📋 Release Notes

🔄 Account Switcher

• New Switch Account option in the header — change GitHub accounts without restarting the server.
• Clears all cached tokens (GitHub, Copilot, disk) and starts a fresh device code flow immediately.

🔐 Auth Race Condition Fix

• Fixed: signing in with a non-Copilot account then switching to the correct account left the UI stuck on "Waiting for authorization..." forever.
• Root cause: background poll thread and client poll raced to call poll_device_code(). Now uses a single-writer pattern — the bg thread is the sole caller, the client reads a shared result.
• NO_COPILOT_ACCESS errors now surface in the UI with the username and links to Switch Account or Sign up for Copilot.
• Client-side poll now times out after 15 minutes (matching GitHub's device code expiry) instead of polling forever.
• Stale Copilot session cache is cleared when starting a new login flow.

🏷️ Version in Header

• The brainstem version now displays in the chat UI header — no more guessing which version you're running.

🔒 Diagnostics Privacy

• The Get Help report now scrubs user_code, device_code, and session_id from diagnostics before publishing to GitHub issues.

🔧 Other Fixes

• Friendly error message when GitHub API quota is exceeded.
• Installer handles force-push divergence and kills existing server on port 7071 before starting.
• 400 error fallback and gh CLI issue creation when API token lacks repo scope.
• Better error telemetry throughout the auth and diagnostics pipeline.

🎯 Local-Only Agents

• Removed the remote agent system — no more hot-loading agents from GitHub repos at runtime.
• Only *_agent.py files directly in the agents/ folder are discovered and loaded.
• Subdirectories like agents/experimental/ are intentionally excluded from auto-discovery.
• Removed the ⚡ Sources sidebar panel, all /repos/* API endpoints, and associated JS/CSS from the chat UI.

🔄 Stateless Agent Loading

• Agents now load fresh on every request — no in-memory cache between calls.
• Removed _agents_cache and reload_agents() — the brainstem is fully stateless per invocation.
• Makes local development and debugging faster: edit an agent file, hit the endpoint, see the change immediately.

📦 Version Tracking & Auto-Upgrade

• Added rapp_brainstem/VERSION file as the single source of truth for the current version.
• New GET /version endpoint returns {"version": "0.1.0"}.
• Version now included in GET /health response and server startup banner.
• The install.sh and install.ps1 one-liner scripts now compare the local VERSION against the remote one hosted on GitHub.
• If the local version matches the remote, the installer exits with "Already up to date" — no unnecessary re-cloning.
• If the remote version is newer, the installer proceeds with a full upgrade (git pull + pip install).

🔧 Other Improvements

• Improved /health endpoint: returns 200 with "status": "unauthenticated" when the Copilot token is missing, instead of a 500 error. The chat UI shows the login overlay instead of failing silently.
• GitHub token refresh and caching improvements (from upstream).
• Copilot Studio converter agent moved to agents/experimental/ (excluded from auto-discovery).