Merge remote-tracking branch 'origin/dev' into fix/registry-namespace-admin-emi

.gitignore

@@ -2,3 +2,9 @@
 # Claude Code worktrees. These are other sessions' checkouts living inside this
 # repo; `git add -A` will otherwise commit them as embedded git repositories.
 .claude/worktrees/
+
+# Per-operator Claude Code settings — local permission grants, not team config.
+# Untracked but not ignored meant one `git add -A` would publish one operator's
+# allowlist (including which secrets they may decrypt) to everyone.
+# `.claude/settings.json` IS tracked and stays the shared, reviewed config.
+.claude/settings.local.json

infra/ACCESS.md

@@ -26,7 +26,7 @@ rather than which environment happens to live there:
 
 | Host | Public IP | Mesh IP | Role |
 |------|-----------|---------|------|
-| **vps1** | `75.119.132.91` | `10.10.0.2` | "Operational programs": Forgejo (git + CI + registry, `git.thermograph.org`), Grafana/Loki/Alloy (`dashboard.thermograph.org`), the `emigriffith.dev` portfolio, and the **dev** environment (mesh-only, `10.10.0.2:8137`, no public DNS/Caddy/TLS) |
+| **vps1** | `75.119.132.91` | `10.10.0.2` | "Operational programs": Forgejo (git + CI + registry, `git.thermograph.org`, also served at `dev.jinemi.com` — migration in progress, `.org` still canonical), Grafana/Loki/Alloy (`dashboard.thermograph.org`), the `emigriffith.dev` portfolio, and the **dev** environment (mesh-only, `10.10.0.2:8137`, no public DNS/Caddy/TLS) |
 | **vps2** | `169.58.46.181` | `10.10.0.1` | "The deployed environment": **prod** and **beta**, as two separate Docker Swarm stacks, plus Centralis, Postfix, the backups, and the one shared TimescaleDB instance both stacks use |
 | **desktop** | — | `10.10.0.3` | AI-model hosting (voice-to-text, an upcoming-feature LLM) plus flex Swarm-worker capacity. Hosts **no** Thermograph environment — `make dev-up` still works there, but only as a laptop-local rehearsal |
 

infra/CLAUDE.md

@@ -12,7 +12,9 @@ Two VPS boxes plus the operator's desktop, on one WireGuard mesh, named by
 ROLE rather than by environment:
 
 - **vps1** — `75.119.132.91`, mesh `10.10.0.2`. "Operational programs": Forgejo
-  (git + CI + registry, `git.thermograph.org`), Grafana/Loki/Alloy
+  (git + CI + registry, `git.thermograph.org`, also served at `dev.jinemi.com`
+  — one Caddy site block, the `.org` name still canonical; migration in
+  progress, see `deploy/forgejo/README.md`), Grafana/Loki/Alloy
   (`dashboard.thermograph.org`), the `emigriffith.dev` portfolio site, and the
   **dev** environment (`/opt/thermograph-dev`, tracks `dev`), with its own
   Postgres container. Dev is mesh-only here — published on `10.10.0.2:8137`,

infra/deploy/Caddyfile.vps1

@@ -3,6 +3,9 @@
 # vps1 serves:
 #   emigriffith.dev            -> static portfolio site (from disk)
 #   git.thermograph.org        -> Forgejo         (see deploy/forgejo/caddy-git.conf)
+#   dev.jinemi.com             -> Forgejo, same site block — migration target for
+#                                 git.thermograph.org, both served, the .org one
+#                                 still canonical (that file's header explains)
 #   dashboard.thermograph.org  -> Grafana         (see observability/caddy-grafana.conf)
 #
 # and hosts the **dev** environment at dev.thermograph.org (below).

infra/deploy/forgejo/README.md

@@ -83,6 +83,53 @@ pattern as its other site blocks, same automatic-HTTPS.
    reach the registry over the mesh, not the public internet — see "Registry
    access from mesh clients" below.
 
