a/index.json":           "application/json",
+	} {
+		if got := p.ContentType(path); got != want {
+			t.Errorf("ContentType(%q) = %q, want %q", path, got, want)
+		}
+	}
+}
+
+func TestRewriteServedEscapesJSON(t *testing.T) {
+	body := `{"@id":"https://api.nuget.org/v3-flatcontainer/"}`
+	got := (&Provider{}).RewriteServed([]byte(body), nugetOrg, `http://evil"<x>`)
+	if want := `{"@id":"http://evil\"\u003cx\u003e/api/v1/remote/nuget-org/v3-flatcontainer/"}`; string(got) != want {
+		t.Errorf("RewriteServed = %s, want %s", got, want)
+	}
+}
+
 func TestUpstreamURL(t *testing.T) {
 	p := &Provider{}
 	tests := []struct {
@@ -3,9 +3,7 @@ package provider
 import (
 	"context"
 	"fmt"
-	"io"
 	"net/http"
-	"time"

 	"git.unkin.net/unkin/artifactapi/pkg/models"
 )
@@ -31,205 +29,6 @@ type FileEntry struct {
 	ContentHash string
 }

-type FileStore interface {
-	ListFilesByPrefix(ctx context.Context, repoName, prefix string) ([]FileEntry, error)
-	ListPackages(ctx context.Context, repoName string) ([]string, error)
-}
-
-type LocalUploader interface {
-	ValidateUpload(filePath string) (storagePath, contentType string, err error)
-	UploadResponse(storagePath, contentHash string, sizeBytes int64) map[string]any
-}
-
-type LocalIndexer interface {
-	ServeLocalIndex(w http.ResponseWriter, r *http.Request, files FileStore, repoName, path string) bool
-	GenerateLocalIndex(ctx context.Context, files FileStore, repoName, path string) ([]byte, error)
-}
-
-type BlobReader interface {
-	Download(ctx context.Context, key string) (io.ReadCloser, int64, error)
-}
-
-type PostUploadHook interface {
-	AfterUpload(ctx context.Context, repoName, storagePath, contentHash string, blobs BlobReader, db MetadataStore)
-}
-
-// PostDeleteHook lets a provider clean up derived state (e.g. RPM metadata that
-// feeds generated repodata) after a local file is removed.
-type PostDeleteHook interface {
-	AfterDelete(ctx context.Context, repoName, storagePath string, db MetadataDeleter) error
-}
-
-type MetadataStore interface {
-	InsertRPMMetadata(ctx context.Context, meta *RPMMetadata) error
-	InsertDebMetadata(ctx context.Context, meta *DebMetadata) error
-}
-
-// RemoteServer lets a remote provider fully answer a request itself instead of
-// going through the byte-proxy engine. It is the remote-side analog of
-// LocalIndexer: a metadata-only remote (e.g. github_rpm) uses it to synthesize
-// repodata from derived per-asset metadata and to redirect package downloads to
-// a backend remote, without ever precaching the packages. Returning false lets
-// the normal proxy path take over.
-type RemoteServer interface {
-	ServeRemote(w http.ResponseWriter, r *http.Request, remote models.Remote, path, proxyBaseURL string, store RemoteMetadataStore) bool
-}
-
-// ServeRewriter rewrites a cached mutable remote response per request, for
-// indexes whose URLs must name the requesting host and so cannot be rewritten
-// before caching.
-type ServeRewriter interface {
-	RewriteServed(body []byte, remote models.Remote, proxyBaseURL string) []byte
-}
-
-// RemoteMetadataStore is the persistence surface a RemoteServer needs to cache
-// and read the metadata it derives per upstream asset. *database.DB satisfies it.
-type RemoteMetadataStore interface {
-	RPMMetadataReader
-	MetadataStore
-	MetadataDeleter
-}
-
-type MetadataDeleter interface {
-	DeleteRPMMetadata(ctx context.Context, repoName, filePath string) error
-	DeleteDebMetadata(ctx context.Context, repoName, filePath string) error
-}
-
-type RPMMetadataReader interface {
-	ListRPMMetadataEntries(ctx context.Context, repoName string) ([]RPMMetadata, error)
-}
-
-// DebMetadataReader is the read surface the deb LocalIndexer needs to
-// regenerate a flat apt repository (Packages/Release) from stored rows.
-// *database.DB satisfies it; ServeLocalIndex type-asserts the FileStore to it,
-// mirroring how the rpm provider reaches its RPMMetadataReader.
