Revamp upload workflow with Git-based submission & UI polish

.github/workflows/chat-proxy-dev.yml

@@ -4,6 +4,7 @@ on:
   push:
     branches-ignore:
       - main
+  pull_request:
   workflow_dispatch:
 
 jobs:
@@ -16,7 +17,7 @@ jobs:
       HYPHA_TOKEN: ${{ secrets.RI_SCALE_TOKEN }}
     steps:
       - name: Checkout
-        uses: actions/checkout@v4
+        uses: actions/checkout@v6
 
       - name: Setup Python
         uses: actions/setup-python@v5

.github/workflows/ci.yml

@@ -2,16 +2,15 @@ name: CI
 
 on:
   push:
-    branches:
-      - main
   pull_request:
+  workflow_dispatch:
 
 jobs:
   frontend-tests:
     runs-on: ubuntu-latest
     steps:
       - name: Checkout
-        uses: actions/checkout@v4
+        uses: actions/checkout@v6
 
       - name: Setup Node
         uses: actions/setup-node@v4
@@ -42,7 +41,7 @@ jobs:
     runs-on: ubuntu-latest
     steps:
       - name: Checkout
-        uses: actions/checkout@v4
+        uses: actions/checkout@v6
 
       - name: Setup Python
         uses: actions/setup-python@v5

.github/workflows/deploy.yml

@@ -22,7 +22,7 @@ jobs:
 
     steps:
       - name: Checkout
-        uses: actions/checkout@v4
+        uses: actions/checkout@v6
 
