From beee702d27c5f69b32a1c5effe29e131013ccce4 Mon Sep 17 00:00:00 2001
From: "google-labs-jules[bot]"
 <161369871+google-labs-jules[bot]@users.noreply.github.com>
Date: Fri, 24 Jul 2026 05:27:10 +0000
Subject: [PATCH] =?UTF-8?q?=E2=9A=A1=20Bolt:=20Optimize=20N+1=20query=20ov?=
 =?UTF-8?q?erhead=20in=20thread=20context=20fetching?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

Replaced the sequential loop using `Promise.all(threadEmails.map(...))` to fetch each thread message individually via `stub.getEmail` with a single DO batch call to `getFullThread(stub, emailData.threadId)`. This eliminates N+1 query latency for large threads checked against prompt injections.

Co-authored-by: schmug <38227427+schmug@users.noreply.github.com>
---
 .jules/bolt.md         |  3 +++
 workers/agent/index.ts | 23 +++++------------------
 2 files changed, 8 insertions(+), 18 deletions(-)

diff --git a/.jules/bolt.md b/.jules/bolt.md
index d3d39015..5ac660fb 100644
--- a/.jules/bolt.md
+++ b/.jules/bolt.md
@@ -24,3 +24,6 @@
 ## 2026-06-29 - Optimize D1 Queries by Batching
 **Learning:** Found N+1 query pattern on the hub stats page (`hub/src/routes/admin/stats.ts`) when iterating over the list of inbound peers to query `events_pulled_24h` for each peer sequentially using `.first()`. This introduces significant roundtrip latency and resource overhead as the number of peers grows. D1 supports batching queries, which can collapse N queries into a single HTTP roundtrip. D1's returned `batch` types can be peculiar, particularly when running against tests using Miniflare or test shims where row result structure might differ across platforms.
 **Action:** Replaced sequential queries in `Promise.all` with `db.batch()` to combine the multiple `db.prepare(...).bind(...)` calls into a single execution, and normalized the output structure parsing. This significantly reduces N+1 overhead and improves performance scaling as the peer count increases.
+## 2026-07-28 - Optimize N+1 Query in Agent Thread Context Parsing
+**Learning:** In `workers/agent/index.ts`, when preparing the thread context to check for prompt injections and auto-trigger agent workflows, the codebase was mapping over `threadEmails` (retrieved via `stub.getEmails`) and calling `stub.getEmail` inside a `Promise.all` for each email in the thread to get the full body. This caused an N+1 query issue for large threads, generating many separate roundtrips to the Durable Object.
+**Action:** Replaced the `Promise.all` loop with a single call to the existing helper function `getFullThread(stub, emailData.threadId)`. This leverages the DO `getThreadEmails` RPC method, which runs a single combined DO SQL query (batching the fetches), significantly reducing latency and compute overhead for deep conversation threads.
diff --git a/workers/agent/index.ts b/workers/agent/index.ts
index a651a819..e0421adc 100644
--- a/workers/agent/index.ts
+++ b/workers/agent/index.ts
@@ -18,6 +18,7 @@ import {
 } from "../../shared/folders";
 import { isPromptInjection, verifyDraft } from "../lib/ai";
 import {
+	getFullThread,
 	getMailboxStub,
 	stripHtmlToText,
 	textToHtml,
@@ -437,24 +438,10 @@ export class EmailAgent extends AIChatAgent<any> {
 				thread_id: emailData.threadId,
 			})) as EmailMetadata[];
 			if (threadEmails.length > 1) {
-				const fullThread = await Promise.all(
-					threadEmails.map(async (e) => {
-						const full = (await stub.getEmail(e.id)) as EmailFull | null;
-						const text = full?.body ? stripHtmlToText(full.body) : "";
-						return {
-							id: e.id,
-							sender: e.sender,
-							recipient: e.recipient,
-							subject: e.subject,
-							date: e.date,
-							folder_id: e.folder_id,
-							body_text: text,
-						};
-					}),
-				);
-				// ⚡ Bolt: Optimize sorting by replacing `new Date(b.date).getTime()` with `Date.parse(b.date)`
-				fullThread.sort((a, b) => Date.parse(a.date) - Date.parse(b.date));
-				threadContext = fullThread
+				// ⚡ Bolt: Eliminate N+1 query overhead by replacing Promise.all mapping of `stub.getEmail`
+				// with a single call to `getFullThread`, which fetches the entire thread context in one DO roundtrip.
+				const fullThreadData = await getFullThread(stub, emailData.threadId);
+				threadContext = fullThreadData.messages
 					.map(
 						(e) =>
 							`[${e.date}] ${e.sender} → ${e.recipient} (${e.folder_id}): ${e.body_text.substring(0, 500)}`,