Ci fixes

.github/workflows/ci.yml

@@ -11,7 +11,7 @@ jobs:
   unit-tests:
     runs-on: ubuntu-latest
     steps:
-      - uses: actions/checkout@v4
+      - uses: actions/checkout@v5
       - name: Install uv
         run: |
           curl -fL https://github.com/astral-sh/uv/releases/latest/download/uv-x86_64-unknown-linux-gnu.tar.gz \
@@ -22,7 +22,7 @@ jobs:
     runs-on: ubuntu-latest
     timeout-minutes: 20
     steps:
-      - uses: actions/checkout@v4
+      - uses: actions/checkout@v5
       - name: Install uv
         run: |
           curl -fL https://github.com/astral-sh/uv/releases/latest/download/uv-x86_64-unknown-linux-gnu.tar.gz \
@@ -44,7 +44,7 @@ jobs:
 
       - name: Upload logs on failure
         if: failure()
-        uses: actions/upload-artifact@v4
+        uses: actions/upload-artifact@v5
         with:
           name: e2e-linux-logs
           path: |
@@ -53,9 +53,9 @@ jobs:
 
   e2e-macos:
     runs-on: macos-latest
-    timeout-minutes: 15
+    timeout-minutes: 20
     steps:
-      - uses: actions/checkout@v4
+      - uses: actions/checkout@v5
       - name: Install uv
         run: |
           curl -fL https://github.com/astral-sh/uv/releases/latest/download/uv-aarch64-apple-darwin.tar.gz \
@@ -68,10 +68,11 @@ jobs:
         run: uv run pytest tests/test_e2e.py -v -s
         env:
           VM_STATE_DIR: ${{ runner.temp }}/vm-state
+          QEMU_ACCEL: tcg  # HVF is unavailable on GitHub-hosted macOS runners
 
       - name: Upload logs on failure
         if: failure()
-        uses: actions/upload-artifact@v4
+        uses: actions/upload-artifact@v5
         with:
           name: e2e-macos-logs
           path: |

README.md

@@ -1,4 +1,4 @@
-# Agent VM
+# less-lethal: userspace LLM agent VM
 
 A sandboxed Debian VM with no direct internet access. All traffic is forced through a host-side [mitmproxy](https://mitmproxy.org/) that enforces an allowlist, giving full visibility and control over what the guest can reach. Runs on macOS (Hypervisor.framework) and Linux (KVM or software emulation). No sudo required.
 

tests/test_e2e.py

@@ -237,10 +237,15 @@ def test_cloud_init_success(running_vm):
     SSH subprocess open during the entire cloud-init run (which includes
     package installation and can take several minutes in TCG mode).
     """
-    deadline = time.monotonic() + 300
+    deadline = time.monotonic() + 600
     last_detail = ""
     while time.monotonic() < deadline:
-        r = _vm_ssh("cloud-init status --long 2>&1", timeout=15)
+        try:
+            r = _vm_ssh("cloud-init status --long 2>&1", timeout=30)
+        except subprocess.TimeoutExpired:
+            remaining = int(deadline - time.monotonic())
+            _progress(f"cloud-init ({remaining}s left): (SSH timed out, retrying)")
+            continue
         remaining = int(deadline - time.monotonic())
         # Compact multi-line status into a single progress line.
         detail = " | ".join(

vm.py

@@ -147,18 +147,27 @@ def launch_qemu(self, qemu_args: list[str]) -> subprocess.Popen:
 # ---------------------------------------------------------------------------
 
 class DarwinBackend(Backend):
-    """macOS backend: HVF acceleration, Homebrew firmware paths."""
+    """macOS backend: HVF acceleration (with TCG fallback), Homebrew firmware paths."""
 
     def __init__(self, brew: Path, arch: Arch, proxy_port: int = PROXY_PORT,
                  ssh_host_port: int = SSH_HOST_PORT) -> None:
         super().__init__(arch, proxy_port, ssh_host_port)
         self._brew = brew
+        override = os.environ.get("QEMU_ACCEL")
+        if override:
+            self._accel = override
+        else:
+            r = subprocess.run(["sysctl", "-n", "kern.hv_support"],
+                               capture_output=True, text=True)
+            self._accel = "hvf" if r.returncode == 0 and r.stdout.strip() == "1" else "tcg"
 
     @property
     def machine_args(self) -> list[str]:
         if self.arch == Arch.ARM64:
-            return ["-machine", "virt,accel=hvf", "-cpu", "host"]
-        return ["-machine", "q35,accel=hvf", "-cpu", "host"]
+            cpu = "host" if self._accel == "hvf" else "cortex-a57"
+            return ["-machine", f"virt,accel={self._accel}", "-cpu", cpu]
+        cpu = "host" if self._accel == "hvf" else "qemu64"
+        return ["-machine", f"q35,accel={self._accel}", "-cpu", cpu]
 
     def prepare_efi(self, state_dir: Path) -> tuple[Path, Path]:
         code_src = self._brew / "share/qemu/edk2-aarch64-code.fd"
@@ -178,7 +187,13 @@ class LinuxBackend(Backend):
     def __init__(self, arch: Arch, proxy_port: int = PROXY_PORT,
                  ssh_host_port: int = SSH_HOST_PORT) -> None:
         super().__init__(arch, proxy_port, ssh_host_port)
-        self._accel = "kvm" if Path("/dev/kvm").exists() else "tcg"
+        override = os.environ.get("QEMU_ACCEL")
+        if override:
+            self._accel = override
+        elif os.access("/dev/kvm", os.R_OK | os.W_OK):
+            self._accel = "kvm"
+        else:
+            self._accel = "tcg"
 
     @property
     def machine_args(self) -> list[str]:
@@ -446,13 +461,17 @@ def start_mitmproxy(proxy_port: int = PROXY_PORT) -> subprocess.Popen:
     log_file = log_path.open("w")
     print(f"Starting mitmproxy on port {proxy_port} (log: .vm/mitmdump.log)...")
     proc = subprocess.Popen(cmd, stdout=log_file, stderr=log_file)
-    time.sleep(1)
-    if proc.poll() is not None:
-        log_file.flush()
-        sys.exit(
-            f"mitmdump failed to start (exit code {proc.returncode}). "
-            f"Check {log_path} — port {proxy_port} may already be in use."
-        )
+
+    # Poll for up to 3 seconds to catch fast failures (e.g. port in use).
+    for _ in range(15):
+        time.sleep(0.2)
+        if proc.poll() is not None:
+            log_file.flush()
+            log_tail = log_path.read_text(errors="replace").strip()
+            sys.exit(
+                f"mitmdump failed to start (exit code {proc.returncode}).\n"
+                f"{log_tail}"
+            )
     return proc