Fix copilot CLI session termination in web terminal

[Copilot]

Copilot CLI sessions launched from the session picker were being destroyed when the copilot process exited, ejecting users back to the session picker. Long-running tasks also caused silent WebSocket disconnects due to nginx's default 60s proxy timeout.

Session persistence (session-picker.py)

Copilot was run as the direct initial tmux command — when the process exits (agent handoff, task phase completion, error), tmux destroys the window and session. Fix wraps the launch in a persistent shell:

# Before: session dies when copilot exits
os.execvp("tmux", ["tmux", "new-session", "-s", name, "copilot", "--continue"])

# After: copilot exits → user drops to zsh, session survives
os.execvp("tmux", ["tmux", "new-session", "-s", name,
                    "-c", "/home/agentbox/workspace",
                    "zsh", "-i", "-c", "copilot --continue; exec zsh"])

Applied to [c], [r], and c<N> (resume-by-id) options.

WebSocket keepalive (Dockerfile, session-manager.py, entrypoint.sh)

• Added proxy_read_timeout 86400s / proxy_send_timeout 86400s to all nginx terminal locations (main /terminal/ and dynamically generated per-session configs)
• Added --ping-interval 30 to all ttyd invocations — default was 300s, well past nginx's 60s cutoff
• Extracted TTYD_PING_INTERVAL and NGINX_PROXY_TIMEOUT constants in session-manager.py for consistency

Original prompt

do you have any idea why the web terminal kills the copilot cli session after it starts doing something? it terminates "copilot" amd gets back to the zsh entry point. I can see the previous commands there

---

💡 You can make Copilot smarter by setting up custom instructions, customizing its development environment and configuring Model Context Protocol (MCP) servers. Learn more Copilot coding agent tips in the docs.