🔥 [Real Incident Report] Composer 1 fully restored a crashed production EC2 with only one human instruction

Showcase
1

Hi Cursor Team and Community,

This report was supervised by Gemini 2.5 AI in “Strict Mode,” which emphasizes only logical rigor and objective facts.

I am sharing a case where Composer 1 autonomously fixed a critical EC2 production incident in a live AWS environment, almost single-handedly.

This is not a demo, but a record of an actual live incident that occurred during normal business hours.

:police_car_light: Incident Summary: The Root Volume Crisis

The root EBS volume reached 100% utilization, causing the OS to freeze, making both SSH and SSM unstable—a truly fatal situation.

The manual recovery procedure usually involves high-risk steps:

  1. Launch a rescue EC2 instance.

  2. Detach the corrupted root volume.

  3. Attach the volume to the rescue instance, then repair and expand the filesystem.

  4. Re-attach the volume to the original EC2 and reboot it.

:light_bulb: My Instruction Was Just One Line

My initial instruction to Composer was simple:

“Root EBS is full and the EC2 can’t boot. Please restore it safely.”

And later in the recovery process, I gave the key command:

“Just run it already” (This casual instruction was executed by Composer and is visible in the execution log below.)

:brain: Composer 1’s Astonishing Workflow

From this single, natural language instruction, Composer 1 autonomously executed all the necessary steps through a sophisticated workflow:

  • Multi-Stage Planning: It designed a complete plan, from diagnosis through recovery to final verification.

  • Long-Term Context Tracking: It remembered session states (EBS Volume ID, Instance ID, Public IP) and passed them accurately to subsequent commands.

  • Safe CLI Generation: It generated high-risk commands like aws ec2 detach-volume and xfs_growfs while incorporating essential safety measures (e.g., error handling and wait-until-available commands).

  • Environmental Adaptability: It used conditional branching to select the correct OS commands (e.g., xfs_growfs vs. resize2fs) based on the filesystem type.

The Ultimate Proof: EBS Volume Swap (The Most Critical Step)

The screenshot below serves as definitive evidence that Composer successfully completed the high-risk EBS detach/attach operation using accurate AWS CLI, and verified its success:

image
image1682Ă—1326 446 KB

:trophy: Conclusion: This is the Reality of AI-Ops

In my daily work, I rely on AI for tasks ranging from MySQL operations to entire server builds. This isn’t just about convenience; it’s about eliminating the hidden risks of human error and prolonged recovery times.

Composer 1 is not just a code generator; it is operating as an autonomous SRE-level agent.

P.S.:

I was hesitant to share this, but I was strongly encouraged by both Google’s Gemini 2.5 AI and ChatGPT GPT-5.1, who both insisted, “This is an invaluable case study that must be shared.” :joy:

The future of AIOps is here.

— Tomoki Shibahara

AI Systems Architect / Japan @ Independent Research

Focused on AI-driven web service development and human-governed orchestration systems integrating LLMs, FastAPI, and AWS automation.