+## Migrating the domain (git.thermograph.org -> dev.jinemi.com)
+
+In progress. **Both hostnames are served**; `git.thermograph.org` is still the
+canonical one.
+
+`caddy-git.conf` lists both names on a *single* site block. That is deliberate,
+not cosmetic: the `/v2/*` mesh-only matcher is per-block, so a second block for
+`dev.jinemi.com` would re-expose the registry API publicly and undo hazard #15.
+Whatever else changes, keep the two names in one block.
+
+Serving a second hostname is the cheap half. Forgejo has exactly one
+`ROOT_URL`, and it builds every absolute URL from it, so on the non-canonical
+host today:
+
+- clone URLs on repo pages read `git.thermograph.org`,
+- the OAuth callback and post-login redirect land on `git.thermograph.org`
+  (login is Google-SSO-only — see `docker-stack.yml`),
+- webhook payload URLs and notification mail links use `git.thermograph.org`.
+
+Browsing, git-over-HTTP and the API all work on `dev.jinemi.com` regardless.
+
+Flipping canonical (`FORGEJO_DOMAIN=dev.jinemi.com`, then `docker stack deploy`)
+needs these **first**, or it breaks login and CI rather than just renaming
+things:
+
+1. **Google OAuth** — add `https://dev.jinemi.com/user/oauth2/<name>/callback`
+   as an authorized redirect URI in the Google Cloud console *before* the flip.
+   Not in this repo; nothing here can verify it. Miss it and no one can log in,
+   including to undo the flip.
+2. **The registry host in image names** — images are named by registry host, and
+   the `git.thermograph.org/` prefix is baked into the runner labels
+   (`register-lan-runner.sh`, `runner-vps2/README.md`), the CI-runner image
+   (`ci-runner/Dockerfile`), `REGISTRY_HOST` in `infra/.env.example`, and every
+   already-pushed tag. Renaming the registry host is a separate migration from
+   renaming the web UI; the two need not happen together, and doing the web UI
+   alone is much the smaller change.
+3. **Mesh `/etc/hosts` pins** — every node in "Registry access from mesh
+   clients" below pins `git.thermograph.org` to `10.10.0.2`. A node pulling
+   under a new hostname needs its own pin, or Caddy's `/v2/*` matcher sees a
+   public source IP and returns 403.
+4. **Runner instance URL** — registered runners hold the instance URL they were
+   registered with. Existing runners keep working against the old name as long
+   as it resolves; a *re*-registration must use whichever name is canonical then.
+
+Until all four are handled, leaving `ROOT_URL` on `git.thermograph.org` is the
+correct state, not an unfinished one.
+
 ## Registry access from mesh clients
 
 Any node that needs `docker login`/push/pull against the registry (the CI

infra/deploy/forgejo/caddy-git.conf

@@ -1,6 +1,6 @@
-# Only the block below (from "git.thermograph.org {") goes into the live
-# Caddyfile — append just that part, not this instructional header, or it
-# ends up as a confusing orphaned comment in production config with no
+# Only the block below (from "git.thermograph.org, dev.jinemi.com {") goes into
+# the live Caddyfile — append just that part, not this instructional header, or
+# it ends up as a confusing orphaned comment in production config with no
 # surrounding context (docker-stack.yml isn't visible from there).
 #
 # Target: /etc/caddy/Caddyfile on vps1 (the box `forgejo` is pinned to — see
@@ -9,19 +9,30 @@
 # Forgejo publishes its web UI to 127.0.0.1:3080 only (host-local), and this
 # block is the only thing that ever talks to it.
 #
-# Registry exposure (hazard #15, see docker-stack.yml's header comment):
-# git.thermograph.org/v2/* is the built-in OCI registry API. It's blocked
-# from anywhere except the WireGuard mesh (10.10.0.0/24) — reachable to CI
-# runners and Swarm nodes over wg0, not from the public internet. Everything
-# else (web UI, git-over-HTTP, PR pages) stays public like the rest of the
-# site.
+# TWO hostnames, ONE block, and that is load-bearing — see "Registry exposure"
+# below. Both names must share a site block so the /v2/* restriction covers
+# both. Splitting dev.jinemi.com into its own block serves the same Forgejo
+# with the registry API open to the public internet.
 #
-# Update the domain below to match whatever you actually pointed at vps1's
-# IP, if not git.thermograph.org.
+# dev.jinemi.com is the migration target for git.thermograph.org; both are
+# served during the transition. Each gets its own Let's Encrypt cert
+# automatically (HTTP-01) — both A records already point at vps1.
+#
+# git.thermograph.org stays CANONICAL until Forgejo's ROOT_URL is flipped
+# (docker-stack.yml, FORGEJO_DOMAIN). Until then dev.jinemi.com serves every
+# page, but Forgejo builds absolute URLs from ROOT_URL, so clone URLs, the
+# OAuth callback and post-login redirects still name git.thermograph.org.
+# Flipping it is NOT just this file — see README.md, "Migrating the domain".
+#
+# Registry exposure (hazard #15, see docker-stack.yml's header comment):
+# /v2/* is the built-in OCI registry API. It's blocked from anywhere except
+# the WireGuard mesh (10.10.0.0/24) — reachable to CI runners and Swarm nodes
+# over wg0, not from the public internet. Everything else (web UI,
+# git-over-HTTP, PR pages) stays public like the rest of the site.
 
