docs: lead decision 92, WebUI preview path; sessions line (#1568)

Co-Authored-By: Claude Opus 5.5 <[email protected]>
2026-10-11 16:55:22 -05:00
parent 1786136002
commit a3ba78b525
2 changed files with 25 additions and 0 deletions
@@ -1942,3 +1942,27 @@ which stay with him. Each item names who decided it and what happened.
      No version floor check runs at start; one would catch a
      downgrade, not a bad upgrade. Re-pinning to a versioned path is one
      `install --claude` away if a release breaks it.
+92. **A WebUI preview path ahead of slice 1 acceptance (2026-10-11).**
+    Jason asked Mos for an ETA for the refactor WebUI at mosaic.woltje.com.
+    No deploy path exists: decision 48 put access beyond localhost and
+    Pocket ID out of v1, and the Console binds and answers on loopback
+    only. Sage proposed cuts and Mos approved them: sign-in at the edge
+    through Pocket ID forward auth, a systemd user unit beside the bus, a
+    preview hostname so the old build stays, read-only, and not held for
+    S7 or Gate E.
+    - Row 74 (#1568, brief `docs/plans/2026-10-11_webui-preview.md`),
+      owner Dewey after their row 59 review, reviewers Darkwing and
+      Filbert. It goes ahead of Sage's rows 64-66.
+    - It is blocked on Q91, Jason's answer on amending 48. No code until
+      then, and nothing is served until he approves ops-01's route from
+      the edge to this host. ops-01 starts the Pocket ID client, the DNS
+      name, the cert and the route proposal now.
+    - Dates given to Mos: preview likely 10-14, worst 10-20, committed
+      10-16, each assuming the ruling on 10-12. Cutover likely 10-23,
+      worst 10-30, not committed until the first S7 piece closes.
+    - The tradeoff: the identity check trusts a header from whatever
+      reaches the loopback port. That is safe only if the proxy strips the
+      header from clients and is the only path in from off the host. A
+      local process can still forge it, the same exposure the Console's
+      reads have today. In-app OIDC would close that and costs several
+      rows; it waits for cutover or later.