refactor: scaffold module registries and default-module layout

Additive change — existing code paths still run via inline fallbacks.
Prepares core for per-module extractions in PR #3 onward.

Four registries added with empty defaults:
  - delivery action handlers (delivery.ts)
  - router inbound gate (router.ts)
  - response dispatcher (index.ts)
  - MCP tool self-registration (container/agent-runner/src/mcp-tools/server.ts)

Default modules moved to src/modules/ for signaling:
  - src/modules/typing/       (extracted from delivery.ts)
  - src/modules/mount-security/ (moved from src/mount-security.ts)

Both are imported directly by core — no hook, no registry. Removal
requires editing core imports.

Migrator now keys applied rows by name (uniqueness) so module
migrations can pick arbitrary version numbers. Stored version column
is auto-assigned as an applied-order sequence.

sqlite_master guards added around core calls into module-owned tables
(user_roles, agent_destinations, pending_questions). No-ops today;
load-bearing after the owning modules are extracted.

MODULE-HOOK markers placed at scheduling's two skill-edit sites
(host-sweep.ts recurrence call, poll-loop.ts pre-task gate). PR #4
replaces the marked blocks when scheduling moves to its module.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

container/agent-runner/src/mcp-tools/agents.ts

@@ -9,6 +9,7 @@
  * (see mcp-tools/index.ts). The host re-checks permission on receive.
  */
 import { writeMessageOut } from '../db/messages-out.js';
+import { registerTools } from './server.js';
 import type { McpToolDefinition } from './types.js';
 
 function log(msg: string): void {
@@ -62,4 +63,4 @@ export const createAgent: McpToolDefinition = {
   },
 };
 
-export const agentTools: McpToolDefinition[] = [createAgent];
+registerTools([createAgent]);

container/agent-runner/src/mcp-tools/core.ts

@@ -12,6 +12,7 @@ import path from 'path';
 import { findByName, getAllDestinations } from '../destinations.js';
 import { getMessageIdBySeq, getRoutingBySeq, writeMessageOut } from '../db/messages-out.js';
 import { getSessionRouting } from '../db/session-routing.js';
+import { registerTools } from './server.js';
 import type { McpToolDefinition } from './types.js';
 
 function log(msg: string): void {
@@ -258,4 +259,4 @@ export const addReaction: McpToolDefinition = {
   },
 };
 
-export const coreTools: McpToolDefinition[] = [sendMessage, sendFile, editMessage, addReaction];
+registerTools([sendMessage, sendFile, editMessage, addReaction]);

container/agent-runner/src/mcp-tools/index.ts

@@ -1,59 +1,21 @@
 /**
- * MCP tools barrel — collects all tool modules and starts the server.
+ * MCP tools barrel — imports each tool module for its side-effect
+ * `registerTools([...])` call, then starts the MCP server.
  *
- * Each module exports a McpToolDefinition[] array. This file registers
- * them all with the MCP server. Adding a new tool module requires only
- * importing it here and spreading its tools array.
+ * Adding a new tool module: create the file, call `registerTools([...])`
+ * at module scope, and append the import here. No central list.
  */
-import { Server } from '@modelcontextprotocol/sdk/server/index.js';
-import { StdioServerTransport } from '@modelcontextprotocol/sdk/server/stdio.js';
-import { CallToolRequestSchema, ListToolsRequestSchema } from '@modelcontextprotocol/sdk/types.js';
-
-import type { McpToolDefinition } from './types.js';
-import { coreTools } from './core.js';
-import { schedulingTools } from './scheduling.js';
-import { interactiveTools } from './interactive.js';
-import { agentTools } from './agents.js';
-import { selfModTools } from './self-mod.js';
+import './core.js';
+import './scheduling.js';
+import './interactive.js';
+import './agents.js';
+import './self-mod.js';
+import { startMcpServer } from './server.js';
 
 function log(msg: string): void {
   console.error(`[mcp-tools] ${msg}`);
 }
 
-const allTools: McpToolDefinition[] = [
-  ...coreTools,
-  ...schedulingTools,
-  ...interactiveTools,
-  ...agentTools,
-  ...selfModTools,
-];
-
-const toolMap = new Map<string, McpToolDefinition>();
-for (const t of allTools) {
-  toolMap.set(t.tool.name, t);
-}
-
-async function startMcpServer(): Promise<void> {
-  const server = new Server({ name: 'nanoclaw', version: '2.0.0' }, { capabilities: { tools: {} } });
-
-  server.setRequestHandler(ListToolsRequestSchema, async () => ({
-    tools: allTools.map((t) => t.tool),
-  }));
-
-  server.setRequestHandler(CallToolRequestSchema, async (request) => {
-    const { name, arguments: args } = request.params;
-    const tool = toolMap.get(name);
-    if (!tool) {
-      return { content: [{ type: 'text', text: `Unknown tool: ${name}` }] };
-    }
-    return tool.handler(args ?? {});
-  });
-
-  const transport = new StdioServerTransport();
-  await server.connect(transport);
-  log(`MCP server started with ${allTools.length} tools: ${allTools.map((t) => t.tool.name).join(', ')}`);
-}
-
 startMcpServer().catch((err) => {
   log(`MCP server error: ${err instanceof Error ? err.message : String(err)}`);
   process.exit(1);

container/agent-runner/src/mcp-tools/interactive.ts

@@ -7,6 +7,7 @@
 import { findQuestionResponse, markCompleted } from '../db/messages-in.js';
 import { writeMessageOut } from '../db/messages-out.js';
 import { getSessionRouting } from '../db/session-routing.js';
+import { registerTools } from './server.js';
 import type { McpToolDefinition } from './types.js';
 
 function log(msg: string): void {
@@ -165,4 +166,4 @@ export const sendCard: McpToolDefinition = {
   },
 };
 