 # --- copy from here down into /etc/caddy/Caddyfile ---
 
-git.thermograph.org {
+git.thermograph.org, dev.jinemi.com {
 	encode zstd gzip
 
 	@registry_external {

infra/deploy/stack/lb/Caddyfile

@@ -13,6 +13,34 @@
 {
 	auto_https off
 	admin off
+
+	# WITHOUT THIS, EVERY OAUTH REDIRECT URI THE BACKEND BUILDS IS http://.
+	#
+	# There are two Caddy hops in front of the app: the HOST Caddy terminates TLS
+	# and sets X-Forwarded-Proto: https, then proxies to 127.0.0.1:8137 over plain
+	# HTTP; this LB is the second hop. Caddy only preserves an incoming
+	# X-Forwarded-* header when the immediate peer is a TRUSTED proxy — otherwise
+	# it overwrites the header with the scheme of the connection it just received,
+	# which here is http. So the app saw http:// no matter how the user arrived.
+	#
+	# accounts/oauth.py:_redirect_uri builds the callback from x-forwarded-proto,
+	# and that URI must match what is registered in the provider's console.
+	# Discord tolerates the http:// form; GOOGLE REJECTS a non-HTTPS redirect URI
+	# for a web client outright, so Google sign-in could not work at all until
+	# this was fixed — independently of whether the credentials were configured.
+	#
+	# Measured on vps2 before the fix, same request to each hop:
+	#   direct to web, XFP=https   -> https://thermograph.org/api/v2/...
+	#   through this LB, XFP=https -> http://thermograph.org/api/v2/...
+	#
+	# private_ranges covers 127.0.0.1/8 and the Docker bridge ranges, which is
+	# where the host Caddy reaches this container from. Nothing outside the box
+	# can connect here — the LB is bound to loopback precisely so it cannot be
+	# reached un-fronted (hazard #6, above) — so trusting the local peer does not
+	# widen anything: the only party that can set these headers is the host Caddy.
+	servers {
+		trusted_proxies static private_ranges
+	}
 }
 
 :8137 {

infra/deploy/stack/lb/Caddyfile.beta

@@ -17,6 +17,17 @@
 {
 	auto_https off
 	admin off
+
+	# Same reason as ./Caddyfile — see the long note there. Without it this second
+	# Caddy hop overwrites the host Caddy's X-Forwarded-Proto: https with http,
+	# and every OAuth redirect URI the backend builds comes out non-HTTPS, which
+	# Google rejects outright for a web client.
+	#
+	# Beta needs it as much as prod does, and arguably first: beta is where an
+	# OAuth provider change gets tested before it reaches thermograph.org.
+	servers {
+		trusted_proxies static private_ranges
+	}
 }
 
 :8137 {