Greptile Summary

This PR fixes Telegram's health-monitor triggering a spurious stale-socket restart every ~35 minutes by skipping the stale-socket check when the channel is "telegram". Because Telegram uses long-polling (each poll acts as its own heartbeat), the half-dead WebSocket detection heuristic doesn't apply to it. The fix is mechanically sound and the three call sites are all updated consistently.

Key observations:

• The core guard logic in evaluateChannelHealth is correct and will prevent the unwanted Telegram restarts.
• channelId is typed as string instead of the existing ChannelId union, leaving the hardcoded "telegram" string unverified by the TypeScript compiler — a typo would silently bypass the guard.
• The wrapping if (policy.channelId !== "telegram") { ... } pattern in channel-health-policy.ts lacks indentation for its body; an early-return/inverse-guard pattern would be cleaner and less error-prone to maintain.
• The readiness.ts change moves policy creation before the !accountSnapshot null guard — a minor inefficiency and style inconsistency. Worth noting that shouldIgnoreReadinessFailure already ignored stale-socket reasons, so Telegram was never surfaced as unhealthy via the readiness endpoint even before this PR.

Confidence Score: 3/5

• The fix is logically correct but has type-safety and formatting concerns that should be addressed before merge.
• The core intent of the PR is sound — skipping the stale-socket check for Telegram long-polling is the right approach. However, using a bare string type for channelId rather than the existing ChannelId union means the hardcoded "telegram" literal has no compile-time safety net, and the wrapping guard brace pattern in channel-health-policy.ts is confusingly formatted. These are style/type-safety issues, not runtime bugs, but they reduce maintainability confidence.
• src/gateway/channel-health-policy.ts — type safety of channelId field and formatting of the guard block both need attention.

_{Last reviewed commit: 65dc78f}

PR #38405 - fix(gateway): skip stale-socket restarts for Telegram polling (#38405)

Merged after verification.

• Merge commit: a5c07fa
• Verified: pnpm build, pnpm check, pnpm test:macmini
• Autoland updates:
  CHANGELOG.md
  src/gateway/channel-health-monitor.ts
  src/gateway/channel-health-policy.test.ts
  src/gateway/channel-health-policy.ts
  src/gateway/server/readiness.test.ts
  src/gateway/server/readiness.ts
• Rationale:
  自动修复 Issue Telegram health-monitor stale-socket restart every ~35min on 2026.3.2 (post-#10795 fix) #38395
• Changelog: CHANGELOG.md updated=true required=true opt_out=false