메뉴
Run a minimal VoidBox VM smoke test locally. Verifies the VM boots and the LLM responds correctly.
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
SOC 직업 분류 기준
Rust coding style guide. Apply automatically when writing or modifying Rust code. Enforces for-loops over iterators, let-else for early returns, variable shadowing, newtypes, explicit matching, and minimal comments.
Run the full quality gate for this repo — format check, clippy, tests, security audit, startup bench regression, and real-workload smoke (HN agent + openclaw gateway) when secrets are available. Invoke before marking any implementation task done or pushing a branch.
Use when investigating performance issues, VM freezes, high CPU usage, or when the user asks to profile a voidbox process. Also invoked as /perf.
Use rust-analyzer's Structural Search and Replace (SSR) to change lots of Rust code. SSR matches by AST structure and semantic meaning, understanding type resolution and path equivalence.
Rust documentation conventions (RFC 1574). Apply when writing doc comments on public Rust items. Covers summary sentences, section headings, type references, and examples.
| name | test-vm |
| description | Run a minimal VoidBox VM smoke test locally. Verifies the VM boots and the LLM responds correctly. |
| disable-model-invocation | true |
Runs examples/smoke/smoke.yaml — the agent replies with "vm-ok 2+2=4". No tool use required; any model that generates text will pass. Success is success: true with output containing vm-ok.
All commands run from the repository root.
Usage: /test-vm [ollama [model] | claude | claude-personal]
ollama [model] — local Ollama (default model: phi4-mini)claude — Anthropic API key (ANTHROPIC_API_KEY must be set)claude-personal — personal Claude account (macOS: extracted from Keychain; Linux: staged from ~/.claude/)ollama phi4-miniRecommended Ollama models (2–4 GB RAM, reliable with claude-code): phi4-mini, gemma3:4b, llama3.2:3b. Avoid *-coder variants — they don't follow the agentic protocol well.
Check for an existing kernel file. macOS needs an uncompressed kernel (vmlinux-*); Linux can use the host kernel (vmlinuz-*):
ls target/vmlinux-arm64 2>/dev/null \
|| ls target/vmlinux-amd64 2>/dev/null \
|| ls /boot/vmlinuz-$(uname -r) 2>/dev/null \
|| echo "MISSING"
If missing, download it (the script caches the result in target/):
scripts/download_kernel.sh
After downloading, run the check again to confirm and record the path — it is needed in Step 4.
ls target/void-box-rootfs.cpio.gz 2>/dev/null || echo "MISSING"
If missing, build it (auto-detects the claude binary on PATH or ~/.local/bin/claude):
scripts/build_claude_rootfs.sh
Run the check again to confirm. If it still fails, tell the user to set CLAUDE_BIN=/path/to/claude and retry.
cargo build --release --bin voidbox
macOS only — Apple Virtualization.framework requires the entitlement:
codesign --force --sign - --entitlements voidbox.entitlements target/release/voidbox
Check Ollama is installed:
which ollama || echo "NOT INSTALLED"
If not installed: macOS → brew install ollama; Linux → curl -fsSL https://ollama.com/install.sh | sh. Stop until installed.
Check Ollama is running and bound to 0.0.0.0 (required so the guest VM can reach it):
curl -sf http://localhost:11434/api/tags > /dev/null && echo "running" || echo "not running"
If not running, start it in the background:
OLLAMA_HOST=0.0.0.0:11434 ollama serve > /tmp/ollama-serve.log 2>&1 &
for i in $(seq 1 10); do curl -sf http://localhost:11434/api/tags > /dev/null && break || sleep 1; done
curl -sf http://localhost:11434/api/tags > /dev/null || { echo "Ollama failed to start:"; cat /tmp/ollama-serve.log; }
If it was already running: warn the user that Ollama must be listening on 0.0.0.0:11434 (not 127.0.0.1) so the guest VM can reach it through the NAT gateway. If unsure, stop Ollama and restart it with OLLAMA_HOST=0.0.0.0:11434 ollama serve.
Check the model is pulled; pull it if not:
ollama list | grep -F "MODEL_NAME" || ollama pull MODEL_NAME
(substitute MODEL_NAME with the actual model, e.g. phi4-mini)
Set the Ollama base URL for use in Step 4:
http://192.168.64.1:11434 (VZ NAT gateway)http://10.0.2.2:11434 (SLIRP gateway)[ -n "$ANTHROPIC_API_KEY" ] && echo "key set" || echo "MISSING — export ANTHROPIC_API_KEY=sk-ant-..."
If not set, stop and ask the user to export it.
Check authentication status using the CLI (exits 0 if logged in, 1 if not):
claude auth status
If not authenticated, stop and tell the user to run claude auth login first, then retry.
No manual credential staging is needed — the claude-personal LLM provider discovers
OAuth credentials automatically (macOS Keychain / Linux ~/.claude/.credentials.json),
stages them to a secure temp directory, and mounts them into the guest VM.
Use the kernel path found in Step 1 and the URL from Step 3 in the commands below.
Ollama:
VOIDBOX_LLM_PROVIDER=ollama \
VOIDBOX_LLM_MODEL=<model> \
VOIDBOX_LLM_BASE_URL=<ollama-url> \
VOID_BOX_KERNEL=<kernel-path> \
VOID_BOX_INITRAMFS=target/void-box-rootfs.cpio.gz \
./target/release/voidbox run --file examples/smoke/smoke.yaml
Claude (API key):
VOID_BOX_KERNEL=<kernel-path> \
VOID_BOX_INITRAMFS=target/void-box-rootfs.cpio.gz \
./target/release/voidbox run --file examples/smoke/smoke.yaml
Claude (personal account):
VOIDBOX_LLM_PROVIDER=claude-personal \
VOID_BOX_KERNEL=<kernel-path> \
VOID_BOX_INITRAMFS=target/void-box-rootfs.cpio.gz \
./target/release/voidbox run --file examples/smoke/smoke.yaml
The run output includes a success: field and an output: section. Check:
success: true — the VM ran the agent without errorvm-ok — the LLM responded correctlyIf success: false or output is missing vm-ok, report the full output for diagnosis.