docs: add architecture reference, update CLAUDE.md with current CI/sidebar state

CLAUDE.md

@@ -3,50 +3,66 @@
 ## Architecture
 - Synology DS224+ (x86_64, 18GB RAM), Tailscale IP: `100.78.131.124`, LAN: `192.168.31.221`
 - Docker: `/volume1/@appstore/ContainerManager/usr/bin/docker`, compose dirs: `/volume1/docker/*`
-- Caddy VPS (`161.33.182.109`, Tailscale: `100.109.198.126`): reverse proxy for `nas.jimmygan.com` → NAS:4000, `music.jimmygan.com` → NAS:4533
-- Build VPS (server2, `158.101.140.85`, Tailscale: `100.70.115.1`): podman, fast network for pulling images
+- VPS Caddy (`161.33.182.109`, Tailscale: `100.109.198.126`): reverse proxy for `*.jimmygan.com`
+- Build VPS (server2, `158.101.140.85`, Tailscale: `100.70.115.1`): podman, fast network
+- NAS Caddy: port 8443 fallback only (SNI broken)
 
 ## Dashboard
 - Backend: Python FastAPI (`dashboard/backend/`)
 - Frontend: Svelte 5 + Tailwind CSS (`dashboard/frontend/`)
 - Svelte 5 runes: use `$state`, `$props`, `$bindable` (not legacy `export let`)
-- Sidebar: categorized as Main (Overview, Docker, Files, Terminal), Media (Navidrome, Jellyfin, Audiobookshelf, Immich), Tools (OpenClaw, Repos, Gitea Web, Stirling PDF, Vaultwarden, Speedtest)
-- New pages: create route in `src/routes/`, import in `App.svelte`, add to appropriate category in `Sidebar.svelte`
+- Sidebar categories: Main, Media, Tools (see `docs/architecture.md` for full layout)
+- New pages: create route in `src/routes/`, import in `App.svelte`, update `Sidebar.svelte`
+
+## CI/CD
+Two workflows in `.gitea/workflows/`:
+- **deploy.yml** — main branch: tests → build → deploy to production
+- **deploy-dev.yml** — dev branch: tests → build → deploy to dev
+
+Two runners:
+- **VPS runner** (`vps` label on server2) — build + test jobs (podman)
+- **NAS runner** (`nas` label on DS224+) — deploy jobs (docker, host networking)
+
+### Build flow
+1. VPS builds Docker image + pushes to local registry `100.78.131.124:5501/`
+2. Streams image to NAS: `podman save | gzip | ssh nasts "gunzip | docker load"`
+3. NAS deploys: `docker compose up -d`
+
+## Local Registry
+- Container: `registry-mirror` on NAS, port 5501
+- Standalone mode (NOT Docker Hub proxy — blocked from China)
+- VPS pushes with `podman push --tls-verify=false`
 
 ## Authentication
-- **Gitea API Access:** Use a dedicated `claude-ci-bot` user with a Personal Access Token (PAT).
-- **Secret Management:** Store the PAT in an environment variable `GITEA_TOKEN`. Avoid hardcoding paths to secret files.
-- **Project Secrets:** Keep sensitive configurations (like the PAT) in a local `.env` file that is ignored by git.
-
-- Build images on VPS (server2) to avoid NAS HDD I/O saturation:
-  1. Build on VPS with Podman
-  2. Stream image to NAS: `podman save <image> | ssh nasts "/volume1/@appstore/ContainerManager/usr/bin/docker load"`
-  3. Deploy: `git push` triggers Gitea Actions CI (builds on VPS, transfers, then `docker compose up`) — do NOT manually restart
-- CI workflows are in `.gitea/workflows/` in this repo
+- **Gitea API Access:** Token stored in `~/.config/tea/config.yml` (user: `jimmy`)
+- **Gitea Token:** `aa8c20329040c56c079d2676318c6e33f14b221d`
+- **Project Secrets:** Gitea repo secrets: CI_TOKEN, CI_DB_PASSWORD, SECRET_KEY, ADMIN_PASSWORD_HASH, TRANSMISSION_USER, TRANSMISSION_PASS
 