-export const interactiveTools: McpToolDefinition[] = [askUserQuestion, sendCard];
+registerTools([askUserQuestion, sendCard]);

container/agent-runner/src/mcp-tools/scheduling.ts

@@ -8,6 +8,7 @@
 import { getInboundDb } from '../db/connection.js';
 import { writeMessageOut } from '../db/messages-out.js';
 import { getSessionRouting } from '../db/session-routing.js';
+import { registerTools } from './server.js';
 import type { McpToolDefinition } from './types.js';
 
 function log(msg: string): void {
@@ -265,4 +266,4 @@ export const updateTask: McpToolDefinition = {
   },
 };
 
-export const schedulingTools: McpToolDefinition[] = [scheduleTask, listTasks, updateTask, cancelTask, pauseTask, resumeTask];
+registerTools([scheduleTask, listTasks, updateTask, cancelTask, pauseTask, resumeTask]);

container/agent-runner/src/mcp-tools/self-mod.ts

@@ -9,6 +9,7 @@
  * the host side (defense in depth).
  */
 import { writeMessageOut } from '../db/messages-out.js';
+import { registerTools } from './server.js';
 import type { McpToolDefinition } from './types.js';
 
 function log(msg: string): void {
@@ -140,4 +141,4 @@ export const requestRebuild: McpToolDefinition = {
   },
 };
 
-export const selfModTools: McpToolDefinition[] = [installPackages, addMcpServer, requestRebuild];
+registerTools([installPackages, addMcpServer, requestRebuild]);

container/agent-runner/src/mcp-tools/server.ts

@@ -0,0 +1,54 @@
+/**
+ * MCP server bootstrap + tool self-registration.
+ *
+ * Each tool module calls `registerTools([...])` at import time. The
+ * barrel (`index.ts`) imports every tool module for side effects, then
+ * calls `startMcpServer()` which uses whatever was registered.
+ *
+ * Default when only `core.ts` is imported: the core `send_message` /
+ * `send_file` / `edit_message` / `add_reaction` tools are available.
+ */
+import { Server } from '@modelcontextprotocol/sdk/server/index.js';
+import { StdioServerTransport } from '@modelcontextprotocol/sdk/server/stdio.js';
+import { CallToolRequestSchema, ListToolsRequestSchema } from '@modelcontextprotocol/sdk/types.js';
+
+import type { McpToolDefinition } from './types.js';
+
+function log(msg: string): void {
+  console.error(`[mcp-tools] ${msg}`);
+}
+
+const allTools: McpToolDefinition[] = [];
+const toolMap = new Map<string, McpToolDefinition>();
+
+export function registerTools(tools: McpToolDefinition[]): void {
+  for (const t of tools) {
+    if (toolMap.has(t.tool.name)) {
+      log(`Warning: tool "${t.tool.name}" already registered, skipping duplicate`);
+      continue;
+    }
+    allTools.push(t);
+    toolMap.set(t.tool.name, t);
+  }
+}
+
+export async function startMcpServer(): Promise<void> {
+  const server = new Server({ name: 'nanoclaw', version: '2.0.0' }, { capabilities: { tools: {} } });
+
+  server.setRequestHandler(ListToolsRequestSchema, async () => ({
+    tools: allTools.map((t) => t.tool),
+  }));
+
+  server.setRequestHandler(CallToolRequestSchema, async (request) => {
+    const { name, arguments: args } = request.params;
+    const tool = toolMap.get(name);
+    if (!tool) {
+      return { content: [{ type: 'text', text: `Unknown tool: ${name}` }] };
+    }
+    return tool.handler(args ?? {});
+  });
+
+  const transport = new StdioServerTransport();
+  await server.connect(transport);
+  log(`MCP server started with ${allTools.length} tools: ${allTools.map((t) => t.tool.name).join(', ')}`);
+}

container/agent-runner/src/poll-loop.ts

@@ -156,11 +156,18 @@ export async function runPollLoop(config: PollLoopConfig): Promise<void> {
     // Pre-task scripts: for any task rows with a `script`, run it before the
     // provider call. Scripts returning wakeAgent=false (or erroring) gate
     // their own task row only — surviving messages still go to the agent.
+    //
+    // MODULE-HOOK:scheduling-pre-task:start
+    // When scheduling is extracted (PR #4), `applyPreTaskScripts` moves
+    // to the scheduling module and the `/add-scheduling` skill replaces
+    // this block with a call to the module. Without scheduling installed,
+    // the block is empty (no script gating) and `keep = normalMessages`.
     const { keep, skipped } = await applyPreTaskScripts(normalMessages);
     if (skipped.length > 0) {
       markCompleted(skipped);
       log(`Pre-task script skipped ${skipped.length} task(s): ${skipped.join(', ')}`);
     }
+    // MODULE-HOOK:scheduling-pre-task:end
 
     if (keep.length === 0) {
       log(`All ${normalMessages.length} non-command message(s) gated by script, skipping query`);

