Fix reload_plugins MCP hot-swap + robustness/metrics pass

[claudiusthebot]

Summary

• Bug: reload_plugins tool call caused hanging responses because the plugin registry updated but the active Claude Agent SDK query retained stale MCP server config.
• Fix: Store active Query reference per chat and call setMcpServers() on it after plugin reload to hot-swap MCP servers immediately.
• Robustness pass: Comprehensive quality improvements across 14 files — eliminated silent error swallowing, added structured logging to every empty catch block, capped unbounded maps, added metrics module, and improved test coverage.

Changes

MCP hot-swap fix

File	Change
src/backend/claude-sdk/handler.ts	Track active Query in a Map, expose getActiveQuery(), clean up in finally block, wire metrics
src/backend/claude-sdk/options.ts	Export buildMcpServers() for reuse outside buildSdkOptions()
src/backend/claude-sdk/index.ts	Re-export getActiveQuery and buildMcpServers
src/core/types.ts	Add optional refreshMcpServers?(chatId) to QueryBackend interface
src/bootstrap.ts	Wire refreshMcpServers — direct setMcpServers() call (no timeout, SDK handles it)
src/core/gateway-actions.ts	Call backend.refreshMcpServers() after reload, prefer body._chatId for non-numeric IDs

Silent error swallowing fixes

File	Change
src/frontend/telegram/actions.ts	Log HTML parse failures, scheduled message failures, and emoji reaction fallbacks instead of swallowing
src/frontend/telegram/handlers.ts	Log admin check failures and rate-limit triggers
src/core/cron.ts	Log invalid cron schedule parse errors
src/storage/media-index.ts	Log media index save failures

Empty catch block fixes

File	Change
src/core/heartbeat.ts	Write log errors to stderr instead of swallowing
src/core/dream.ts	Write log errors to stderr instead of swallowing
src/util/trace.ts	Write trace failures to stderr (avoids recursion)
src/util/cleanup-registry.ts	Write cleanup handler errors to stderr

Robustness

File	Change
src/frontend/telegram/handlers.ts	Cap verifiedGroups (1000) and unauthorizedCooldown (5000) maps to prevent unbounded growth
src/core/tools/mcp-server.ts	Add unhandledRejection handler to prevent silent MCP subprocess crashes

Metrics

File	Change
src/util/metrics.ts	New — lightweight in-process counters + histograms (p50/p95/p99)
src/backend/claude-sdk/handler.ts	Wire response_latency_ms, queries_total, tool_calls.*, errors.* metrics

Tests

File	Change
src/__tests__/metrics.test.ts	New — 5 tests (counters, percentiles, histogram cap, reset, empty state)
src/__tests__/sessions.test.ts	Delete expect(true).toBe(true) placeholder test
src/__tests__/reload-plugins.test.ts	Tests for _chatId override, String(chatId) fallback, MCP refresh lifecycle

Test plan

• All 1334 tests pass (47 files)
• TypeScript type-check passes (tsc --noEmit)
• Manual test: reload plugins mid-conversation and verify new plugin tools work immediately
• Verify structured logs appear for previously-silent error paths
• Check metrics via handler after a few queries

[Copilot]

Pull request overview

Fixes Claude Agent SDK MCP server hot-swap during reload_plugins by tracking the in-flight SDK Query per chat and updating its MCP server config immediately after plugin reload, so newly reloaded plugin tools are usable in the same turn.

Changes:

• Track and expose the active Claude SDK Query per chat to enable mid-turn control operations (e.g., setMcpServers()).
• Export/reuse MCP server construction (buildMcpServers) and wire a new optional backend hook refreshMcpServers(chatId).
• Invoke refreshMcpServers after reload_plugins and surface added/removed/errors info in the action response.

Reviewed changes

Copilot reviewed 6 out of 6 changed files in this pull request and generated 2 comments.

Show a summary per file

File	Description
src/backend/claude-sdk/handler.ts	Stores active in-flight Query in a map and exposes getActiveQuery() for hot-swap operations.
src/backend/claude-sdk/options.ts	Exports buildMcpServers() for reuse outside the options builder.
src/backend/claude-sdk/index.ts	Re-exports getActiveQuery and buildMcpServers from the backend barrel.
src/core/types.ts	Adds optional refreshMcpServers(chatId) to the QueryBackend interface.
src/bootstrap.ts	Implements refreshMcpServers for the Claude SDK backend by calling Query.setMcpServers().
src/core/gateway-actions.ts	Calls backend.refreshMcpServers() after plugin reload and appends a status summary to the response.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[Copilot]

Pull request overview

Copilot reviewed 7 out of 7 changed files in this pull request and generated 2 comments.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[Copilot]

Pull request overview

Copilot reviewed 7 out of 7 changed files in this pull request and generated 1 comment.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[Copilot]

Pull request overview

Copilot reviewed 19 out of 19 changed files in this pull request and generated 3 comments.

Comments suppressed due to low confidence (1)

src/frontend/telegram/actions.ts:76

• In sendText, the catch assumes any sendMessage(..., { parse_mode: "HTML" }) failure is an HTML parse error and then retries without parse_mode. This can be misleading in logs (network/403/etc will also be reported as “HTML parse failed”) and may cause an unintended second send attempt for non-parse failures. Consider only falling back when the error is a Telegram “can't parse entities”/400 parse error, and otherwise rethrow (or at least log the original error as a generic send failure).

  try {
    const sent = await bot.api.sendMessage(chatId, html, {
      parse_mode: "HTML",
      reply_parameters: replyTo ? { message_id: replyTo } : undefined,
    });
    return sent.message_id;
  } catch (err) {
    logWarn(
      "bot",
      `sendText HTML parse failed (chat=${chatId}): ${err instanceof Error ? err.message : err}`,
    );
    const sent = await bot.api.sendMessage(chatId, text, {
      reply_parameters: replyTo ? { message_id: replyTo } : undefined,
    });
    return sent.message_id;
  }

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[Copilot]

Pull request overview

Copilot reviewed 19 out of 19 changed files in this pull request and generated 2 comments.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[Copilot]

Pull request overview

Copilot reviewed 19 out of 19 changed files in this pull request and generated 4 comments.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[Copilot]

Pull request overview

Copilot reviewed 19 out of 19 changed files in this pull request and generated 4 comments.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[Copilot]

Pull request overview

Copilot reviewed 19 out of 19 changed files in this pull request and generated 2 comments.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[Copilot]

Pull request overview

Copilot reviewed 19 out of 19 changed files in this pull request and generated 1 comment.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[Copilot]

Pull request overview

Copilot reviewed 23 out of 23 changed files in this pull request and generated 1 comment.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[Copilot]

Pull request overview

Copilot reviewed 25 out of 25 changed files in this pull request and generated 4 comments.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.