 ## claude-dev CI Contract
-- Source package lives in `claude-dev/` (`Dockerfile`, `required-tools.txt`, and smoke scripts under `claude-dev/scripts/`)
-- CI-first only: update `claude-dev/*` in git, then push; do not do ad-hoc local image rebuild/redeploy for normal changes
-- Required capabilities are declared in `claude-dev/required-tools.txt`; adding tools must update this file and related smoke checks
-- Build/deploy gate (`.gitea/workflows/deploy-claude-dev-dev.yml`) uses NAS runner with `DOCKER_BUILDKIT=0`, then runs:
-  - tools smoke (`smoke-tools.sh`)
-  - network smoke (`smoke-network.sh`)
-  - auth/mount smoke (`smoke-auth.sh` with required runtime mounts)
-- Deployment only happens after smoke pass, using immutable tag format `claude-dev:dev-<timestamp>-<sha>`
-- Runtime compose definition is `claude-dev/docker-compose.yml`; CI syncs it to `/volume1/docker/claude-dev/docker-compose.yml`
-- Runtime requirements preserved: `network_mode: host`, `/volume1/repos` mount, `/root/.claude` mount, `/root/.claude.json` mount, `/root/.gitea_token` mount
-- Rollback: run workflow manually with `rollback_image` (for example the value in `/volume1/docker/claude-dev/previous-image.txt`) to redeploy a known-good immutable tag via CI
+- Source in `claude-dev/` (Dockerfile, required-tools.txt, scripts/)
+- CI-first: update `claude-dev/*` in git, then push
+- Workflow: `.gitea/workflows/deploy-claude-dev-dev.yml`
+- Uses NAS runner with `DOCKER_BUILDKIT=0`
+- Immutable tag format: `claude-dev:dev-<timestamp>-<sha>`
+- Runtime compose: `claude-dev/docker-compose.yml`
+- Rollback: run workflow with `rollback_image` using value from `/volume1/docker/claude-dev/previous-image.txt`
 
 ## Synology Quirks
 - No `crontab` — edit `/etc/crontab` with `sudo` + `printf`, restart crond
 - No `scp` subsystem — use `cat file | ssh ... 'cat > dest'`
-- ACLs override Unix perms — use `sudo chmod 777` + `synoacltool -enforce-inherit` for bind mounts
+- ACLs override Unix perms — use `sudo chmod 777` + `synoacltool -enforce-inherit`
+- Tailscale userspace mode — cannot bind Docker ports to Tailscale IPs
 
 ## DNS
 - dnsmasq on NAS resolves `*.jimmygan.com` → `161.33.182.109` (Caddy VPS)
 - Tailscale split DNS: `jimmygan.com` → NAS (queries hit dnsmasq)
-- All paths (LAN, Tailscale, public) route through Caddy for TLS
 
 ## Proxy API
 - URL: `https://www.bytecatcode.org`
-- Models: `claude-opus-4-6`, `claude-sonnet-4-6`, `claude-haiku-4-5-20251001` (no date-suffixed names)
+- Models: `claude-opus-4-6`, `claude-sonnet-4-6`, `claude-haiku-4-5-20251001`
+
+## Key Files
+- `docs/architecture.md` — full architecture reference
+- `docs/TESTING.md` — test patterns and conventions
+- `.gitea/workflows/deploy.yml` — main CI
+- `.gitea/workflows/deploy-dev.yml` — dev CI