-type DebMetadataReader interface {
-	ListDebMetadataEntries(ctx context.Context, repoName string) ([]DebMetadata, error)
-}
-
-// DebMetadata is the derived per-package metadata for a Debian .deb, carrying
-// the full raw control stanza so the Packages index can be regenerated
-// faithfully alongside the computed size/md5/sha256 apt requires.
-type DebMetadata struct {
-	RepoName     string
-	FilePath     string
-	ContentHash  string
-	Name         string
-	Version      string
-	Architecture string
-	Control      string
-	Size         int64
-	MD5          string
-	SHA256       string
-	// CreatedAt is the persisted insert time; the Release Date: is derived from
-	// the newest value so the index is byte-identical across replicas and
-	// regenerations (issue #117) rather than stamped from wall clock.
-	CreatedAt time.Time
-}
-
-// AlpineMetadataStore / AlpineMetadataDeleter / AlpineMetadataReader are the
-// Alpine-specific persistence surfaces. They are kept separate from the shared
-// RPM/Deb metadata interfaces so the apk provider can type-assert the generic
-// MetadataStore/MetadataDeleter/FileStore it is handed without widening (and
-// thus perturbing the test doubles of) the rpm and deb providers. *database.DB
-// satisfies all three.
-type AlpineMetadataStore interface {
-	InsertAlpineMetadata(ctx context.Context, meta *AlpineMetadata) error
-}
-
-type AlpineMetadataDeleter interface {
-	DeleteAlpineMetadata(ctx context.Context, repoName, filePath string) error
-}
-
-type AlpineMetadataReader interface {
-	ListAlpineMetadataEntries(ctx context.Context, repoName string) ([]AlpineMetadata, error)
-}
-
-// AlpineMetadata is the derived per-package metadata for an Alpine .apk, holding
-// the fields an APKINDEX record carries plus the apk pull checksum (Q1…, the
-// sha1 of the control gzip stream) and the download/installed sizes.
-type AlpineMetadata struct {
-	RepoName         string
-	FilePath         string
-	ContentHash      string
-	Checksum         string   // C: "Q1" + base64(sha1(control gzip stream))
-	Name             string   // P:
-	Version          string   // V:
-	Arch             string   // A:
-	DownloadSize     int64    // S: on-disk .apk size
-	InstalledSize    int64    // I: unpacked size from .PKGINFO
-	Description      string   // T:
-	URL              string   // U:
-	License          string   // L:
-	Origin           string   // o:
-	Maintainer       string   // m:
-	BuildTime        int64    // t:
-	Commit           string   // c:
-	ProviderPriority string   // k:
-	Depends          []string // D:
-	Provides         []string // p:
-	InstallIf        []string // i:
-}
-
-type RPMMetadata struct {
-	RepoName      string
-	FilePath      string
-	ContentHash   string
-	Name          string
-	Epoch         int
-	Version       string
-	Release       string
-	Arch          string
-	Summary       string
-	Description   string
-	RPMSize       int64
-	InstalledSize int64
-	License       string
-	Vendor        string
-	Group         string
-	BuildHost     string
-	SourceRPM     string
-	URL           string
-	Packager      string
-	Requires      []RPMDep
-	Provides      []RPMDep
-	Conflicts     []RPMDep
-	Obsoletes     []RPMDep
-	Files         []RPMFile
-	Changelogs    []RPMChangelog
-	// CreatedAt is the persisted upload timestamp; used as a stable, replica-independent
-	// value for the repodata <time>/<revision> fields so generated indexes are deterministic.
-	CreatedAt time.Time
-}
-
-type RPMDep struct {
-	Name    string `json:"name"`
-	Flags   string `json:"flags,omitempty"`
-	Epoch   string `json:"epoch,omitempty"`
-	Version string `json:"version,omitempty"`
-	Release string `json:"release,omitempty"`
-}
-
-type RPMFile struct {
-	Path string `json:"path"`
-	Type string `json:"type,omitempty"`
-}
-
-type RPMChangelog struct {
-	Author string `json:"author"`
-	Date   int64  `json:"date"`
-	Text   string `json:"text"`
-}
-
-type IndexMerger interface {
-	MergeIndexes(members []MemberIndex, proxyBaseURL string) ([]byte, error)
-}
-
-type MemberIndex struct {
-	RemoteName string
-	Body       []byte
-}
-
 var registry = map[models.PackageType]Provider{}

 func Register(p Provider) {
@@ -247,53 +46,3 @@ func Get(t models.PackageType) (Provider, error) {
 func All() map[models.PackageType]Provider {
 	return registry
 }
-
-type HelmMetadataStore interface {
-	InsertHelmMetadata(ctx context.Context, meta *HelmMetadata) error
-}
-
-type HelmMetadataDeleter interface {
-	DeleteHelmMetadata(ctx context.Context, repoName, filePath string) error
-}
-
-type HelmMetadataReader interface {
-	ListHelmMetadataEntries(ctx context.Context, repoName string) ([]HelmMetadata, error)
-}
-
-// HelmMetadata is a local chart's Chart.yaml, kept verbatim so index.yaml
-// entries carry every field helm reads.
