Summary

Salvage of PR #6263 by @Helmi (cherry-picked onto current main) with two bug fixes.

Adds a pre-timeout warning notification to the gateway inactivity timeout. When the agent has been idle for gateway_timeout_warning seconds (default 900s / 15 min), a single notification is sent. If inactivity continues to gateway_timeout (default 1800s / 30 min), the full timeout fires as before.

This gives users on slow API providers a heads-up instead of a surprise timeout.

Bug fixes on top of the original PR

① Unbound exception variable — the original had except Exception: followed by logger.debug('...%s', _ne). The _ne variable wasn't bound to this exception (it existed in an outer scope from a different try/except). Fixed: except Exception as _warn_err:

② Warning message remaining-time math — the original said "timed out in {_warn_mins} min" where _warn_mins = _agent_warning // 60. That's the elapsed time, not the remaining time. With defaults (warning=15min, timeout=30min) it happened to be correct, but with warning=10min / timeout=30min it would say "timed out in 10 min" when 20 minutes remain. Fixed: _remaining_mins = (_agent_timeout - _agent_warning) // 60

Changes

• gateway/run.py — Warning check in poll loop, config bridge for env var
• hermes_cli/config.py — gateway_timeout_warning: 900 in DEFAULT_CONFIG
• cli-config.yaml.example — Documented both timeout keys
• tests/gateway/test_gateway_inactivity_timeout.py — 8 tests, all pass

Config

agent:
  gateway_timeout: 1800          # full timeout (seconds, 0 = unlimited)
  gateway_timeout_warning: 900   # warning before timeout (seconds, 0 = disable)

Example message at 15 min idle:

⚠️ No activity for 15 min. If the agent does not respond soon, it will be timed out in 15 min. You can continue waiting or /reset.

Fixes #6260