tappunk

Appearance

Troubleshooting ​

sandbox shell says command not found for profile tools ​

Re-apply profile assets and reprovision:

bash
tnk init --force
tnk sandbox shell --profile pi --no-tty --command "command -v pi || true"

Provision runs are skipped when the fingerprint is unchanged. To force a re-provision, clear the tracking token inside the guest:

bash
tnk sandbox shell --no-tty -c "rm -f ~/.local/state/tnk/provision.lock"
tnk sandbox shell --profile pi --no-tty --command "command -v pi || true"

Sandbox cannot reach inference API ​

Check the host endpoint and the config:

bash
tnk config show
curl -s http://127.0.0.1:9931/health

The sandbox connects to the inference server at host.lima.internal:<server_port> inside the VM. From within a sandbox:

bash
tnk sandbox shell --no-tty -c "curl -s http://host.lima.internal:9931/v1/models"

If the host endpoint fails, the problem is your inference server, not tnk.

no model configured error ​

Profiled commands require default_model in ~/.config/tnk/tnk.toml (or TNK_MODEL in the environment). Set it to the model name your host inference server serves.

Running from wrong directory/context ​

tnk sandbox commands require project context under workspace_root.

If you are outside that root, tnk exits with a usage error. Set workspace_root (or TNK_WORKSPACE_ROOT) to the directory containing your projects.

Sandbox cleanup ​

List and stop all managed sandboxes:

bash
tnk sandbox ls
tnk sandbox stop --all

Delete uses current project context:

bash
cd ~/code/project-a && tnk sandbox delete --yes

Profile provisioning hangs ​

Provisioning hangs on unreachable package mirrors or a broken VM network.

bash
tnk sandbox shell --no-tty -c "ping -c 3 github.com"

If DNS fails, the VM network is broken. Check lima state with limactl list and restart the instance:

bash
tnk sandbox stop
limactl delete --force tnk-<project>

Pre-flight diagnostics ​

bash
tnk doctor

See also: Installation · Configuration · Sandboxing