src/container-runner.ts

@@ -13,11 +13,12 @@ import { CONTAINER_IMAGE, DATA_DIR, GROUPS_DIR, IDLE_TIMEOUT, ONECLI_URL, TIMEZO
 import { readContainerConfig, writeContainerConfig } from './container-config.js';
 import { CONTAINER_RUNTIME_BIN, hostGatewayArgs, readonlyMountArgs, stopContainer } from './container-runtime.js';
 import { getAgentGroup } from './db/agent-groups.js';
+import { getDb, hasTable } from './db/connection.js';
 import { getAdminsOfAgentGroup, getGlobalAdmins, getOwners } from './db/user-roles.js';
 import { initGroupFilesystem } from './group-init.js';
-import { stopTypingRefresh } from './delivery.js';
+import { stopTypingRefresh } from './modules/typing/index.js';
 import { log } from './log.js';
-import { validateAdditionalMounts } from './mount-security.js';
+import { validateAdditionalMounts } from './modules/mount-security/index.js';
 // Provider host-side config barrel — each provider that needs host-side
 // container setup self-registers on import.
 import './providers/index.js';
@@ -286,10 +287,16 @@ async function buildContainerArgs(
   // Users allowed to run admin commands (e.g. /clear) inside this container.
   // Computed at wake time: owners + global admins + admins scoped to this
   // agent group. Role changes take effect on next container spawn.
+  //
+  // Guarded: if the permissions module isn't installed, `user_roles`
+  // doesn't exist and the set stays empty — the formatter treats an
+  // empty admin set as permissionless (every sender is admin).
   const adminUserIds = new Set<string>();
-  for (const r of getOwners()) adminUserIds.add(r.user_id);
-  for (const r of getGlobalAdmins()) adminUserIds.add(r.user_id);
-  for (const r of getAdminsOfAgentGroup(agentGroup.id)) adminUserIds.add(r.user_id);
+  if (hasTable(getDb(), 'user_roles')) {
+    for (const r of getOwners()) adminUserIds.add(r.user_id);
+    for (const r of getGlobalAdmins()) adminUserIds.add(r.user_id);
+    for (const r of getAdminsOfAgentGroup(agentGroup.id)) adminUserIds.add(r.user_id);
+  }
   if (adminUserIds.size > 0) {
     args.push('-e', `NANOCLAW_ADMIN_USER_IDS=${Array.from(adminUserIds).join(',')}`);
   }

src/db/connection.ts

@@ -31,3 +31,18 @@ export function closeDb(): void {
   _db?.close();
   _db = null;
 }
+
+/**
+ * Check whether a table exists. Used by core code that touches
+ * module-owned tables so that an uninstalled module degrades silently
+ * instead of raising SQLite errors. Cheap: a single indexed lookup on
+ * sqlite_master. Results are not cached — a module install adds the
+ * table at runtime (next service start), and callers may run before
+ * or after that boundary.
+ */
+export function hasTable(db: Database.Database, name: string): boolean {
+  const row = db.prepare(`SELECT 1 FROM sqlite_master WHERE type='table' AND name = ? LIMIT 1`).get(name) as
+    | { '1': number }
+    | undefined;
+  return row !== undefined;
+}

src/db/migrations/index.ts

@@ -32,29 +32,34 @@ export function runMigrations(db: Database.Database): void {
       name    TEXT NOT NULL,
       applied TEXT NOT NULL
     );
+    CREATE UNIQUE INDEX IF NOT EXISTS idx_schema_version_name ON schema_version(name);
   `);
 
-  const currentVersion =
-    (db.prepare('SELECT MAX(version) as v FROM schema_version').get() as { v: number | null })?.v ?? 0;
-
-  const pending = migrations.filter((m) => m.version > currentVersion);
+  // Uniqueness is keyed on `name`, not `version`. This lets module
+  // migrations (added later by install skills) pick arbitrary version
+  // numbers without coordinating across modules. `version` stays on
+  // the Migration object as an ordering hint within the barrel array;
+  // the stored `version` column is auto-assigned at insert time as an
+  // applied-order number.
+  const applied = new Set<string>(
+    (db.prepare('SELECT name FROM schema_version').all() as { name: string }[]).map((r) => r.name),
+  );
+  const pending = migrations.filter((m) => !applied.has(m.name));
   if (pending.length === 0) return;
 
-  log.info('Running migrations', {
-    from: currentVersion,
-    to: pending[pending.length - 1].version,
-    count: pending.length,
-  });
+  log.info('Running migrations', { count: pending.length });
 
   for (const m of pending) {
     db.transaction(() => {
       m.up(db);
+      const next =
+        (db.prepare('SELECT COALESCE(MAX(version), 0) + 1 AS v FROM schema_version').get() as { v: number }).v;
       db.prepare('INSERT INTO schema_version (version, name, applied) VALUES (?, ?, ?)').run(
-        m.version,
+        next,
         m.name,
         new Date().toISOString(),
       );
     })();
-    log.info('Migration applied', { version: m.version, name: m.name });
+    log.info('Migration applied', { name: m.name });
   }
 }

