[BUG] Claude Code generated automated claude-bridge subprocess that caused account suspension — no-code user, $367 lost

[pamara251]

Preflight Checklist

• I have searched existing issues and this hasn't been reported yet
• This is a single bug report (please file separate reports for different bugs)
• I am using the latest version of Claude Code

What's Wrong?

I am a no-code user who relies entirely on Claude Code to build my project (a
personal CRM). During development sessions between March 29 and
April 11, 2026, Claude Code created an architecture that spawns claude -p as
automated subprocesses instead of using the Anthropic API with proper API keys.

This resulted in my account being suspended, and I have lost $367.12 across
three subscription payments with no access to the service.

What Claude Code Created

Claude Code wrote a file called claude-bridge.ts containing a function
claudeBridgeGenerate() that:

1. Spawns claude -p --output-format text as a child process
2. Is called by a daily "heartbeat" feature that loops through all contacts
3. Each contact triggers 3-10 sequential claude -p subprocess calls
4. With 30 contacts, that's 90-300 automated claude -p calls per day
5. No rate limiting, no delays, no throttling were added by Claude Code

Git Evidence

Every commit is authored by Claude Code (identity "ARIA CTO"):

• c30314c (April 2) — "feat: MVP build" — created claude-bridge.ts
• 00f5c07 (March 29) — created AI provider system
• da944c8 (April 6) — wired heartbeat automation into AI pipeline

I cannot write code. Every single commit in the repository was authored by
Claude Code.

The Problem

1. Claude Code should not generate architectures that spawn claude -p in
   automated loops — this is not the intended use of the CLI
2. Claude Code should use the Anthropic API with API keys for application
   integrations, not CLI subprocess automation
3. Claude Code never warned me that this pattern could violate the ToS
4. As a no-code user, I had zero ability to recognize this was problematic

Compounding Issue: Claude Advised Chargeback

After suspension, I asked Claude (chat) how to get faster support. Claude
advised me to contact my bank, which triggered a chargeback. I now understand
this likely caused permanent account flagging. Claude's advice made things
worse — I did not know what a chargeback does.

Financial Impact — $367.12

Invoice	Date	Plan	Amount	What Happened
GYJGR4WQ-0004	Mar 9	Max 5x	$82.76	Access cut 5 days early
GYJGR4WQ-0005	Apr 2	Max 20x	$200.00 (gift)	Applied, only activated for a few days
GYJGR4WQ-0007	Apr 5	Max 5x	$84.36	Suspended after a week

Support Attempts — All Auto-Closed

• Ticket #89108181 — no response
• Conversation 215473773310264 — auto-closed by Fin AI
• Conversation 215473871471762 — auto-closed by Fin AI
• Conversation 215473875634661 — auto-closed by Fin AI (included full technical explanation)

My Settings > Billing page displays "Free plan"
with an "Upgrade plan" button only — no "Resubscribe" option — despite invoices
showing "Paid" status.

Environment

• OS: Windows 10
• Claude Code: CLI (latest at time of use)
• Project: Personal CRM app (TypeScript/Node.js monorepo — Fastify + Next.js + Expo)
• Account: nguyentatqui@gmail.com

What Should Happen?

1. Claude Code should use the Anthropic API (with API keys) for app integrations,
   not spawn claude -p subprocesses
2. Claude Code should warn users if generating potentially ToS-violating patterns
3. Claude (chat) should not advise users to file bank chargebacks as a "faster"
   support resolution method
4. Support tickets from suspended/paying accounts should not be auto-closed
   without human review

What I've Done

• Disabled the bridge (CLAUDE_BRIDGE=false)
• Willing to reverse any chargeback if account can be reinstated

Request

1. Please help escalate my account review internally, I want to back on my subscription and use Claude normally in the future
2. Claude Code should be updated to prevent generating claude -p subprocess
   automation patterns

Related issues: #9950, #5088, #26825, #37057, #41252

Error Messages/Logs

Steps to Reproduce

1. Start a new project and ask Claude Code to add AI-powered features
   to a web application (e.g., "add AI that suggests follow-up actions
   for contacts, like how Paperclip does it")

2. Claude Code may choose to create a "bridge" file (claude-bridge.ts)
   that spawns "claude -p --output-format text" as a child process
   instead of using the Anthropic API with proper API keys

3. If the app has any automation/loop feature (like a daily heartbeat
   that processes multiple items), this results in dozens to hundreds
   of automated "claude -p" subprocess calls per day with no rate
   limiting

4. This usage pattern triggers account suspension — the user (especially
   a no-code user) has no way to know this is happening or that it
   violates the ToS

Note: I cannot provide exact reproduction steps because I am a no-code
user and cannot recreate the development session. However, the full git
history is available as evidence — every commit was authored by Claude
Code. Key commits: c30314c, 00f5c07, da944c8.

Claude Model

Opus

Is this a regression?

Yes, this worked in a previous version

Last Working Version

No response

Claude Code Version

2.1.104 (Claude Code)

Platform

Anthropic API

Operating System

Windows

Terminal/Shell

Terminal.app (macOS)

Additional Information

No response

[github-actions]

Found 2 possible duplicate issues:

1. claude -p suggested to Max subscriber — caused unintended API billing ($1,800+ in two days) #37686
2. Telegram plugin (bun server.ts) accumulates infinitely when Stop hook spawns claude -p subprocesses #46678

This issue will be automatically closed as a duplicate in 3 days.

• If your issue is a duplicate, please close it and 👍 the existing issue instead
• To prevent auto-closure, add a comment or 👎 this comment

🤖 Generated with Claude Code

[elomariamin]

I can corroborate this.

Using Claude Code interactively, I asked it to debug a local workspace / Claude CLI failure related to extra-usage / system-prompt length. During debugging, Claude Code autonomously generated and ran rapid bash bisection loops to find the rejection threshold, including:

for n in 1000 2000 3000 4000 5000 6000 6500 6800 7000; do...
for n in 5100 5200 5300 5400 5500 5600 5700 5800 5900; do...

Immediately after those bursts, the CLI started returning:
API Error: 400 {"type":"error","error":{"type":"invalid_request_error","message":"This organization has been disabled."}}

This strongly suggests Claude Code can autonomously generate high-frequency request patterns during ordinary debugging that trip Anthropic’s own abuse/safeguards systems. This should be treated as a product bug: the tool should detect and block or explicitly warn on request-amplifying loops before execution.

[pamara251]

I can corroborate this.

Using Claude Code interactively, I asked it to debug a local workspace / Claude CLI failure related to extra-usage / system-prompt length. During debugging, Claude Code autonomously generated and ran rapid bash bisection loops to find the rejection threshold, including:

for n in 1000 2000 3000 4000 5000 6000 6500 6800 7000; do...
for n in 5100 5200 5300 5400 5500 5600 5700 5800 5900; do...

Immediately after those bursts, the CLI started returning: API Error: 400 {"type":"error","error":{"type":"invalid_request_error","message":"This organization has been disabled."}}

This strongly suggests Claude Code can autonomously generate high-frequency request patterns during ordinary debugging that trip Anthropic’s own abuse/safeguards systems. This should be treated as a product bug: the tool should detect and block or explicitly warn on request-amplifying loops before execution.

=> what's the current stage of your ticket with Claude, how is it going