fix: bound message history queries to prevent memory/token overflow (v0.7.0)

getNewMessages and getMessagesSince were loading unbounded results from
SQLite. Added LIMIT (default 200) via DESC subquery, per-prompt cap
(MAX_MESSAGES_PER_PROMPT=50), and cursor recovery from last bot message
timestamp to prevent full history dump on restart or corrupted state.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Author: ziggy

CHANGELOG.md

@@ -1,5 +1,12 @@
 # Changelog
 
+## v0.7.0 (2026-03-29) — Message history overflow fix
+
+### Fixes
+- **Message history queries now have LIMIT** — `getNewMessages()` and `getMessagesSince()` were loading unbounded results from SQLite. As conversations grew, this caused escalating memory usage and token costs. Both queries now cap at 200 rows (configurable), using a DESC subquery to keep the most recent messages.
+- **Per-prompt message cap** — `MAX_MESSAGES_PER_PROMPT` (default 50, configurable via env var) limits how many messages are bundled into a single agent prompt.
+- **Cursor recovery on restart** — when `lastAgentTimestamp` is missing (new group, corrupted state, or restart), the system now recovers from the last bot message timestamp instead of falling back to empty string (which would send the entire conversation history).
+
 ## v0.6.9 (2026-03-29) — Structured memory + process watchdog
 
 ### New

package-lock.json

@@ -1,6 +1,6 @@
 {
   "name": "ghostclaw",
-  "version": "0.6.9",
+  "version": "0.7.0",
   "description": "Personal AI assistant. Bare metal, Telegram-first, no containers.",
   "type": "module",
   "main": "dist/index.js",

package.json

@@ -66,6 +66,11 @@ export const MAX_CONCURRENT_CONTAINERS = Math.max(
   parseInt(process.env.MAX_CONCURRENT_CONTAINERS || '5', 10) || 5,
 );
 
+export const MAX_MESSAGES_PER_PROMPT = Math.max(
+  1,
+  parseInt(process.env.MAX_MESSAGES_PER_PROMPT || '50', 10) || 50,
+);
+
 function escapeRegex(str: string): string {
   return str.replace(/[.*+?^${}()|[\]\\]/g, '\\$&');
 }

src/config.ts

@@ -302,24 +302,29 @@ export function getNewMessages(
   jids: string[],
   lastTimestamp: string,
   botPrefix: string,
+  limit = 200,
 ): { messages: NewMessage[]; newTimestamp: string } {
   if (jids.length === 0) return { messages: [], newTimestamp: lastTimestamp };
 
   const placeholders = jids.map(() => '?').join(',');
   // Filter bot messages using both the is_bot_message flag AND the content
   // prefix as a backstop for messages written before the migration ran.
+  // Use a subquery to get the most recent N messages, then re-sort chronologically.
   const sql = `
-    SELECT id, chat_jid, sender, sender_name, content, timestamp
-    FROM messages
-    WHERE timestamp > ? AND chat_jid IN (${placeholders})
-      AND is_bot_message = 0 AND content NOT LIKE ?
-      AND content != '' AND content IS NOT NULL
-    ORDER BY timestamp
+    SELECT * FROM (
+      SELECT id, chat_jid, sender, sender_name, content, timestamp
+      FROM messages
+      WHERE timestamp > ? AND chat_jid IN (${placeholders})
+        AND is_bot_message = 0 AND content NOT LIKE ?
+        AND content != '' AND content IS NOT NULL
+      ORDER BY timestamp DESC
+      LIMIT ?
+    ) ORDER BY timestamp
   `;
 
   const rows = db
     .prepare(sql)
-    .all(lastTimestamp, ...jids, `${botPrefix}:%`) as NewMessage[];
+    .all(lastTimestamp, ...jids, `${botPrefix}:%`, limit) as NewMessage[];
 