src/delivery.ts

@@ -21,6 +21,7 @@ import {
 } from './db/sessions.js';
 import { getAgentGroup, createAgentGroup, updateAgentGroup, getAgentGroupByFolder } from './db/agent-groups.js';
 import { createDestination, getDestinationByName, hasDestination, normalizeName } from './db/agent-destinations.js';
+import { getDb, hasTable } from './db/connection.js';
 import { getMessagingGroup, getMessagingGroupByPlatform } from './db/messaging-groups.js';
 import { pickApprovalDelivery, pickApprover } from './access.js';
 import {
@@ -38,18 +39,16 @@ import {
 import { log } from './log.js';
 import { normalizeOptions, type RawOption } from './channels/ask-question.js';
 import {
-  heartbeatPath,
   openInboundDb,
   openOutboundDb,
   sessionDir,
-  inboundDbPath,
   resolveSession,
   writeDestinations,
   writeSessionMessage,
-  writeSystemResponse,
 } from './session-manager.js';
 import { resetContainerIdleTimer, wakeContainer } from './container-runner.js';
 import { initGroupFilesystem } from './group-init.js';
+import { pauseTypingRefreshAfterDelivery, setTypingAdapter } from './modules/typing/index.js';
 import type { OutboundFile } from './channels/adapter.js';
 import type { AgentGroup, Session } from './types.js';
 
@@ -93,6 +92,9 @@ let sweepPolling = false;
 
 export function setDeliveryAdapter(adapter: ChannelDeliveryAdapter): void {
   deliveryAdapter = adapter;
+  // Forward to the typing module so it can fire setTyping on its own
+  // interval. Direct call, not a registry — typing is a default module.
+  setTypingAdapter(adapter);
 }
 
 /**
@@ -194,148 +196,6 @@ async function requestApproval(
   log.info('Approval requested', { action, approvalId, agentName, approver: target.userId });
 }
 
-/** Show typing indicator on a channel. Called when a message is routed to the agent. */
-export async function triggerTyping(channelType: string, platformId: string, threadId: string | null): Promise<void> {
-  try {
-    await deliveryAdapter?.setTyping?.(channelType, platformId, threadId);
-  } catch {
-    // Typing is best-effort — don't fail routing if it errors
-  }
-}
-
-// ── Typing refresh ──
-// Most platforms expire a typing indicator after 5–10s, so a one-shot call
-// on message arrival goes stale long before the agent finishes thinking.
-// We keep it alive by re-firing setTyping on a short interval — but only
-// while the agent is actually WORKING, not just while the container is
-// alive. The agent-runner touches `heartbeat` on every SDK event, so we
-// gate each tick on "is the heartbeat file fresh?". If it goes stale (agent
-// finished its turn and is idle-polling), the refresh stops on its own
-// without waiting for the container to exit.
-//
-// After delivering a user-facing message, the refresh is paused for
-// POST_DELIVERY_PAUSE_MS — long enough for the client-side typing
-// indicator to visually clear (Discord ~10s, Telegram ~5s). If the agent
-// keeps touching heartbeat past the pause window, typing resumes
-// naturally on the next refresh tick.
-//
-// `startTypingRefresh` is idempotent per session. `stopTypingRefresh` is
-// called from container-runner.ts on container exit as a fast-path cleanup
-// (the heartbeat-staleness path would catch it within one tick anyway).
-const TYPING_REFRESH_MS = 4000;
-// Grace window from startTypingRefresh: fire typing unconditionally for
-// this long regardless of heartbeat state. Covers container spawn/wake
-// latency, which can be 5–12s on a cold start before the first heartbeat
-// touch lands.
-const TYPING_GRACE_MS = 15000;
-// After the grace window, a heartbeat must be mtimed within this many
-// milliseconds of now to count as "agent is working." Heartbeats are
-// touched on every SDK event (tool calls, result chunks), so during
-// active work they land every few hundred ms. 6s is well above that
-// while still being small enough to stop typing quickly when the agent
-// goes idle.
-const HEARTBEAT_FRESH_MS = 6000;
-// After we deliver a user-facing message, pause typing for this long so
-// the client-side indicator has time to visually clear. Tuned for the
-// longest common client expiry (Discord ~10s). The interval stays
-// running; ticks inside the pause just skip the setTyping call.
-const POST_DELIVERY_PAUSE_MS = 10000;
-
-interface TypingTarget {
-  agentGroupId: string;
-  channelType: string;
-  platformId: string;
-  threadId: string | null;
-  interval: NodeJS.Timeout;
-  startedAt: number;
-  pausedUntil: number; // epoch ms; 0 = not paused
-}
-
-const typingRefreshers = new Map<string, TypingTarget>();
-
-function isHeartbeatFresh(agentGroupId: string, sessionId: string): boolean {
-  const hbPath = heartbeatPath(agentGroupId, sessionId);
-  try {
-    const stat = fs.statSync(hbPath);
-    return Date.now() - stat.mtimeMs < HEARTBEAT_FRESH_MS;
-  } catch {
-    return false;
-  }
-}
-
-export function startTypingRefresh(
-  sessionId: string,
-  agentGroupId: string,
-  channelType: string,
-  platformId: string,
-  threadId: string | null,
-): void {
-  const existing = typingRefreshers.get(sessionId);
-  if (existing) {
-    // Already refreshing. Fire an immediate tick for the new inbound
-    // event and reset the grace window — the new message restarts the
-    // container-wake latency budget. Also clear any lingering
-    // post-delivery pause: a new inbound means the user expects typing
-    // to show immediately.
-    triggerTyping(channelType, platformId, threadId).catch(() => {});
-    existing.startedAt = Date.now();
-    existing.pausedUntil = 0;
-    return;
-  }
-
-  // Immediate tick + periodic refresh.
-  triggerTyping(channelType, platformId, threadId).catch(() => {});
-  const startedAt = Date.now();
-  const interval = setInterval(() => {
-    const entry = typingRefreshers.get(sessionId);
-    if (!entry) return; // stopped externally since this tick was scheduled
-
-    // Inside a post-delivery pause: skip setTyping but keep the interval
-    // running so we resume automatically once the pause expires.
-    if (entry.pausedUntil > Date.now()) return;
-
-    const withinGrace = Date.now() - entry.startedAt < TYPING_GRACE_MS;
-    if (withinGrace || isHeartbeatFresh(entry.agentGroupId, sessionId)) {
-      triggerTyping(entry.channelType, entry.platformId, entry.threadId).catch(() => {});
-      return;
-    }
-
-    // Out of grace AND heartbeat stale — agent is idle, stop refreshing.
-    clearInterval(entry.interval);
-    typingRefreshers.delete(sessionId);
-  }, TYPING_REFRESH_MS);
-  // unref so a stale refresher can't hold the event loop alive.
-  interval.unref();
-  typingRefreshers.set(sessionId, {
-    agentGroupId,
-    channelType,
-    platformId,
-    threadId,
-    interval,
-    startedAt,
-    pausedUntil: 0,
-  });
-}
-
-/**
- * Pause the typing refresh for POST_DELIVERY_PAUSE_MS. Called after a
- * user-facing message is delivered so the client-side indicator has a
- * chance to visually clear before the agent's next SDK event pushes it
- * back on. No-op if no refresh is active for this session.
- */
-export function pauseTypingRefreshAfterDelivery(sessionId: string): void {
-  const entry = typingRefreshers.get(sessionId);
-  if (!entry) return;
-  entry.pausedUntil = Date.now() + POST_DELIVERY_PAUSE_MS;
-}
-
-export function stopTypingRefresh(sessionId: string): void {
-  const entry = typingRefreshers.get(sessionId);
-  if (!entry) return;
-  clearInterval(entry.interval);
-  typingRefreshers.delete(sessionId);
-}
-
 /** Start the active container poll loop (~1s). */
 export function startActiveDeliveryPoll(): void {
   if (activePolling) return;
@@ -553,15 +413,22 @@ async function deliverMessage(
       throw new Error(`unknown messaging group for ${msg.channel_type}/${msg.platform_id} (message ${msg.id})`);
     }
     const isOriginChat = session.messaging_group_id === mg.id;
-    if (!isOriginChat && !hasDestination(session.agent_group_id, 'channel', mg.id)) {
+    // Guarded: without the agent-to-agent module, `agent_destinations`
+    // doesn't exist and we permit all non-origin channel sends (the
+    // origin-chat case is always allowed regardless).
+    const checkDestinations = hasTable(getDb(), 'agent_destinations');
+    if (!isOriginChat && checkDestinations && !hasDestination(session.agent_group_id, 'channel', mg.id)) {
       throw new Error(
         `unauthorized channel destination: ${session.agent_group_id} cannot send to ${mg.channel_type}/${mg.platform_id}`,
       );
     }
   }
 
