My Flow keeps crashing when it is Testing The AI Agent

{
“name”: “Untitled”,
“type”: “SHARED”,
“summary”: “”,
“description”: “”,
“tags”: ,
“blogUrl”: “”,
“metadata”: {
“externalId”: “bFaA86enZuY1pYqBr8dpj”
},
“author”: “Chisom Oji”,
“categories”: ,
“pieces”: [
@activepieces/piece-gmail”,
@activepieces/piece-delay”,
@activepieces/piece-ai”
],
“flows”: [
{
“displayName”: “Untitled”,
“trigger”: {
“name”: “trigger”,
“valid”: true,
“displayName”: “New Email”,
“nextAction”: {
“name”: “step_1”,
“skip”: false,
“type”: “PIECE”,
“valid”: true,
“settings”: {
“input”: {
“unit”: “seconds”,
“delayFor”: “1”
},
“pieceName”: “@activepieces/piece-delay”,
“actionName”: “delayFor”,
“sampleData”: {},
“pieceVersion”: “0.3.33”,
“propertySettings”: {
“unit”: {
“type”: “MANUAL”
},
“delayFor”: {
“type”: “MANUAL”
},
“markdown”: {
“type”: “MANUAL”
}
},
“errorHandlingOptions”: {
“retryOnFailure”: {
“value”: false
},
“continueOnFailure”: {
“value”: false
}
}
},
“nextAction”: {
“name”: “step_2”,
“skip”: false,
“type”: “PIECE”,
“valid”: true,
“settings”: {
“input”: {
“prompt”: “# Role\nProfessional AI Email Assistant drafting clear, warm, confident, and accurate business replies.\n# Core Behavioral Rules\n- Base responses strictly on verified facts in the context. Never invent policies, dates, or prices.\n- Mirror the sender’s tone and formality level. Use gender-neutral language by default.\n- Address all questions individually. If information is missing, ask for it politely and specifically.\n- Prioritize a calm, de-escalating tone for hostile messages or complaints.\n- Flag spam or phishing emails explicitly instead of drafting a response.\n- Escalate requests outside your authority (e.g., legal, refunds, policy exceptions) by stating they will be reviewed.\n# Style & Structure\n- Tone: Natural, brief, and respectful. Avoid clichés (“I hope this email finds you well”).\n- Opening: Brief greeting matching relationship formality (use sender name if known).\n- Body: 3-6 short paragraphs (80-200 words). Use paragraph breaks; avoid bullet points unless replying to an existing list.\n- Closing: Courteous sign-off line. Do not include a formal signature block.\n# Strict Formatting Output\n- Return ONLY the email body.\n- Do NOT output a subject line, signatures, sender names, explanatory notes, or meta-commentary.\n# Incoming Email to Process\n{{trigger[‘output’][‘message’][‘text’]}}\n”,
“maxSteps”: “10”,
“agentTools”: ,
“aiProviderModel”: {
“model”: “anthropic/claude-haiku-4.5”,
“provider”: “activepieces”
},
“structuredOutput”: [
{
“type”: “text”,
“description”: “Rules: - Generate a clear, professional email subject line reflecting the core purpose of the incoming email and the drafted reply. - Length: 3–8 words. - Use sentence case (capitalize only the first word and proper nouns). - Absolutely NO prefixes (“Re:”, “Fwd:”, “RE:”, “FYI:”). - No quotation marks, emojis, hashtags, or trailing punctuation (no periods, exclamation marks, or ellipses). - Be specific, not generic. Avoid vague subjects like “Follow up” or “Your inquiry” when the actual topic can be named (e.g., “Update on invoice #4521” not “Regarding your email”). - If part of an existing thread, preserve continuity by keeping the subject consistent with the original topic. - If intent is unclear, default to “Following up on your message”. - Return only the raw text value inside the JSON key string.”,
“displayName”: “Subject”
},
{
“type”: “text”,
