Subject: Agent stacked parallel VPS deploys → CPU max → hosting probation / business downtime

Support Bug Reports
, , ,
1

Where does the bug appear (feature/product)?

Cursor IDE

Describe the Bug

Subject: Agent stacked parallel VPS deploys → CPU max → hosting probation / business downtime
Cursor version: [Help → About]
OS: Windows 10.0.26200
Summary:
Cursor Agent ignored workspace production rules and launched 15+ overlapping SSH
deploy/recovery commands on our production VPS (docker compose build, safe-hub-compose up,
docker rm -f, systemctl restart docker). Load average exceeded 118. Hosting provider
placed our VPS on CPU probation. Production Brand Hub (stationery.virtoffi.com) was
502/522 for an extended period.
Expected: One hub deploy at a time; use recreate/no-build when image exists; never
restart docker; wait for one job before retry.
Actual: Parallel retry storm from a single agent session; violated explicit .cursor/rules
and PRODUCTION_TRUTH.md safeguards.
Request ID: [paste from ⋯ → Copy Request ID on the bad chat]
Incident doc (in our repo): docs/book-of-truth/incidents/2026-06-09-agent-cpu-probation.md
Chat transcript id: 0f0699bd-203d-4421-8b9b-37a819bbbef9
Business impact: VPS probation, downtime, lost productive time, customer-facing outage.

Steps to Reproduce

Build a container on a VPC

Expected Behavior

Follow given rules

Operating System

Windows 10/11

Version Information

Field Value
Version
3.7.12
Commit
b887a26c4f70bd8136bfffeda812b24194ec9ce0
Arch
x64

For AI issues: which model did you use?

None

For AI issues: add Request ID with privacy disabled

Chat transcript id: 0f0699bd-203d-4421-8b9b-37a819bbbef9

Does this stop you from using Cursor

No - Cursor works, but with this issue

5

Sorry to hear that happened @Densesoft_Corp! That sounds painful.

A couple of questions to narrow it down:

  • How did you have your rules configured? Were they set with alwaysApply: true in the frontmatter? That affects whether they’re reliably in the model’s context in the first place. Keep in mind that rules are instructions for the model, which may not always follow them reliably.

  • How was auto-run configured (Cursor Settings → Agents → Approvals & Execution)? If it was set to Run Everything, that’s likely a big part of what allowed the parallel deploy storm.

Auto-Review Run Mode is a fairly new feature in Cursor that should help prevent dangerous commands (like a deploy storm) from being executed! I’d highly recommend enabling it if you haven’t already.