   let newTimestamp = lastTimestamp;
   for (const row of rows) {
@@ -333,20 +338,36 @@ export function getMessagesSince(
   chatJid: string,
   sinceTimestamp: string,
   botPrefix: string,
+  limit = 200,
 ): NewMessage[] {
   // Filter bot messages using both the is_bot_message flag AND the content
   // prefix as a backstop for messages written before the migration ran.
+  // Use a subquery to get the most recent N messages, then re-sort chronologically.
   const sql = `
-    SELECT id, chat_jid, sender, sender_name, content, timestamp
-    FROM messages
-    WHERE chat_jid = ? AND timestamp > ?
-      AND is_bot_message = 0 AND content NOT LIKE ?
-      AND content != '' AND content IS NOT NULL
-    ORDER BY timestamp
+    SELECT * FROM (
+      SELECT id, chat_jid, sender, sender_name, content, timestamp
+      FROM messages
+      WHERE chat_jid = ? AND timestamp > ?
+        AND is_bot_message = 0 AND content NOT LIKE ?
+        AND content != '' AND content IS NOT NULL
+      ORDER BY timestamp DESC
+      LIMIT ?
+    ) ORDER BY timestamp
   `;
   return db
     .prepare(sql)
-    .all(chatJid, sinceTimestamp, `${botPrefix}:%`) as NewMessage[];
+    .all(chatJid, sinceTimestamp, `${botPrefix}:%`, limit) as NewMessage[];
+}
+
+export function getLastBotMessageTimestamp(chatJid: string): string | null {
+  const sql = `
+    SELECT timestamp FROM messages
+    WHERE chat_jid = ? AND is_bot_message = 1
+    ORDER BY timestamp DESC
+    LIMIT 1
+  `;
+  const row = db.prepare(sql).get(chatJid) as { timestamp: string } | undefined;
+  return row?.timestamp ?? null;
 }
 
 export function createTask(

src/db.ts

@@ -10,6 +10,7 @@ import {
   DATA_DIR,
   IDLE_TIMEOUT,
   MAIN_GROUP_FOLDER,
+  MAX_MESSAGES_PER_PROMPT,
   POLL_INTERVAL,
   TELEGRAM_BOT_TOKEN,
   TELEGRAM_ONLY,
@@ -30,6 +31,7 @@ import {
   getAllRegisteredGroups,
   getAllSessions,
   getAllTasks,
+  getLastBotMessageTimestamp,
   getMessagesSince,
   getNewMessages,
   getRouterState,
@@ -151,11 +153,13 @@ async function processGroupMessages(chatJid: string): Promise<boolean> {
 
   const isMainGroup = group.folder === MAIN_GROUP_FOLDER;
 
-  const sinceTimestamp = lastAgentTimestamp[chatJid] || '';
+  const sinceTimestamp =
+    lastAgentTimestamp[chatJid] || getLastBotMessageTimestamp(chatJid) || '';
   const missedMessages = getMessagesSince(
     chatJid,
     sinceTimestamp,
     ASSISTANT_NAME,
+    MAX_MESSAGES_PER_PROMPT,
   );
 
   if (missedMessages.length === 0) return true;
@@ -440,8 +444,11 @@ async function startMessageLoop(): Promise<void> {
 
           const allPending = getMessagesSince(
             chatJid,
-            lastAgentTimestamp[chatJid] || '',
+            lastAgentTimestamp[chatJid] ||
+              getLastBotMessageTimestamp(chatJid) ||
+              '',
             ASSISTANT_NAME,
+            MAX_MESSAGES_PER_PROMPT,
           );
           const messagesToSend =
             allPending.length > 0 ? allPending : groupMessages;
@@ -474,8 +481,14 @@ async function startMessageLoop(): Promise<void> {
 
 function recoverPendingMessages(): void {
   for (const [chatJid, group] of Object.entries(registeredGroups)) {
-    const sinceTimestamp = lastAgentTimestamp[chatJid] || '';
-    const pending = getMessagesSince(chatJid, sinceTimestamp, ASSISTANT_NAME);
+    const sinceTimestamp =
+      lastAgentTimestamp[chatJid] || getLastBotMessageTimestamp(chatJid) || '';
+    const pending = getMessagesSince(
+      chatJid,
+      sinceTimestamp,
+      ASSISTANT_NAME,
+      MAX_MESSAGES_PER_PROMPT,
+    );
     if (pending.length > 0) {
       logger.info(
         { group: group.name, pendingCount: pending.length },