hermes - 💡(How to fix) Fix feat: configurable auto-resume delay for paused gateway platforms [1 pull requests]

Problem

When the Hermes Gateway loses connection to a platform (e.g. Telegram), the circuit breaker pauses it after 10 consecutive failures. The pause is permanent — next_retry is set to float("inf") — so even after the underlying network issue is resolved (e.g. Clash/VPN reconnects), the platform stays paused until the user manually runs /platform resume <name> or restarts the gateway.

This is a common pain point for users behind dynamic proxy/VPN setups (Clash, Surge, etc.), where network interruptions are temporary but the gateway has no way to recover autonomously.

Proposed Solution

Add a config option gateway.platform_resume_delay (in minutes, default 0 = backward compatible) that controls how long the gateway waits before automatically resuming a paused platform.

When > 0:

• _pause_failed_platform() sets a finite next_retry = now + delay_min * 60
• The reconnect watcher checks a new auto_resume_at field on each tick
• When the timeout expires, the platform is auto-resumed and the watcher attempts reconnection
• /platform pause (manual) does NOT set auto_resume_at, so manually paused platforms still require explicit /platform resume

Environment variable override: HERMES_PLATFORM_RESUME_DELAY

Implementation

The changes are contained in gateway/run.py (4 touch points):

1. __init__: load the config value into self._platform_resume_delay
2. _load_platform_resume_delay(): new static method reading from config.yaml or env var
3. _pause_failed_platform(): use the delay instead of float("inf") when > 0, log auto-resume info
4. Reconnect watcher: check auto_resume_at on paused platforms and auto-resume when expired

Example Config

gateway:
  platform_resume_delay: 10  # auto-resume after 10 minutes

Backward Compatibility

Default 0 preserves the current behavior (never auto-resume). No config change needed for existing setups.

PR: #pending