       - name: Setup Node
         uses: actions/setup-node@v4

CLAUDE.md

@@ -0,0 +1,165 @@
+# Model Hub - CLAUDE.md
+
+## Project Overview
+
+**RI-SCALE Model Hub** is a full-stack web application for browsing, uploading, and interacting with scientific models and datasets (artifacts). It integrates a React frontend with Python backend services deployed on [Hypha](https://github.com/amun-ai/hypha) infrastructure.
+
+## Tech Stack
+
+| Layer | Technology |
+|-------|-----------|
+| Frontend | React 18 + TypeScript 5.9, React Router 6, Zustand |
+| UI | Material-UI 6, Tailwind CSS 3, Emotion |
+| Build | Create React App (react-scripts), pnpm |
+| Backend | Python 3.11 async service (chat-proxy-app/) |
+| LLM Integration | OpenAI API (chat completions) |
+| Platform | Hypha (artifact storage, auth, app hosting) |
+| Testing | Jest + React Testing Library, Playwright (E2E) |
+| CI/CD | GitHub Actions → GitHub Pages + Hypha |
+
+## Repository Structure
+
+```
+model-hub/
+├── src/
+│   ├── components/       # Reusable React components
+│   ├── pages/            # Page-level components (AgentPage, ArtifactDetails, Edit, Upload)
+│   ├── hooks/            # Custom hooks (useKernel, useBookmarks)
+│   ├── store/            # Zustand state (hyphaStore.ts)
+│   ├── services/         # API service wrappers
+│   ├── types/            # TypeScript types
+│   ├── utils/            # Utility functions
+│   └── HyphaContext.tsx  # Hypha backend provider
+├── chat-proxy-app/
+│   └── app.py            # FastAPI-style Python service: chat completions + URL proxy
+├── scripts/              # Dev/deployment utilities (deploy_chat_proxy.py, diagnose_hub.py, etc.)
+├── docs/                 # Documentation (chat-proxy-cicd.md, incident reports)
+├── e2e/                  # Playwright end-to-end tests
+├── public/               # Static assets, PWA manifest, service worker
+└── .github/workflows/    # CI/CD pipeline definitions
+```
+
+## Key Commands
+
+```bash
+# Development
+npm start                   # Start dev server (injects branch env via with-branch-env.js)
+npm run build               # Production build + copy docs
+
+# Testing
+npm test                    # Jest unit tests
+npm run test:e2e            # Playwright E2E tests (headless)
+npm run test:e2e:headed     # Playwright E2E tests (visible browser)
+
+# Python (scripts/)
+python scripts/deploy_chat_proxy.py     # Deploy/update chat-proxy Hypha app
+python scripts/test_chat_proxy.py       # Health check chat-proxy
+python scripts/diagnose_hub.py          # Inspect hub config + permissions
+python scripts/fix_hub_permissions.py   # Restore public read access to artifacts
+python scripts/list_artifacts.py        # List all artifacts
+python scripts/upload_sample.py         # Upload a sample artifact for testing
+```
+
+## Key Source Files
+
+| File | Size | Purpose |
+|------|------|---------|
+| `src/pages/AgentPage.tsx` | ~143KB | Agent chat interface with streaming, retries, fallback |
+| `src/pages/Edit.tsx` | ~95KB | Artifact editing with RDF metadata support |
+| `src/pages/Upload.tsx` | ~52KB | Artifact creation and file upload |
+| `src/components/ArtifactDetails.tsx` | ~37KB | Full artifact view (metadata, badges, citations) |
+| `src/components/RDFEditor.tsx` | ~34KB | RDF metadata editor |
+| `chat-proxy-app/app.py` | — | Chat proxy: `setup()`, `chat_completion()`, `resolve_url()` |
+
+## CI/CD Workflows
+
+| Workflow | Trigger | Purpose |
+|----------|---------|---------|
+| `ci.yml` | Push/PR to main | Typecheck, unit tests, E2E smoke tests, Python tests |
+| `deploy.yml` | Push to main | Build + deploy frontend to GitHub Pages |
+| `chat-proxy-dev.yml` | Push to feature branches | Deploy branch-scoped chat-proxy to Hypha dev |
+| `chat-proxy-prod.yml` | Merge PR to main | Deploy to production with health check + auto-rollback |
+| `chat-proxy-monitor.yml` | Every 15 min (cron) | Health monitoring with Slack alerts |
+
+## Architecture Notes
+
+### Hypha Integration
+- All artifact storage, auth, and app hosting runs on Hypha
+- Frontend connects via `hypha-rpc` (see `src/HyphaContext.tsx` and `src/store/hyphaStore.ts`)
+- Chat proxy deployed as a Hypha app (app ID pattern: `chat-proxy[-dev-<branch-slug>]`)
+
+### Chat Proxy
+- Injects OpenAI API keys server-side so they never reach the browser
+- `resolve_url()` endpoint acts as a safe HTTP proxy with allowlist validation
+- Dev apps are per-branch; prod app auto-rolls back on health check failure
+
+### Agent Architecture
+- `AgentPage.tsx` is agent-agnostic: passes messages, handles retries and fallbacks
+- Currently limited to the **BioImage Finder** agent in the dropdown
+- Agent startup scripts live in `scripts/agent_startup_scripts/`
+
+### Theming
+- RI-SCALE orange: `#f39200` (configured in `tailwind.config.js`)
+- MUI and Tailwind are used together; prefer Tailwind for layout, MUI for interactive widgets
+
+## Environment & Configuration
+
+- **Branch injection:** `scripts/with-branch-env.js` injects `REACT_APP_BRANCH` at build time
+- **Tailwind config:** `tailwind.config.js` — custom color palette
+- **TypeScript:** `tsconfig.json` — `baseUrl: "src"` for absolute imports
+- **E2E:** `playwright.config.ts` — Chrome at 1366×900, targets localhost dev server
+
+## External Dependencies
+
+- **Hypha** — backend platform (artifact store, app hosting, authentication)
+- **OpenAI API** — LLM chat completions via chat-proxy
+- **BioImage Archive** — scientific image data source for BioImage Finder agent
+- **GitHub Pages** — static frontend hosting
+
+## Lessons Learned (Session Notes)
+
+### hypha-rpc `_rkwargs: true` — CRITICAL rule
+Every call to an artifact manager or server service that passes a **dict of named params** needs `_rkwargs: true` as the last key in the dict. Without it, the entire dict is sent as a single positional argument (bound to the first parameter, e.g. `alias`), silently breaking all real parameters.
+
+```typescript
+// ✅ CORRECT — multiple named params
+await artifactManager.list({ parent_id: ..., filters: ..., limit: ..., _rkwargs: true });
+await artifactManager.create({ alias: ..., type: ..., manifest: ..., _rkwargs: true });
+await artifactManager.read({ artifact_id: ..., _rkwargs: true });
+await artifactManager.list_files({ artifact_id: ..., _rkwargs: true });
+
+// ✅ EXCEPTION — server.generateToken passes a single dict as positional `config: TokenConfig`
+// Pydantic auto-converts the dict. NO _rkwargs here.
+await server.generateToken({ expires_in: expiresIn });
+```
+
+Reference: `../hypha/hypha/templates/ws/index.html` — the canonical JS client example.
+
+### REST API pagination wrapper
+The hypha REST endpoints (e.g. `/artifacts/{alias}/files/`, `/artifacts/{alias}/children`) return a **paginated wrapper**:
+```json
+{ "items": [...], "total": N, "offset": 0, "limit": 1000 }
+```
+Always unpack `.items`:  `const data = await res.json(); setFiles(data.items ?? data);`
+
+### Local dev serving — production build + npx serve
+- `npm start` (webpack-dev-server) fails through the svamp tunnel due to `ERR_CONTENT_DECODING_FAILED` — the tunnel proxy mangles gzip `Content-Encoding`.
+- **Workaround**: use a production build: `DISABLE_ESLINT_PLUGIN=true npm run build` then `nohup npx serve -s build -l 3000 --no-clipboard > /tmp/serve.log 2>&1 &`
+- The svamp tunnel is then started with `nohup svamp service expose model-hub --port 3000 > /tmp/tunnel.log 2>&1 &`
+- Use `nohup` (not bare `&`) so the process survives shell exit.
+
+### Stale tunnel backends cause intermittent failures
+`svamp service list` shows backend count. If >1, kill all and restart:
+```bash
+pkill -9 -f "svamp service expose model-hub"
+# wait for 0 backends, then restart
+```
+
+### Git push — use `fork` remote
+`git push origin` fails (no write access to `RI-SCALE/model-hub`). Always use:
+```bash
+git push fork feat/upload-git-workflow-ui-improvements
+```
+
+### Large static assets break tunnel loading
+PNG logos or images >1MB load very slowly through the tunnel (even if they technically work). Keep navbar logos ≤50KB. Use `sips -Z 800 original.png --out optimized.png` (macOS) to resize.