Instance UNHEALTHY + locked out: SSH key mismatch with no recovery path (reset/stop unsupported)

[makaylatang]

Bug Report

Summary

Instance is UNHEALTHY and completely locked out due to SSH public key mismatch. None of the self-serve recovery options work for this instance type.

Instance Details

• Name: srivatsanlab-kin
• ID: 6jda575vz
• Org: Genome-Scale Diffusion Models for Multimodal Genetic Design
• Machine: gcp.h100x8.sxm.academic-grant
• Status: UNHEALTHY (BUILD: COMPLETED, SHELL: READY)

Error

nvidia@srivatsanlab-kin: Permission denied (publickey).

This occurs on both nvidia@srivatsanlab-kin and nvidia@srivatsanlab-kin-host.

Root Cause

The instance has an old SSH public key in authorized_keys. After running brev login, a new brev.pem is generated locally but the new public key cannot be pushed to the instance because it is already UNHEALTHY — a deadlock.

Steps to Reproduce

1. Instance becomes UNHEALTHY (for any reason)
2. brev login regenerates ~/.brev/brev.pem with a new key pair
3. New public key cannot be pushed to the UNHEALTHY instance
4. All SSH access is permanently lost

Recovery Attempts

Command	Result
brev login --token <token>	Succeeds locally, but new key not pushed to instance
brev refresh	Refreshes local SSH config only
brev reset srivatsanlab-kin	❌ "This instance does not support reset"
brev stop srivatsanlab-kin	❌ "instance does not support stop"
brev exec srivatsanlab-kin ...	❌ Permission denied (publickey)
brev exec srivatsanlab-kin --host ...	❌ Permission denied (publickey)

My Current Public Key

Please add this to /home/nvidia/.ssh/authorized_keys on the instance (ID: 6jda575vz):

ssh-rsa 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 makaylatang23@gmail.com

Feature Request

Even if immediate help isn't possible, this exposes a gap: instances that don't support reset or stop have no self-serve recovery path when SSH keys get out of sync. A brev repair-ssh <instance> command that re-pushes the public key via the backend (bypassing SSH) would prevent this deadlock entirely.

Environment

• OS: macOS Darwin 25.5.0 (arm64)
• brev CLI version: v0.6.331