-  // Track pending questions for ask_user_question flow
-  if (content.type === 'ask_question' && content.questionId) {
+  // Track pending questions for ask_user_question flow.
+  // Guarded: without the interactive module, `pending_questions` doesn't
+  // exist and we skip persistence — the card still delivers to the user,
+  // but the response path has nowhere to land and will log unclaimed.
+  if (content.type === 'ask_question' && content.questionId && hasTable(getDb(), 'pending_questions')) {
     const title = content.title as string | undefined;
     const rawOptions = content.options as unknown;
     if (!title || !Array.isArray(rawOptions)) {
@@ -637,6 +504,34 @@ async function deliverMessage(
   return platformMsgId;
 }
 
+/**
+ * Delivery action registry.
+ *
+ * Modules register handlers for system-kind outbound message actions via
+ * `registerDeliveryAction`. Core checks the registry first in
+ * `handleSystemAction` and falls through to the inline switch when no
+ * handler is registered. The switch will shrink as modules are extracted
+ * (scheduling, approvals, agent-to-agent) and eventually only its default
+ * branch remains.
+ *
+ * Default when no handler registered and the switch doesn't match: log
+ * "Unknown system action" and return.
+ */
+export type DeliveryActionHandler = (
+  content: Record<string, unknown>,
+  session: Session,
+  inDb: Database.Database,
+) => Promise<void>;
+
+const actionHandlers = new Map<string, DeliveryActionHandler>();
+
+export function registerDeliveryAction(action: string, handler: DeliveryActionHandler): void {
+  if (actionHandlers.has(action)) {
+    log.warn('Delivery action handler overwritten', { action });
+  }
+  actionHandlers.set(action, handler);
+}
+
 /**
  * Handle system actions from the container agent.
  * These are written to messages_out because the container can't write to inbound.db.
@@ -650,6 +545,12 @@ async function handleSystemAction(
   const action = content.action as string;
   log.info('System action from agent', { sessionId: session.id, action });
 
+  const registered = actionHandlers.get(action);
+  if (registered) {
+    await registered(content, session, inDb);
+    return;
+  }
+
   switch (action) {
     case 'schedule_task': {
       const taskId = content.taskId as string;

src/host-sweep.ts

@@ -100,8 +100,14 @@ async function sweepSession(session: Session): Promise<void> {
       detectStaleContainers(inDb, outDb, session, agentGroup.id);
     }
 
