railiance-platform/workplans/RPF-WP-0047-bao-coulomb-social-notice-page.md
codex c3486be3c2
Some checks failed
CI Smoke / host-smoke (push) Successful in 0s
CI Smoke / container-smoke (push) Has been cancelled
RPF-WP-0047 done: bao.coulomb.social serves a trusted static notice
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

Assistant: claude-code
Assistant-Model: opus
Assistant-Process: 150322@bnt-lap001
Assistant-Session: 16a7b788-374e-4915-a1df-fc87ffd9a5e4
2026-09-24 00:57:34 +02:00

109 lines
4.3 KiB
Markdown

---
id: RPF-WP-0047
type: workplan
title: "bao.coulomb.social: static not-publicly-available page with a trusted certificate"
domain: financials
repo: railiance-platform
status: finished
owner: railiance-platform
topic_slug: railiance
created: "2026-09-24"
updated: "2026-09-24"
related: [RPF-WP-0025, RMASTER-WP-0020]
state_hub_workstream_id: "6ab44339-66d3-522b-9d34-fc9ff6f20d1b"
---
## Why
The founder decided on 2026-09-23 (relayed by user-engine, hub message
`7d8982c2`) that OpenBao stays non-public. The RMASTER-WP-0020-T09 /
RPF-WP-0025-T03 retraction stands. But the published name must not present a
self-signed Traefik certificate and a bare 404. **Requested end state:** a
static page saying the resource is not publicly available, over a trusted
certificate. It names no internal host, port, tunnel or version, offers no
login form or redirect, and proxies nothing to OpenBao. This reverses the DNS
withdrawal handed to railiance-infra on 2026-09-22; a hold was sent
(`a4bca724`). A static notice is not an exposure grant under ADR-0008.
## T01 Build the page and manifests
```task
id: RPF-WP-0047-T01
status: done
priority: high
state_hub_task_id: "caa019c6-dfdb-5f17-b860-67e1feaebd51"
```
`argocd/platform-addons/bao-notice/` holds:
- A namespace `bao-notice` with a NetworkPolicy: ingress only from Traefik
(kube-system) on 8080, no egress at all, so there is no path to OpenBao.
- Nginx `nginxinc/nginx-unprivileged` pinned by digest (already running in
the cluster), running non-root with a read-only root filesystem and all
capabilities dropped.
- `/` returns 200 with the notice. Every other path returns 404 with the same
notice, so a probe of `/v1/sys/health` is never a false 200.
- `server_tokens off`, CSP `default-src 'none'`, `noindex`, `no-store`.
- An Ingress on `websecure` with `letsencrypt-prod`, and an HTTP→HTTPS
redirect at priority 1, so ACME HTTP-01 still wins.
Verified on 2026-09-24:
- Local container probe: `/` 200, `/v1/sys/health` and `/ui/` 404 with the
notice, `Server: nginx` with no version.
- Server-side dry run on railiance01: all eight objects valid.
- No existing Ingress claims the host.
- The `railiance-tenants` AppProject already whitelists every kind.
## T02 Adopt through ArgoCD (live, founder go-ahead)
```task
id: RPF-WP-0047-T02
status: done
priority: high
state_hub_task_id: "3af7f95f-4689-547e-90dc-529e65c7b9e2"
```
The Application is `argocd/railiance01/applications/bao-notice.application.yaml`,
pinned to `960ed09`. Sync the root by hand to create it, then sync
`bao-notice` by hand. cert-manager issues `bao-notice-tls` over HTTP-01.
## T03 Verify from outside and close the loop
```task
id: RPF-WP-0047-T03
status: done
priority: high
state_hub_task_id: "26bf3291-e74e-51dd-89e5-c38169d471b0"
```
From the workstation: `curl -sv https://bao.coulomb.social/` must show a
trusted chain with no `-k`, HTTP 200 and the notice.
`https://bao.coulomb.social/v1/sys/health` must return 404, and
`http://bao.coulomb.social/` must redirect to https. Record the evidence, then
tell user-engine, railiance-infra (keep the A record) and railiance-master
(T09 posture unchanged).
## Rollback
Delete the `bao-notice` Application with prune. That restores today's state:
the Traefik default certificate and a 404.
## Execution record — 2026-09-23/24 (founder go-ahead)
- **T02 done.** The root and `bao-notice` were hand-synced; the app is
Synced/Healthy.
- **First issuance returned 502 on the HTTP-01 challenge.** The isolation
NetworkPolicy also covered cert-manager's solver pod (port 8089). Fixed by a
narrow `bao-notice-acme-solver` policy, Traefik to solver pods on 8089 only,
and `bao-notice-tls` was issued.
- **HSTS added** (`max-age=31536000`, this host only). The single-replica pod
swap caused a few seconds of 502 through Traefik.
- **T03 done.** Checked from the workstation with no `-k`:
- Certificate `CN=bao.coulomb.social` from Let's Encrypt, verify ok, expires
2026-12-22 (cert-manager renews).
- `/` returns 200 with the notice; `/v1/sys/health` and `/ui/` return 404;
`http://` returns a 301 to https.
- HSTS and CSP `default-src 'none'` are set, and the `Server` header carries
no version.
- No internal identifiers, form or redirect in the body.
- Evidence: `docs/evidence/2026-09-24-bao-coulomb-social-notice.json`.