diff --git a/src/app/slack/webhook/route.ts b/src/app/slack/webhook/route.ts
index 0bb05087d..74f223986 100644
--- a/src/app/slack/webhook/route.ts
+++ b/src/app/slack/webhook/route.ts
@@ -36,6 +36,44 @@ export const maxDuration = 800;
 const PROCESSING_REACTION = 'hourglass_flowing_sand';
 const COMPLETE_REACTION = 'white_check_mark';
 
+/**
+ * Split a mrkdwn string into Slack section blocks, each within the character limit.
+ * Splits on newline boundaries to avoid breaking mid-line/mid-URL.
+ */
+function toSectionBlocks(
+  text: string,
+  charLimit: number
+): Array<{ type: 'section'; text: { type: 'mrkdwn'; text: string } }> {
+  if (text.length <= charLimit) {
+    return [{ type: 'section', text: { type: 'mrkdwn', text } }];
+  }
+
+  const blocks: Array<{ type: 'section'; text: { type: 'mrkdwn'; text: string } }> = [];
+  let remaining = text;
+
+  while (remaining.length > 0) {
+    if (remaining.length <= charLimit) {
+      blocks.push({ type: 'section', text: { type: 'mrkdwn', text: remaining } });
+      break;
+    }
+
+    // Find the last newline within the limit to avoid splitting mid-line
+    let splitAt = remaining.lastIndexOf('\n', charLimit);
+    if (splitAt <= 0) {
+      // No newline found; hard-split at the limit as a fallback
+      splitAt = charLimit;
+    }
+
+    blocks.push({
+      type: 'section',
+      text: { type: 'mrkdwn', text: remaining.slice(0, splitAt) },
+    });
+    remaining = remaining.slice(splitAt + 1);
+  }
+
+  return blocks;
+}
+
 /**
  * Post an ephemeral message to the user with a button to view the cloud agent session
  */
@@ -215,20 +253,19 @@ async function processSlackMessage(event: AppMentionEvent | GenericMessageEvent,
   const responseWithDevInfo = result.response + getDevUserSuffix();
   const slackFormattedMessage = markdownToSlackMrkdwn(responseWithDevInfo);
 
+  // Slack section blocks have a 3000-character limit for the text field.
+  // When the response exceeds this (common for cloud agent results that include
+  // PR/MR URLs near the end), the block gets silently truncated, cutting off the URL.
+  // For long messages, split into multiple section blocks to preserve the full content.
+  const SLACK_SECTION_CHAR_LIMIT = 3000;
+  const blocks = toSectionBlocks(slackFormattedMessage, SLACK_SECTION_CHAR_LIMIT);
+
   // Post the response in the thread
   const slackResponse = await postSlackMessageByAccessToken(accessToken, {
     channel,
     text: slackFormattedMessage,
     thread_ts: replyThreadTs,
-    blocks: [
-      {
-        type: 'section',
-        text: {
-          type: 'mrkdwn',
-          text: slackFormattedMessage,
-        },
-      },
-    ],
+    blocks,
   });
 
   console.log(
diff --git a/src/lib/bot/run.ts b/src/lib/bot/run.ts
index d45854af5..b7d200c13 100644
--- a/src/lib/bot/run.ts
+++ b/src/lib/bot/run.ts
@@ -90,6 +90,9 @@ ${formatGitLabRepositoriesForPrompt(gitlabContext)}
 
 Treat this context as authoritative. Prefer selecting a repo from the provided repository list. If the user requests work on a repo that isn't in the list, ask them to confirm the exact owner/repo (or group/project for GitLab) and ensure it's accessible to the integration. Never invent repository names.
 
+## Sharing results
+- After calling a tool in "code" mode, check the result for a PR/MR URL and share it with the user — this is the most important output. Always include the full PR/MR URL in your response so the user can click it directly.
+
 ## Accuracy & safety
 - Don't claim you ran tools, changed code, or created a PR/MR unless the tool results confirm it.
 - Don't fabricate links (including PR/MR URLs).
diff --git a/src/lib/slack-bot.ts b/src/lib/slack-bot.ts
index 841adfea6..f8ad9f0aa 100644
--- a/src/lib/slack-bot.ts
+++ b/src/lib/slack-bot.ts
@@ -113,6 +113,9 @@ Your prompt to the agent should usually include:
 - any constraints (keep changes minimal, follow existing patterns, etc.)
 - a request to open a PR (GitHub) or MR (GitLab) and return the URL
 
+### After the tool returns
+When the tool returns, check the result for a PR/MR URL and share it with the user — this is the most important output. Always include the full PR/MR URL in your response so the user can click it directly.
+
 ## Accuracy & safety
 - Don't claim you ran tools, changed code, or created a PR/MR unless the tool results confirm it.
 - Don't fabricate links (including PR/MR URLs).
@@ -126,7 +129,7 @@ const SPAWN_CLOUD_AGENT_TOOL: OpenAI.Chat.Completions.ChatCompletionTool = {
   function: {
     name: 'spawn_cloud_agent',
     description:
-      'Spawn a Cloud Agent session to perform coding tasks on a GitHub repository or GitLab project. Provide exactly one of githubRepo or gitlabProject.',
+      'Spawn a Cloud Agent session to perform coding tasks on a GitHub repository or GitLab project. Provide exactly one of githubRepo or gitlabProject. After the tool returns, if mode was "code", check the result for a PR/MR URL and share it with the user — this is the most important output.',
     parameters: {
       type: 'object',
       properties: {
@@ -281,10 +284,22 @@ async function spawnCloudAgentSession(
     kilocodeOrganizationId = owner.id;
   }
 
+  const mode = args.mode || 'code';
+  const isGitLab = !!args.gitlabProject;
+
+  // For "code" mode, explicitly instruct the agent to open a PR/MR and return the URL
+  const promptWithPrInstruction =
+    mode === 'code'
+      ? args.prompt +
+        (isGitLab
+          ? '\n\nOpen a merge request with your changes and return the MR URL.'
+          : '\n\nOpen a pull request with your changes and return the PR URL.')
+      : args.prompt;
+
   // Append PR/MR signature to the prompt if we have requester info
   const promptWithSignature = requesterInfo
-    ? args.prompt + buildPrSignature(requesterInfo)
-    : args.prompt;
+    ? promptWithPrInstruction + buildPrSignature(requesterInfo)
+    : promptWithPrInstruction;
 
   // Build platform-specific prepareInput and initiateInput
   let prepareInput: PrepareSessionInput;