When Your Account Gets Blocked, Your System Knows First — AI Agent Health Detection in Practice
How It Started
We use MindThread to manage 21 Threads accounts, auto-publishing 35+ posts daily.
Yesterday, I casually opened the Threads App to check one account and saw:
"Please confirm you're a human"
The account was blocked.
But the system showed no errors — API returned 200, scheduler kept running. The posts just went into a void where nobody could see them.
This is the most dangerous blind spot in automation: broken but looks fine.
The Scale Problem
One account? Just check manually.
But with 21 accounts, it's impossible to open each App daily to verify.
More critically — if an account is blocked for 3 days before you notice, that's 3 days of wasted content and a broken schedule.
What I needed: the system detects it and tells me.
Solution: Container Creation Probe
The Threads API has a useful property:
You can attempt to "create a container" without actually publishing. Normal accounts return a container_id, but blocked accounts return:
``
error_subcode: 2207051
error_user_title: "Operation blocked"
One API call determines if an account can post.
`python
def _check_posting_blocked(token, user_id):
r = requests.post(
f'https://graph.threads.net/v1.0/{user_id}/threads',
data={
'media_type': 'TEXT',
'text': 'health_check_probe',
'access_token': token,
},
timeout=15,
)
data = r.json()
if 'id' in data:
return False, 'ok'
err = data.get('error', {})
if err.get('error_subcode') == 2207051:
return True, 'Operation blocked'
return False, f'other: {err.get("message")}'
`
This code doesn't actually post anything (container created but not published), making it safe for health checks.
Architecture: Background Worker
We wrapped this into a Background Worker that runs alongside the main system:
Frequency: Every 6 hours, scan all accounts
Flow:
1. Load all account configurations
2. Test each account with Container Creation probe
3. Save results to account_health.json (persistent)
4. Compare with previous state to detect new blocks
5. New block detected → instant Telegram notification
Key Design Decisions:
- State change detection: Only notifies on ok → blocked transitions, no repeated alerts
- Persistence: Results survive system restarts
- Rate limiting: 3-second delay between accounts
- Graceful shutdown: Supports stop events
Instant Telegram Alerts
When a block is detected, the system pushes directly to my personal Telegram:
`
⚠️ Account Block Alert
These accounts cannot post:
• rent.calc.tw — Operation blocked
• tw.salary — Operation blocked
Scan time: 2026-03-05 18:30
``
From "didn't know the account was down" to "auto-detected within 6 hours."
Lessons Learned
After scanning all 21 accounts, 2 were blocked. They shared common traits:
| Trait | rent.calc.tw | tw.salary |
|-------|-------------|-----------|
| Created | 3 days ago | 3 days ago |
| Post frequency | 3-4x daily | 3-4x daily |
| Result | Blocked | Blocked |
Conclusion: New account + high frequency = triggers Threads' automation detection.
For new accounts, cut posting frequency in half for the first 7 days. Build trust first, then accelerate.
The Next Layer of Automation
Most people think "automation" means: let the system do things for you.
But mature automation means: let the system take care of itself.
- Content automation ✅ (Scheduler)
- Reply automation ✅ (Reply Monitor)
- Data automation ✅ (Insights Worker)
- Health detection automation ✅ (Added today)
When you manage more and more accounts, manual inspection becomes impossible. The system must self-diagnose and self-report.
This is the real value of AI Agents — not just generating content, but watching the system and preventing problems.
Technical Summary
| Item | Spec |
|------|------|
| Detection | Threads API Container Creation Probe |
| Frequency | Every 6 hours |
| Notification | Telegram Personal Chat |
| Persistence | account_health.json |
| API Endpoints | 4 (status / scan-now / start / stop) |
| Deployment | Background Worker Thread (auto-starts with main system) |
MindThread isn't just a scheduling tool. It's a complete automation infrastructure — from content generation, scheduled publishing, reply management, to account health detection.
Each automation layer pushes you from "manual management" toward "system autonomy."
This is our methodology. Continuously building, continuously sharing.