-type HelmMetadata struct {
-	RepoName    string
-	FilePath    string
-	ContentHash string
-	Name        string
-	Version     string
-	ChartYAML   string
-	CreatedAt   time.Time
-}
-
-type PuppetMetadataStore interface {
-	InsertPuppetMetadata(ctx context.Context, meta *PuppetMetadata) error
-}
-
-type PuppetMetadataDeleter interface {
-	DeletePuppetMetadata(ctx context.Context, repoName, filePath string) error
-}
-
-type PuppetMetadataReader interface {
-	ListPuppetMetadataEntries(ctx context.Context, repoName, slug string) ([]PuppetMetadata, error)
-}
-
-// PuppetMetadata is a local module release; MetadataJSON is the tarball's
-// metadata.json verbatim so Forge release objects carry every field.
-type PuppetMetadata struct {
-	RepoName     string
-	FilePath     string
-	ContentHash  string
-	Slug         string
-	Version      string
-	MetadataJSON string
-	Size         int64
-	MD5          string
-	CreatedAt    time.Time
-}
@@ -39,6 +39,8 @@ func (p *Provider) UpstreamURL(remote models.Remote, path string) string {
 	return strings.TrimRight(remote.BaseURL, "/") + "/" + strings.TrimLeft(path, "/")
 }

+func (p *Provider) RewritesResponse() {}
+
 func (p *Provider) RewriteResponse(body []byte, remote models.Remote, proxyBaseURL string) ([]byte, error) {
 	if proxyBaseURL == "" {
 		return nil, nil
@@ -52,6 +52,8 @@ func (p *Provider) UpstreamURL(remote models.Remote, path string) string {
 	return strings.TrimRight(remote.BaseURL, "/") + "/" + strings.TrimLeft(path, "/")
 }

+func (p *Provider) RewritesResponse() {}
+
 func (p *Provider) RewriteResponse(body []byte, remote models.Remote, proxyBaseURL string) ([]byte, error) {
 	if proxyBaseURL == "" {
 		return nil, nil
@@ -73,6 +73,8 @@ func (p *Provider) UpstreamURL(remote models.Remote, path string) string {
 	return strings.TrimRight(remote.BaseURL, "/") + "/v1/providers/" + strings.TrimLeft(path, "/")
 }

+func (p *Provider) RewritesResponse() {}
+
 func (p *Provider) RewriteResponse(body []byte, remote models.Remote, proxyBaseURL string) ([]byte, error) {
 	if remote.ReleasesRemote == "" {
 		return nil, nil
--- a/Show More
+++ b/Show More
Author	SHA1	Message	Date
unkin-agent	2c671a1be7	Add npm virtual repositories (#152 ) ci/woodpecker/tag/docker Pipeline was successful Details npm clients need one registry for internal and public packages, but npm virtuals have no merger, so a local npm repo and npmjs cannot be served together. - merge packuments across members: union versions (member order wins), first-defined dist-tags minus ones naming missing versions - serve full and abbreviated (`install-v1`) packuments, scoped names included, via the shared virtual merge cache - point tarballs at `<virtual>/<member>/<path>` and 302 them to the owning member - reject an invalid Host with 400; serve virtual and local packuments `no-store` with `Vary: Host, X-Forwarded-Proto, Accept` - return 502 when a member errors; skip members without the package - add `docs/npm/virtual.md` and a stock `node:22` e2e-docker case Reviewed-on: #152 Co-authored-by: unkin-agent <unkin-agent@unkin.net> Co-committed-by: unkin-agent <unkin-agent@unkin.net>	2026-10-17 21:52:55 +11:00
unkin-agent	bf8228e9f2	Harden nuget per-request index rewrite (#149 ) NuGet index rewrites embed the client Host and X-Forwarded-Proto into per-request JSON, so a shared cache could store one host's URLs, a forged header could inject content, HEAD skipped the rewrite, and immutable flat-container blobs were re-validated as mutable. - set `no-store`, `Vary: Host, X-Forwarded-Proto` and the rewritten Content-Length on GET and HEAD - share `v1.Scheme()` (first X-Forwarded-Proto value, `http`/`https` only) across nuget, cargo, virtual redirects and the terraform registry - reject non host[:port] Host headers and JSON-escape the injected base URL - classify `.snupkg`, `readme` and `icon` flat-container blobs as immutable requires unkin/artifactapi#148 merged first Reviewed-on: #149 Co-authored-by: unkin-agent <unkin-agent@unkin.net> Co-committed-by: unkin-agent <unkin-agent@unkin.net>	2026-10-17 21:39:42 +11:00