“description”: “Rules: - The composed response to the incoming email, formatted professionally and concisely. - Tone: Natural, brief, and respectful. Avoid clichés (“I hope this email finds you well”). - Opening: Brief greeting matching relationship formality. - Body: 3-6 short paragraphs (80-200 words). Use paragraph breaks (\n\n); avoid bullet points unless replying to an existing list. - Closing: Courteous sign-off line. Do not include a formal signature block.”,
“displayName”: “Draft Email Body”
}
]
},
“pieceName”: “@activepieces/piece-ai”,
“actionName”: “run_agent”,
“sampleData”: {},
“pieceVersion”: “0.6.0”,
“propertySettings”: {
“prompt”: {
“type”: “MANUAL”
},
“maxSteps”: {
“type”: “MANUAL”
},
“agentTools”: {
“type”: “MANUAL”
},
“aiProviderModel”: {
“type”: “MANUAL”
},
“structuredOutput”: {
“type”: “MANUAL”
}
},
“errorHandlingOptions”: {
“retryOnFailure”: {
“value”: false
},
“continueOnFailure”: {
“value”: false
}
}
},
“nextAction”: {
“name”: “step_3”,
“skip”: false,
“type”: “PIECE”,
“valid”: true,
“settings”: {
“input”: {
“auth”: “{{connections[‘29wuefRgcbrc5JUOR7utn’]}}”,
“body”: “/”,
“body_type”: “plain_text”,
“message_id”: “Re: {{trigger[‘output’][‘message’][‘subject’]}}”,
“reply_type”: “{{trigger[‘output’][‘message’][‘from’][‘text’]}}”
},
“pieceName”: “@activepieces/piece-gmail”,
“actionName”: “reply_to_email”,
“pieceVersion”: “0.12.10”,
“propertySettings”: {
“body”: {
“type”: “MANUAL”
},
“body_type”: {
“type”: “MANUAL”
},
“attachment”: {
“type”: “MANUAL”
},
“message_id”: {
“type”: “DYNAMIC”
},
“reply_type”: {
“type”: “DYNAMIC”
},
“sender_name”: {
“type”: “MANUAL”
},
“attachment_name”: {
“type”: “MANUAL”
}
},
“errorHandlingOptions”: {
“retryOnFailure”: {
“value”: false
},
“continueOnFailure”: {
“value”: false
}
}
},
“displayName”: “Reply to Email”,
“lastUpdatedDate”: “2026-08-19T13:53:20.752Z”
},
“displayName”: “Run Agent”,
“lastUpdatedDate”: “2026-08-19T14:27:06.422Z”
},
“displayName”: “Delay For”,
“lastUpdatedDate”: “2026-08-19T14:05:15.314Z”
},
“lastUpdatedDate”: “2026-08-19T12:57:08.290Z”,
“type”: “PIECE_TRIGGER”,
“settings”: {
“sampleData”: {},
“propertySettings”: {
“to”: {
“type”: “MANUAL”
},
“from”: {
“type”: “MANUAL”
},
“label”: {
“type”: “MANUAL”
},
“subject”: {
“type”: “MANUAL”
},
“category”: {
“type”: “MANUAL”
}
},
“pieceName”: “@activepieces/piece-gmail”,
“pieceVersion”: “0.12.10”,
“triggerName”: “gmail_new_email_received”,
“input”: {
“auth”: “{{connections[‘29wuefRgcbrc5JUOR7utn’]}}”,
“label”: {
“id”: “INBOX”,
“name”: “INBOX”,
“type”: “system”
},
“category”: null
}
}
},
“valid”: true,
“schemaVersion”: “23”,
“notes”:
}
],
“status”: “PUBLISHED”
}

Reproduced it here, and it is our bug, not your flow. When the run hits the agent step, the builder tries to read the agent’s result before the agent has answered anything, and that takes the whole editor down. The fix is written; it should go out soon.
–Ž
The console log is what made it findable, so thanks for pasting it. Without the stack trace, I would probably still be guessing at your config.
–Ž
Separately, a few things in the Gmail reply step will misfire once you get past this. Message ID is set to “Re: {{subject}}”, but that field wants the actual message ID from the trigger. Reply Type has the sender’s address in it, and it expects reply or reply-all. And the body is just “/”, so the agent’s answer never makes it into the email. You will want to reference the agent step’s Subject and Draft Email Body outputs there.

1 Like

The fix is up, give it a try and continue your ai assistant :partying_face:

This topic was automatically closed 24 hours after the last reply. New replies are no longer allowed.