Summary

Hermes gateway can fail to start on Discord because the platform connect timeout is fixed/defaulted to 30 seconds. In my setup, Discord slash command registration/sync takes around 28-31 seconds. Hermes kills the connect attempt at 30s, even though Discord is still making progress, then launchd restarts the service. This can leave the gateway in a restart loop and make Hermes stop replying on Discord.

Observed behaviour

Gateway logs repeatedly show:

Connecting to discord...
[Discord] Registered /skill command with 173 skill(s) via autocomplete
✗ discord error: discord connect timed out after 30s
[Discord] Disconnected
Gateway failed to connect any configured messaging platform: discord: discord connect timed out after 30s

On one run, the gateway did connect successfully just after the old timeout window:

[Discord] Registered /skill command with 173 skill(s) via autocomplete
[Discord] Connected as Hermes#0401
✓ discord connected
Gateway running with 1 platform(s)

But because restarts were already in progress, that process was then terminated and launchd continued cycling.

Expected behaviour

If the Discord client is actively progressing, for example slash commands were registered/synced, the gateway should not be killed at exactly 30 seconds. Either:

1. the default platform connect timeout should be longer for Discord;
2. the timeout should be configurable in documented config, not only via env var;
3. slash command sync should happen after the bot is considered connected; or
4. progress during connect should refresh/extend the startup watchdog.

Reproduction notes

Environment:

Hermes installed via ~/.hermes/hermes-agent
Gateway managed by launchd on macOS
Platform: Discord
Configured slash commands: 173 skills

Failure pattern:

2026-05-04 15:11:41 Connecting to discord...
2026-05-04 15:12:09 [Discord] Registered /skill command with 173 skill(s) via autocomplete
2026-05-04 15:12:11 ✗ discord error: discord connect timed out after 30s

The key point is that Discord command registration completed at about 28s, then the whole connect attempt was killed at 30s.

Workaround

Setting this env var fixed the immediate issue:

HERMES_GATEWAY_PLATFORM_CONNECT_TIMEOUT=90

After restarting the launchd service with that env var, Discord connected successfully:

[Discord] Connected as Hermes#0401
✓ discord connected
Gateway running with 1 platform(s)

Suggested fix

Make the platform connect timeout longer by default for Discord, or move Discord slash command sync outside the critical connect timeout. A 30s hard/default timeout seems too tight for accounts with many skills/slash commands or slow Discord API responses.

It would also help to document HERMES_GATEWAY_PLATFORM_CONNECT_TIMEOUT, or expose it as something like:

gateway:
  platform_connect_timeout: 90

The current failure mode is confusing because launchd keeps the service running, but Hermes is not actually connected long enough to respond.