-    // 4. Handle recurrence for completed messages
+    // 4. Handle recurrence for completed messages.
+    // MODULE-HOOK:scheduling-recurrence:start
+    // When scheduling is extracted (PR #4), `handleRecurrence` moves to
+    // `src/modules/scheduling/` and the `/add-scheduling` skill replaces
+    // this block with a call to the module. Without scheduling
+    // installed, the block is empty and recurrence is a no-op.
     await handleRecurrence(inDb, session);
+    // MODULE-HOOK:scheduling-recurrence:end
   } finally {
     inDb.close();
     outDb?.close();

src/index.ts

@@ -33,10 +33,55 @@ import { writeSessionMessage } from './session-manager.js';
 import { wakeContainer, buildAgentGroupImage, killContainer } from './container-runner.js';
 import { log } from './log.js';
 
+/**
+ * Response handler registry.
+ *
+ * Button-click / question responses arrive via the channel adapter's
+ * `onAction` callback. Core iterates registered handlers in registration
+ * order; the first one that returns `true` claims the response.
+ * Unclaimed responses fall through to the inline `handleQuestionResponse`
+ * below (which handles OneCLI credential approvals, pending_approvals,
+ * and pending_questions). As those modules are extracted, the inline
+ * function will shrink and the registry will own the full dispatch.
+ */
+export interface ResponsePayload {
+  questionId: string;
+  value: string;
+  userId: string | null;
+  channelType: string;
+  platformId: string;
+  threadId: string | null;
+}
+
+export type ResponseHandler = (payload: ResponsePayload) => Promise<boolean>;
+
+const responseHandlers: ResponseHandler[] = [];
+
+export function registerResponseHandler(handler: ResponseHandler): void {
+  responseHandlers.push(handler);
+}
+
+async function dispatchResponse(payload: ResponsePayload): Promise<void> {
+  for (const handler of responseHandlers) {
+    try {
+      const claimed = await handler(payload);
+      if (claimed) return;
+    } catch (err) {
+      log.error('Response handler threw', { questionId: payload.questionId, err });
+    }
+  }
+  // Unclaimed — fall through to inline handler.
+  await handleQuestionResponse(payload.questionId, payload.value, payload.userId ?? '');
+}
+
 // Channel barrel — each enabled channel self-registers on import.
 // Channel skills uncomment lines in channels/index.ts to enable them.
 import './channels/index.js';
 
+// Modules barrel — default modules (typing, mount-security) ship here; skills
+// append registry-based modules. Imported for side effects (registrations).
+import './modules/index.js';
+
 import type { ChannelAdapter, ChannelSetup, ConversationConfig } from './channels/adapter.js';
 import { initChannelAdapters, teardownChannelAdapters, getChannelAdapter } from './channels/channel-registry.js';
 
@@ -82,7 +127,18 @@ async function main(): Promise<void> {
         });
       },
       onAction(questionId, selectedOption, userId) {
-        handleQuestionResponse(questionId, selectedOption, userId).catch((err) => {
+        dispatchResponse({
+          questionId,
+          value: selectedOption,
+          userId,
+          channelType: adapter.channelType,
+          // platformId/threadId aren't surfaced by the current onAction
+          // signature — the inline fallback looks them up from the
+          // pending_question / pending_approval row. Registered handlers
+          // typically do the same.
+          platformId: '',
+          threadId: null,
+        }).catch((err) => {
           log.error('Failed to handle question response', { questionId, err });
         });
       },

src/modules/index.ts

@@ -0,0 +1,16 @@
+/**
+ * Modules barrel.
+ *
+ * Each module self-registers at import time. This barrel is imported by
+ * src/index.ts for side effects (registry registrations, typing impl setup,
+ * etc.). Core runs with an empty barrel — the registries have inline
+ * fallbacks and `sqlite_master` guards.
+ *
+ * Default modules (ship with main, direct core import):
+ *   - src/modules/typing/        → imported directly by router/delivery/container-runner
+ *   - src/modules/mount-security/ → imported directly by container-runner
+ *
+ * Registry-based modules (installed via /add-<name> skills, pulled from the
+ * `modules` branch): append imports below.
+ */
+export {};

src/modules/mount-security/index.ts

@@ -9,8 +9,8 @@
 import fs from 'fs';
 import os from 'os';
 import path from 'path';
-import { MOUNT_ALLOWLIST_PATH } from './config.js';
-import { log } from './log.js';
+import { MOUNT_ALLOWLIST_PATH } from '../../config.js';
+import { log } from '../../log.js';
 
 export interface AdditionalMount {
   hostPath: string;