dashboard/backend/tests/conftest.py

@@ -71,7 +71,7 @@ def temp_rbac_file(temp_volume_root):
     data = {
         "role_defaults": {
             "admin": {"pages": "*", "sidebar_links": "*"},
-            "member": {"pages": ["dashboard", "docker"], "sidebar_links": "*"},
+            "member": {"pages": ["dashboard"], "sidebar_links": "*"},
             "viewer": {"pages": ["dashboard"], "sidebar_links": ["dashboard"]},
         },
         "user_overrides": {},
@@ -470,11 +470,11 @@ def mock_rbac_data():
     return {
         "role_defaults": {
             "admin": {"pages": "*", "sidebar_links": "*"},
-            "member": {"pages": ["dashboard", "docker", "files"], "sidebar_links": "*"},
+            "member": {"pages": ["dashboard", "files"], "sidebar_links": "*"},
             "viewer": {"pages": ["dashboard"], "sidebar_links": ["dashboard"]},
         },
         "user_overrides": {
-            "customuser": {"pages": ["dashboard", "docker"], "sidebar_links": ["dashboard", "docker", "files"]}
+            "customuser": {"pages": ["dashboard"], "sidebar_links": ["dashboard", "files"]}
         },
     }
 

dashboard/backend/tests/integration/test_docker.py

@@ -1,158 +0,0 @@
-"""
-Integration tests for Docker operations.
-"""
-
-import pytest
-
-
-class TestDockerContainerList:
-    """Test listing Docker containers."""
-
-    def test_list_containers_success(self, test_app, admin_headers):
-        """Test listing containers successfully."""
-        response = test_app.get("/api/docker/containers", headers=admin_headers)
-
-        assert response.status_code == 200
-        data = response.json()
-        assert isinstance(data, list)
-        if len(data) > 0:
-            container = data[0]
-            assert "id" in container
-            assert "name" in container
-            assert "status" in container
-
-    def test_list_containers_without_auth(self, test_app):
-        """Test listing containers without authentication."""
-        response = test_app.get("/api/docker/containers")
-
-        assert response.status_code == 401
-
-
-class TestDockerContainerActions:
-    """Test Docker container start/stop/restart actions."""
-
-    def test_start_container_success(self, test_app, admin_headers, mock_docker_client):
-        """Test starting a container."""
-        container_id = "abc123"
-
-        response = test_app.post(f"/api/docker/containers/{container_id}/start", headers=admin_headers)
-
-        assert response.status_code == 200
-        data = response.json()
-        assert data["ok"] is True
-
-    def test_stop_container_success(self, test_app, admin_headers, mock_docker_client):
-        """Test stopping a container."""
-        container_id = "abc123"
-
-        response = test_app.post(f"/api/docker/containers/{container_id}/stop", headers=admin_headers)
-
-        assert response.status_code == 200
-
-    def test_restart_container_success(self, test_app, admin_headers, mock_docker_client):
-        """Test restarting a container."""
-        container_id = "abc123"
-
-        response = test_app.post(f"/api/docker/containers/{container_id}/restart", headers=admin_headers)
-
-        assert response.status_code == 200
-
-    def test_container_action_without_auth(self, test_app):
-        """Test container action without authentication."""
-        response = test_app.post("/api/docker/containers/abc123/start")
-
-        assert response.status_code == 401
-
-    def test_container_action_invalid_action(self, test_app, admin_headers):
-        """Test container action with invalid action."""
-        response = test_app.post("/api/docker/containers/abc123/invalid_action", headers=admin_headers)
-
