I need a browser automation that acts like a front-desk assistant on our client portal. The automation will: Log in with provided credentials. Monitor client messages. Draft and send replies using my GPT API (I will handle the GPT integration myself). Respect strict rules around when to reply (e.g., never interfere once a human takes over). This will allow us to handle inbound client conversations smoothly while keeping humans in control when needed. --- Key Responsibilities Login & Session Handling Secure login with credentials from .env or secret manager. Handle sessions, cookies, re-authentication gracefully. Inbox Monitoring Detect new/unread threads. Parse sender, timestamps, and full conversation history. Reply Automation (GPT API provided by me) Expose clean hooks to fetch draft replies from my GPT API. Post replies back into the portal reliably, verifying success. Human-Lock Rule If a thread is started by a human, the bot will never reply. If a human joins any ongoing thread, the bot will stop replying and lock that thread. Locked threads remain off-limits unless explicitly re-enabled by an admin. Headless Execution Must run in headless mode by default (for deployment). Support a debug option (HEADLESS=false) to run visibly for testing. Compatible with Docker/container deployment. Resilience & State Management Track thread states (bot vs. human ownership). Persist state so crashes/restarts don’t cause duplicate replies. Implement retries, backoff, and DOM-change handling. Observability Structured logs (JSON preferred). Events: new message detected, draft fetched, reply sent, locked thread skipped, errors. Simple summary dashboard/CLI output.