src/modules/typing/index.ts

@@ -0,0 +1,165 @@
+/**
+ * Typing indicator refresh — default module.
+ *
+ * Most platforms expire a typing indicator after 5–10s, so a one-shot
+ * call on message arrival goes stale long before the agent finishes
+ * thinking. This module keeps it alive by re-firing `setTyping` on a
+ * short interval — but only while the agent is actually WORKING, gated
+ * on the heartbeat file's mtime after an initial grace period.
+ *
+ * After delivering a user-facing message, the refresh is paused for
+ * POST_DELIVERY_PAUSE_MS so the client-side indicator can visually
+ * clear.
+ *
+ * Default module status:
+ *   - Lives in src/modules/ for signaling (not really core), but ships
+ *     on main and is imported directly by core. No registry, no hook.
+ *   - Removing requires editing src/router.ts, src/delivery.ts, and
+ *     src/container-runner.ts to drop the calls.
+ */
+import fs from 'fs';
+
+import { heartbeatPath } from '../../session-manager.js';
+
+const TYPING_REFRESH_MS = 4000;
+/**
+ * Grace window from startTypingRefresh: fire typing unconditionally
+ * for this long regardless of heartbeat state. Covers container
+ * spawn/wake latency (5–12s on cold start before first heartbeat).
+ */
+const TYPING_GRACE_MS = 15000;
+/**
+ * After the grace window, a heartbeat must be mtimed within this
+ * many ms of now to count as "agent is working." Heartbeats land
+ * every few hundred ms during active work, so 6s is well above
+ * the working floor and small enough to stop typing quickly when
+ * the agent goes idle.
+ */
+const HEARTBEAT_FRESH_MS = 6000;
+/**
+ * After we deliver a user-facing message, pause typing for this
+ * long so the client-side indicator has time to visually clear.
+ * Tuned for the longest common expiry (Discord ~10s). The interval
+ * stays running; ticks inside the pause just skip the setTyping call.
+ */
+const POST_DELIVERY_PAUSE_MS = 10000;
+
+interface TypingAdapter {
+  setTyping?(channelType: string, platformId: string, threadId: string | null): Promise<void>;
+}
+
+interface TypingTarget {
+  agentGroupId: string;
+  channelType: string;
+  platformId: string;
+  threadId: string | null;
+  interval: NodeJS.Timeout;
+  startedAt: number;
+  pausedUntil: number; // epoch ms; 0 = not paused
+}
+
+let adapter: TypingAdapter | null = null;
+const typingRefreshers = new Map<string, TypingTarget>();
+
+/**
+ * Bind the typing module to the channel delivery adapter so it can
+ * call `setTyping`. Called once by `src/delivery.ts` inside
+ * `setDeliveryAdapter`. Passing a fresh adapter replaces the prior
+ * binding and leaves active refreshers in place (they'll use the
+ * new adapter on their next tick).
+ */
+export function setTypingAdapter(a: TypingAdapter): void {
+  adapter = a;
+}
+
+async function triggerTyping(channelType: string, platformId: string, threadId: string | null): Promise<void> {
+  try {
+    await adapter?.setTyping?.(channelType, platformId, threadId);
+  } catch {
+    // Typing is best-effort — don't let it fail delivery or routing.
+  }
+}
+
+function isHeartbeatFresh(agentGroupId: string, sessionId: string): boolean {
+  const hbPath = heartbeatPath(agentGroupId, sessionId);
+  try {
+    const stat = fs.statSync(hbPath);
+    return Date.now() - stat.mtimeMs < HEARTBEAT_FRESH_MS;
+  } catch {
+    return false;
+  }
+}
+
+export function startTypingRefresh(
+  sessionId: string,
+  agentGroupId: string,
+  channelType: string,
+  platformId: string,
+  threadId: string | null,
+): void {
+  const existing = typingRefreshers.get(sessionId);
+  if (existing) {
+    // Already refreshing. Fire an immediate tick for the new inbound
+    // event and reset the grace window — the new message restarts
+    // the container-wake latency budget. Also clear any lingering
+    // post-delivery pause: a new inbound means the user expects
+    // typing to show immediately.
+    triggerTyping(channelType, platformId, threadId).catch(() => {});
+    existing.startedAt = Date.now();
+    existing.pausedUntil = 0;
+    return;
+  }
+
+  // Immediate tick + periodic refresh.
+  triggerTyping(channelType, platformId, threadId).catch(() => {});
+  const startedAt = Date.now();
+  const interval = setInterval(() => {
+    const entry = typingRefreshers.get(sessionId);
+    if (!entry) return; // stopped externally since this tick was scheduled
+
+    // Inside a post-delivery pause: skip setTyping but keep the
+    // interval running so we resume automatically once the pause
+    // expires.
+    if (entry.pausedUntil > Date.now()) return;
+
+    const withinGrace = Date.now() - entry.startedAt < TYPING_GRACE_MS;
+    if (withinGrace || isHeartbeatFresh(entry.agentGroupId, sessionId)) {
+      triggerTyping(entry.channelType, entry.platformId, entry.threadId).catch(() => {});
+      return;
+    }
+
+    // Out of grace AND heartbeat stale — agent is idle, stop refreshing.
+    clearInterval(entry.interval);
+    typingRefreshers.delete(sessionId);
+  }, TYPING_REFRESH_MS);
+  // unref so a stale refresher can't hold the event loop alive.
+  interval.unref();
+  typingRefreshers.set(sessionId, {
+    agentGroupId,
+    channelType,
+    platformId,
+    threadId,
+    interval,
+    startedAt,
+    pausedUntil: 0,
+  });
+}
+
+/**
+ * Pause the typing refresh for POST_DELIVERY_PAUSE_MS. Called after
+ * a user-facing message is delivered so the client-side indicator
+ * has a chance to visually clear before the agent's next SDK event
+ * pushes it back on. No-op if no refresh is active for this session.
+ */
+export function pauseTypingRefreshAfterDelivery(sessionId: string): void {
+  const entry = typingRefreshers.get(sessionId);
+  if (!entry) return;
+  entry.pausedUntil = Date.now() + POST_DELIVERY_PAUSE_MS;
+}
+
+export function stopTypingRefresh(sessionId: string): void {
+  const entry = typingRefreshers.get(sessionId);
+  if (!entry) return;
+  clearInterval(entry.interval);
+  typingRefreshers.delete(sessionId);
+}