-        assert response.status_code == 400
-        data = response.json()
-        assert "detail" in data
-
-
-class TestDockerContainerLogs:
-    """Test Docker container logs retrieval."""
-
-    def test_get_container_logs_success(self, test_app, admin_headers, mock_docker_client):
-        """Test getting container logs."""
-        container_id = "abc123"
-
-        response = test_app.get(f"/api/docker/containers/{container_id}/logs", headers=admin_headers)
-
-        assert response.status_code == 200
-        data = response.json()
-        assert "logs" in data or isinstance(data, str)
-
-    def test_get_container_logs_with_tail(self, test_app, admin_headers, mock_docker_client):
-        """Test getting container logs with tail limit."""
-        container_id = "abc123"
-
-        response = test_app.get(f"/api/docker/containers/{container_id}/logs?tail=100", headers=admin_headers)
-
-        assert response.status_code == 200
-
-    def test_get_container_logs_without_auth(self, test_app):
-        """Test getting logs without authentication."""
-        response = test_app.get("/api/docker/containers/abc123/logs")
-
-        assert response.status_code == 401
-
-    def test_get_container_logs_nonexistent(self, test_app, admin_headers, mock_docker_client):
-        """Test getting logs for nonexistent container."""
-        # Note: In this test setup, the mock always returns a container
-        # Testing actual error handling would require a different approach
-        # For now, verify the endpoint works with the mock
-        response = test_app.get("/api/docker/containers/nonexistent/logs", headers=admin_headers)
-
-        # With the current mock setup, this will succeed
-        assert response.status_code == 200
-
-
-class TestDockerPermissions:
-    """Test Docker operation permissions."""
-
-    @pytest.fixture
-    def member_token(self, mock_config, temp_auth_file, temp_rbac_file):
-        """Create token for member role."""
-        from datetime import timedelta
-
-        from auth_service import create_access_token
-
-        return create_access_token(data={"sub": "memberuser", "role": "member"}, expires_delta=timedelta(minutes=30))
-
-    @pytest.fixture
-    def member_headers(self, member_token):
-        """Headers with member access token."""
-        return {"Authorization": f"Bearer {member_token}"}
-
-    def test_member_can_list_containers(self, test_app, member_headers):
-        """Test that member can list containers."""
-        response = test_app.get("/api/docker/containers", headers=member_headers)
-
-        # Members should be able to view containers
-        assert response.status_code == 200
-
-    def test_member_cannot_start_container(self, test_app, member_headers):
-        """Test that member cannot start containers (admin only)."""
-        response = test_app.post("/api/docker/containers/abc123/start", headers=member_headers)
-
-        # Should be forbidden for non-admin
-        assert response.status_code == 403
-
-    def test_viewer_can_list_containers(self, test_app, mock_config, temp_auth_file, temp_rbac_file):
-        """Test that viewer can list containers if they have docker page access."""
-        from datetime import timedelta
-
-        from auth_service import create_access_token
-
-        viewer_token = create_access_token(
-            data={"sub": "vieweruser", "role": "viewer"}, expires_delta=timedelta(minutes=30)
-        )
-        headers = {"Authorization": f"Bearer {viewer_token}"}
-
-        response = test_app.get("/api/docker/containers", headers=headers)
-
-        # Viewer may not have docker page access by default
-        assert response.status_code in [200, 403]

