Problem

A channel that fails its initial connection rethrows from IHostedService.StartAsync. An unhandled exception from StartAsync aborts the .NET Generic Host and triggers Akka CoordinatedShutdown — so a single misconfigured channel takes the entire daemon down (Slack, reminders, HTTP API, everything).

This surfaced in #1033: a Discord bot without the Message Content privileged intent gets gateway close 4014 ("Disallowed intent(s)"), and the whole daemon shuts down.

Both DiscordChannel.StartAsync and SlackChannel.StartAsync had this throw;. The catch blocks emitted a Warning-severity ChannelDisconnected alert and then rethrew — warn, then kill the process. That mismatch shows the daemon-wide blast radius was never intended. This is V1 prototype behavior we are reversing: a misconfigured channel should degrade in isolation, not crash the process.

Desired behavior

• A channel that can't connect leaves the daemon and all other subsystems running.
• The failed channel reports disconnected (with a reason) via netclaw status and a channel.disconnected operational alert.
• Transient failures (network/gateway blip) retry on a background reconnect loop with capped backoff.
• Fatal failures (bad token, disallowed/invalid intents, missing OAuth scope) stay disconnected until the operator fixes config and restarts — retrying would loop forever.

Status

Fix implemented (commit 21dc4f00):

• ChannelConnectException + per-channel Fatal/Transient classifiers (Discord gateway close codes; Slack auth error codes).
• DiscordNetGatewayClient observes the gateway Disconnected event and fails fast on a fatal close instead of blocking on the 30s readiness timeout.
• DiscordChannel/SlackChannel StartAsync degrade instead of rethrowing; transient failures reconnect with backoff.

Related: #1033