src/router.ts

@@ -21,7 +21,7 @@ import { getChannelAdapter } from './channels/channel-registry.js';
 import { isMember } from './db/agent-group-members.js';
 import { getMessagingGroupByPlatform, createMessagingGroup, getMessagingGroupAgents } from './db/messaging-groups.js';
 import { upsertUser, getUser } from './db/users.js';
-import { startTypingRefresh } from './delivery.js';
+import { startTypingRefresh } from './modules/typing/index.js';
 import { log } from './log.js';
 import { resolveSession, writeSessionMessage } from './session-manager.js';
 import { wakeContainer } from './container-runner.js';
@@ -45,6 +45,34 @@ export interface InboundEvent {
   };
 }
 
+/**
+ * Inbound gate registry.
+ *
+ * A module (permissions, today) can register a single gate function that
+ * owns sender resolution + access decision. Without a registered gate,
+ * core falls back to the inline `extractAndUpsertUser` +
+ * `enforceAccess` + `handleUnknownSender` chain.
+ *
+ * Takes the raw event so the gate can read sender fields from
+ * `event.message.content`. Returns either allowed=true with a `userId`
+ * (null if unresolved) or allowed=false with a reason; core drops the
+ * message on refusal.
+ */
+export type InboundGateResult =
+  | { allowed: true; userId: string | null }
+  | { allowed: false; userId: string | null; reason: string };
+
+export type InboundGateFn = (event: InboundEvent, mg: MessagingGroup, agentGroupId: string) => InboundGateResult;
+
+let inboundGate: InboundGateFn | null = null;
+
+export function setInboundGate(fn: InboundGateFn): void {
+  if (inboundGate) {
+    log.warn('Inbound gate overwritten');
+  }
+  inboundGate = fn;
+}
+
 /**
  * Route an inbound message from a channel adapter to the correct session.
  * Creates messaging group + session if they don't exist yet.
@@ -81,13 +109,8 @@ export async function routeInbound(event: InboundEvent): Promise<void> {
     });
   }
 
-  // 2. Resolve sender → user id. Upsert into users table on first sight so
-  //    subsequent messages find an existing row. `userId` is null if the
-  //    adapter didn't give us enough to identify a sender (the gate will
-  //    then apply unknown_sender_policy).
-  const userId = extractAndUpsertUser(event);
-
-  // 3. Resolve agent groups wired to this messaging group
+  // 2. Resolve agent groups wired to this messaging group. (The gate runs
+  //    after this so it can decide based on the target agent group.)
   const agents = getMessagingGroupAgents(mg.id);
   if (agents.length === 0) {
     log.warn('MESSAGE DROPPED — no agent groups wired to this channel. Run setup register step to configure.', {
@@ -128,13 +151,31 @@ export async function routeInbound(event: InboundEvent): Promise<void> {
     return;
   }
 
-  // 4. Access gate. Public channels skip the gate entirely.
-  if (mg.unknown_sender_policy !== 'public') {
-    const gate = enforceAccess(userId, match.agent_group_id);
-    if (!gate.allowed) {
-      handleUnknownSender(mg, userId, match.agent_group_id, gate.reason, event);
+  // 3. Inbound gate: sender resolution + access decision. If a module
+  //    registered a gate, it owns the whole thing (it can upsert users,
+  //    check roles, etc.). Otherwise fall back to the inline chain.
+  let userId: string | null;
+  if (inboundGate) {
+    const result = inboundGate(event, mg, match.agent_group_id);
+    userId = result.userId;
+    if (!result.allowed) {
+      log.info('MESSAGE DROPPED — inbound gate refused', {
+        messagingGroupId: mg.id,
+        agentGroupId: match.agent_group_id,
+        userId,
+        reason: result.reason,
+      });
       return;
     }
+  } else {
+    userId = extractAndUpsertUser(event);
+    if (mg.unknown_sender_policy !== 'public') {
+      const gate = enforceAccess(userId, match.agent_group_id);
+      if (!gate.allowed) {
+        handleUnknownSender(mg, userId, match.agent_group_id, gate.reason, event);
+        return;
+      }
+    }
   }
 
   // 5. Resolve or create session.

src/session-manager.ts

@@ -17,6 +17,7 @@ import path from 'path';
 import { DATA_DIR } from './config.js';
 import { getAgentGroup } from './db/agent-groups.js';
 import { getDestinations } from './db/agent-destinations.js';
+import { getDb, hasTable } from './db/connection.js';
 import { getMessagingGroup } from './db/messaging-groups.js';
 import { createSession, findSession, findSessionByAgentGroup, getSession, updateSession } from './db/sessions.js';
 import {
@@ -183,6 +184,11 @@ export function writeDestinations(agentGroupId: string, sessionId: string): void
   const dbPath = inboundDbPath(agentGroupId, sessionId);
   if (!fs.existsSync(dbPath)) return;
 
+  // Guarded: when the agent-to-agent module isn't installed, the
+  // `agent_destinations` table doesn't exist. Skip silently — core
+  // container spawn continues without projecting destinations.
+  if (!hasTable(getDb(), 'agent_destinations')) return;
+
   const rows = getDestinations(agentGroupId);
   const resolved: DestinationRow[] = [];