docs/architecture.md

@@ -0,0 +1,100 @@
+# NAS Tools — Architecture & Operations
+
+## Overview
+
+NAS Dashboard is a web UI for managing a Synology DS224+ NAS and its ecosystem of services. It runs as a Docker container on the NAS with a FastAPI backend + Svelte 5 frontend.
+
+## Architecture
+
+```
+Browser → VPS Caddy (161.33.182.109, port 443)
+         → Tailscale → NAS Caddy (port 8443, fallback only)
+         → NAS Dashboard (port 4000)
+```
+
+DNS: `*.jimmygan.com` → VPS Caddy → Tailscale → NAS services.
+
+## Services
+
+### Media
+| Service | URL | Port | Description |
+|---------|-----|------|-------------|
+| Navidrome | music.jimmygan.com | 4533 | Music streaming |
+| Jellyfin | media.jimmygan.com | 8097 | Media server |
+| Audiobookshelf | books.jimmygan.com | 13378 | Audiobooks & podcasts |
+| Immich | photos.jimmygan.com | 2283 | Photo/video backup |
+| t-youtube | yt.jimmygan.com | 8001 | YouTube subscription reader |
+| Transmission | (sidebar route) | 9091 | BitTorrent client |
+
+### Tools
+| Service | URL | Description |
+|---------|-----|-------------|
+| Gitea | git.jimmygan.com | Git hosting + CI (Gitea Actions) |
+| LiteLLM | (dashboard route) | LLM proxy |
+| OPC | (dashboard route) | Kanban + agents |
+| n8n | n8n.jimmygan.com | Workflow automation |
+| Vaultwarden | vault.jimmygan.com | Password manager |
+| Hermes | hermes-agent.nousresearch.com/docs | AI Agent docs |
+
+## CI/CD (Gitea Actions)
+
+### Workflows
+- **deploy.yml** — Main branch: tests → build → deploy to production
+- **deploy-dev.yml** — Dev branch: tests → build → deploy to dev
+
+### Runners
+- **VPS runner** (`vps` label) — server2 (Oracle Tokyo, 158.101.140.85). Used for build + test jobs. Uses podman.
+- **NAS runner** (`nas` label) — DS224+. Used for deploy jobs. Uses Docker via `/volume1/@appstore/ContainerManager/usr/bin/docker`. Runs with `nas:host` label (host networking).
+
+### Build & Deploy Flow
+1. VPS runner builds Docker image
+2. Pushes to local registry: `100.78.131.124:5501/nas-dashboard:latest` (podman push --tls-verify=false)
+3. Streams image to NAS via SSH pipe: `podman save | gzip | ssh nasts "gunzip | docker load"`
+4. NAS runner deploys: `docker compose up -d`
+
+### CI Speed Notes
+- Build is the bottleneck (~12 min first run due to npm ci + Vite build)
+- npm ci is cached between runs (node_modules saved to disk on VPS)
+- pip is cached between runs
+- Registry push enables future layer caching
+- Synology Docker daemon doesn't support `--tls-verify=false` for pulls; SSH pipe used instead
+
+### Recent CI Fixes (July 2026)
+| Fix | Date | Description |
+|-----|------|-------------|
+| Checkout URL | Jul 8 | Changed from `gitea:3000` to `127.0.0.1:3300` (NAS host networking) |
+| TRANSMISSION_USER/PASS | Jul 8 | Added missing env vars to deploy step |
+| Gzip transfer | Jul 8 | Added gzip compression to SSH pipe |
+| Local registry | Jul 9 | Build pushes to local registry for layer caching |
+| NAS runner crash | Jul 8 | `.runner` address changed from `gitea:3000` to `127.0.0.1:3300` |
+
+## Local Registry
+
+Registry container: `registry-mirror` on NAS, port 5501.
+- Standalone mode (NOT Docker Hub proxy — Hub is blocked from China)
+- Listens on `0.0.0.0:5501`
+- Images pushed by VPS CI, pulled by NAS fallback via SSH pipe
+- API: `http://100.78.131.124:5501/v2/`
+
+## Dashboard Sidebar Layout
+
+### Main
+Overview, LiteLLM, OPC, Files, Terminal, Security
+
+### Media
+Navidrome, Jellyfin, Audiobookshelf, Immich, t-youtube, Media Management, Transmission
+
+### Tools
+Hermes, Chat Digest, Code Sessions, Repos, Gitea Web, Stirling PDF, Vaultwarden, n8n, Speedtest
+
+All external links use clean HTTPS URLs (no `:8443` port).
+
+## Cmd+K Palette
+Press `⌘K` (Mac) or `Ctrl+K` to fuzzy-search all sidebar items. New in July 2026.
+
+## NAS Quirks
+- No `crontab` — edit `/etc/crontab` with `sudo` + `printf`, restart crond
+- No `scp` subsystem — use `cat file | ssh ... 'cat > dest'`
+- ACLs override Unix perms — use `sudo chmod 777` + `synoacltool -enforce-inherit`
+- Docker at `/volume1/@appstore/ContainerManager/usr/bin/docker`
+- Tailscale userspace mode — cannot bind to Tailscale IP directly in Docker