Greptile Summary

This PR excludes Discord channels from the stale-socket health check in channel-health-policy.ts, using the same pattern as the existing Telegram exclusion. The fix correctly addresses 23+ false-positive daily restarts caused by Carbon's heartbeat ACK not updating lastEventAt.

• channel-health-policy.ts: Adds policy.channelId !== "discord" to the stale-socket guard. The fix is sound — Discord dead connections are already caught by the connected === false check above, so skipping the event-age threshold for Discord does not introduce a blindspot.
• channel-health-policy.test.ts: Correctly migrates the existing stale-socket test to use "slack" and adds a new Discord exclusion test. Minor quality concern: two pre-existing tests (webhook mode and null-lastEventAt) still use evaluateDiscordHealth's default channelId: "discord", so they now pass via the new Discord exclusion rather than the guards they were designed to exercise.

Confidence Score: 5/5

Safe to merge — the production fix is correct and the only finding is a P2 test quality issue.

The implementation change is minimal, well-motivated, and follows an established exclusion pattern. All 15 policy tests and 30 monitor tests pass. The sole comment is P2: two existing tests silently exercise the wrong guard after the Discord exclusion lands, but this does not affect runtime correctness.

Minor test quality issue in src/gateway/channel-health-policy.test.ts — webhook-mode and null-lastEventAt tests should use "slack" instead of the "discord" default to validate their intended guards.

This is a comment left during a code review.
Path: src/gateway/channel-health-policy.test.ts
Line: 179-202

Comment:
**Webhook and null-`lastEventAt` tests now pass for the wrong reason**

Both of these tests use the `evaluateDiscordHealth` helper, which defaults to `channelId: "discord"`. After this fix, Discord is excluded from stale-socket detection unconditionally, so these tests succeed because of the new `channelId !== "discord"` short-circuit — not because of the `mode !== "webhook"` or `lastEventAt != null` guards they were written to cover.

If either of those guards were accidentally deleted or broken, these tests would still pass (they silently stopped validating what their names say). Consider using a non-excluded channel ID (e.g. `"slack"`) for these two cases:

```ts
// "skips stale-socket detection for channels in webhook mode"
const evaluation = evaluateDiscordHealth(
  { running: true, connected: true, enabled: true, configured: true, lastStartAt: 0, lastEventAt: 0, mode: "webhook" },
  100_000,
  "slack",   // ← use a channel that isn't Discord-excluded
);

// "does not flag stale sockets for channels without event tracking"
const evaluation = evaluateDiscordHealth(
  { running: true, connected: true, enabled: true, configured: true, lastStartAt: 0, lastEventAt: null },
  100_000,
  "slack",
);
```

This also applies to any future test added via `evaluateDiscordHealth` that intends to test a path other than the Discord exclusion itself.

How can I resolve this? If you propose a fix, please make it concise.

_{Reviews (1): Last reviewed commit: "fix(gateway): skip stale-socket detectio..." | Re-trigger Greptile}

cc @steipete @Takhoffman — requesting review. Tak, you authored the connected === true guard in #38643, so you have the most context on this code path.

@jacobtomlinson would you be able to review this when you get a chance? Small fix — excludes Discord from stale-socket detection (same pattern as the existing Telegram exclusion). Greptile gave it 5/5 confidence.