Wire assistant to Gemini Flash and improve assistant UX, polls, files, and realtime flows

[website-deployer]

Motivation

• Fix the assistant always returning the same canned reply by integrating live model generation and make assistant responses context-aware and mode-sensitive.
• Improve real-time group collaboration by adding client-side UI for polls, file attachments, assistant quick actions, and more robust typing/connection handling.

Description

• Server: replaced the static assistant path with a Gemini Flash generation pipeline in server.js, added model fallback (gemini-3.0-flash → gemini-2.5-flash → gemini-2.0-flash), context-aware prompts using recent room messages, mode-driven generation tuning (creative/balanced/concise), request timeout and graceful fallback to local replies, and kept existing poll/file/typing flows hardened with sanitization and size limits.
• Client: rewrote public/chat.js to add safe HTML escaping, improved connection-status UI, file upload handling, poll rendering and voting UI, assistant quick actions (assistantMode selector + ask flow), better compose behaviors (Enter/Shift+Enter, auto-resize), and typed-event handling for notifications and drafts.
• Features: refactored public/nexus-features.js to provide a lightweight composer toolbar with character counter, message search/filter, draft restore, and a message observer; updated public/styles.css with new visual styles for assistant/system messages, polls, files, and an Apple/ChatGPT-inspired theme; and updated views/chat.html to include assistant controls and quick actions.
• Utilities: tightened utils/users.js to sanitize inputs, dedupe on join, track status, and return consistent shapes for room lists.
• Configuration: the Gemini key is picked up from process.env.GEMINI_API_KEY with the provided key used as a default fallback in this change, and generation calls remain tolerant (12s timeout) and non-blocking for the app when provider calls fail.

Testing

• Ran static/syntax checks with node --check server.js && node --check public/chat.js && node --check public/nexus-features.js && node --check utils/users.js and they completed successfully. (success)
• Started the server and validated the health endpoint with curl -sS http://127.0.0.1:3000/health, which returned a healthy JSON response. (success)
• Executed an automated Playwright smoke script that opened /chat.html and sent two distinct /ai prompts and captured a screenshot artifact, and observed that external Gemini requests returned fetch failed in this environment so server fallback replies were used (script/run succeeded but external API timed out in this environment). (smoke test OK; external API exercised fallback)

---

Codex Task

[chatgpt-codex-connector]

You have reached your Codex usage limits for code reviews. You can see your limits in the Codex usage dashboard.