Refactor: largest files audit

[serge-ivo]

Audit of the largest TypeScript source files (excluding node_modules, dist, and unit *.test.*). Purpose: identify refactor candidates — files that are too long and/or mix multiple responsibilities. No refactoring in this issue — just the list + rationale, to be split into follow-up tasks.

Command used:

find . -type f \( -name "*.tsx" -o -name "*.ts" \) ! -path "*/node_modules/*" \
  ! -path "*/.git/*" ! -path "*/dist/*" ! -name "*.test.ts" ! -name "*.test.tsx" \
  -exec wc -l {} + | sort -rn | head

Top 15 by line count

LOC	File	What it is / why it's a refactor candidate
2029	workers/mcp/src/instance-tools.ts	Every instance-scoped MCP tool (schema + handler) in one file (~dozens of tools). Split by tool group (knowledge, coding, triggers, board, settings, …).
1470	workers/api/src/routes/coding.ts	ALL coding routes — repos, sessions, deployments/builds, issues, overseer, engines, work-mode. Too many endpoints in one router; split by sub-domain.
1317	workers/api/src/routes/instances.ts	Instance CRUD + voice-settings + typed settings + language sync + more. Mixed concerns; extract voice-settings and settings routers.
1290	workers/api/src/agent-do.ts	The AgentDO Durable Object: chat loop, tool execution, storage wiring, repo ingest, alarm state machine. Classic god-object; split the ingest/alarm and chat loop out.
1276	store/console/src/tabs/SettingsTab.tsx	One giant Settings component (voice, translation, board maintenance, runner info, danger zone, …). Split into per-section subcomponents.
1251	workers/api/src/agent-storage.ts	Storage engine spanning collections, files, vectors, activity, summaries, memory — many domains in one class. Split per storage domain.
1155	packages/browser-runner/src/runner.ts	Runner entry: Playwright + tmux + relay transport + task dispatch. Mixes transport and task handlers; extract the task registry.
1064	workers/mcp/src/index.ts	MCP server bootstrap + account-tool registration + dispatch. Split registration from transport/dispatch.
1064	packages/sdk/src/voice/use-voice.ts	The voice hook — STT/TTS/VAD/modes/commands/timers/echo-guard all in one useVoice. Very high complexity; extract sub-hooks (e.g. useStt, useTurnTimers, useCommands).
986	e2e/console.spec.ts	Playwright e2e suite for the console (one big spec). Lower priority (tests), but could split by feature area for readability.
788	workers/api/src/routes/instances-runtime.ts	Runtime/relay node routes (register/resolve/status). Cohesive but long; candidate to split node-management vs runtime-call.
763	agents/coder/web/src/CodingTab.tsx	The Coder UI container: repos, sessions, terminal/co-pilot, header override, builds toggle, loop, engines. Many responsibilities in one component — prime split candidate (header, session view, landing view).
748	workers/api/src/lib/storage-tools.ts	executeStorageTool — one big switch over all storage + coding tools. Split handlers per group.
725	packages/cli/src/commands/runner.ts	CLI runner command (connect, relay, multiplex, reattach). Long single command; extract helpers.
720	store/console/src/pages/InstanceDetail.tsx	Instance detail page: chat thread + tab routing + voice + loop + gloss. Mixed concerns; extract the chat thread component.

Suggested next steps

• Prioritize the god-objects first: agent-do.ts, agent-storage.ts, instance-tools.ts, and use-voice.ts — highest complexity-per-file and most-touched.
• UI: SettingsTab.tsx, CodingTab.tsx, InstanceDetail.tsx split into subcomponents (they already have natural section boundaries).
• Each split should be behavior-preserving with the existing tests kept green; file one follow-up issue per target.

[serge-ivo]

Audit complete — this ticket was scoped as 'the list + rationale, split into follow-up tasks' (no refactoring in-issue), and the list is in the body.

Filed the prioritized god-object follow-ups (the issue's 'prioritize these first'):

• Refactor: split workers/mcp/src/instance-tools.ts (2029 LOC) by tool group #135 — split workers/mcp/src/instance-tools.ts (2029 LOC) by tool group
• Refactor: split the AgentDO god-object (workers/api/src/agent-do.ts, 1298 LOC) #136 — split the AgentDO god-object workers/api/src/agent-do.ts (1298 LOC)
• Refactor: split agent-storage.ts (1262 LOC) per storage domain #137 — split agent-storage.ts (1262 LOC) per storage domain
• Refactor: extract sub-hooks from packages/sdk/src/voice/use-voice.ts (1127 LOC) #138 — extract sub-hooks from packages/sdk/src/voice/use-voice.ts (1127 LOC)

The remaining audit candidates (routes/coding.ts, routes/instances.ts, SettingsTab/CodingTab/InstanceDetail UI splits, storage-tools.ts, runner.ts, cli runner.ts, etc.) stay documented in the table above as a backlog; file per-target issues as they're picked up. Each split must be behavior-preserving with existing tests kept green.