unkin-agent	ad72c3d630	Add HEAD route for remotes and gems remote type (#148 ) Remotes answer only GET, so HEAD probes get 405, and there is no RubyGems remote. Synthesized (github_) and rewritten (npm, helm, pypi, puppet, terraform, nuget) mutable responses need the GET path for HEAD lengths to match. - add a HEAD route for /api/v1/remote/{name}/; rewritten and synthesized mutable HEADs go through the GET path - mark rewriting providers with a ResponseRewriter capability; set Content-Length on proxied GETs - use the remote's upstream client for upstream HEAD - add a gems remote type: compact index and specs mutable, .gem and gemspec immutable - add gems docs, UI snippet and e2e caching case; add HEAD-vs-GET e2e checks Reviewed-on: #148 Co-authored-by: unkin-agent <unkin-agent@unkin.net> Co-committed-by: unkin-agent <unkin-agent@unkin.net>	2026-10-17 21:23:29 +11:00
unkin-agent	42998a80f4	Add local npm registry publish and install (#139 ) In-house npm packages have nowhere to publish: npm locals only accept raw file uploads, which `npm publish` and `npm install` cannot use. - accept `npm publish` (`PUT /api/v1/local/<repo>/<pkg>`, scoped included) on local npm repos, capping the body at 256MiB and verifying the tarball against dist integrity - write tarball, then manifest, then dist-tags; publishing an existing version returns 409 and leaves tags alone, while a matching leftover tarball from a failed publish is reused so retries complete - serve packuments with tarball URLs pointing back at the local repo, dropping dist-tags that name unpublished versions - document the `.npmrc` registry + `_authToken` setup in the UI and `docs/npm/local.md` - add a stock `node:22` publish/install e2e-docker case Reviewed-on: #139 Co-authored-by: unkin-agent <unkin-agent@unkin.net> Co-committed-by: unkin-agent <unkin-agent@unkin.net>	2026-10-17 17:18:05 +11:00
unkin-agent	d69a58e54a	Fix mutable path refresh: default mutable_ttl, revalidate after expiry (#151 ) A remote created through the API without `mutable_ttl` was stored as 0, so its mutable paths never refreshed. With `check_mutable`, the ETag expired together with the freshness key, so an expired mutable path was always refetched in full and never revalidated. - default an omitted `mutable_ttl` to 3600 on create; keep the stored value on update; explicit 0 still means never expire - keep mutable ETags for `max(10 × mutable_ttl, 7d)`, re-armed on each 304, only when `check_mutable` is set and `mutable_ttl` is non-zero - drop the stored ETag when a refetch returns none; never store one for immutable paths - flush a remote's Redis keys on delete - update docs/common/caching.md Reviewed-on: #151 Co-authored-by: unkin-agent <unkin-agent@unkin.net> Co-committed-by: unkin-agent <unkin-agent@unkin.net>	2026-10-17 17:17:41 +11:00
unkin-agent	6e6481e09a	Split per-type registration points into per-type files (#150 ) Parallel package-type PRs conflict because every type edits the same shared lines (README tables, provider.go, server.go imports, the package type list, UI maps, the e2e caching table). A new type now only adds files. - move per-type README sections into docs/<type>/{remote,local,virtual}.md - split provider.go capability and metadata types into cap_.go and metadata_.go - import providers via internal/provider/all, one file per type - declare each package type in package_type_<type>.go, registered via init() - split the e2e caching cases and UI colours/snippets into per-type files Reviewed-on: #150 Co-authored-by: unkin-agent <unkin-agent@unkin.net> Co-committed-by: unkin-agent <unkin-agent@unkin.net>	2026-10-17 15:54:47 +11:00