fix: sanitize subprocess call in runner.py (#2149)

* fix: V-001 security vulnerability

Automated security fix generated by OrbisAI Security

* fix: sanitize subprocess call in runner.py

The runner

* fix: address PR review comments on V-001 allowlist and test coverage

Remove dangerous interpreters (python, python3, node, curl, wget) from
ALLOWED_SETUP_EXECUTABLES — they can execute arbitrary code via argument
flags and are not needed for sandbox setup. Rewrite test_invariant_runner
to call _setup_sandbox directly instead of spawning runner.py as a
subprocess (which had no __main__ entrypoint and never exercised the fix).

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

---------

Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>

skills/skill-comply/scripts/runner.py

@@ -15,6 +15,11 @@ from scripts.scenario_generator import Scenario
 
 SANDBOX_BASE = Path("/tmp/skill-comply-sandbox")
 ALLOWED_MODELS = frozenset({"haiku", "sonnet", "opus"})
+ALLOWED_SETUP_EXECUTABLES = frozenset({
+    "git", "npm", "pip", "pip3",
+    "touch", "mkdir", "cp", "mv", "echo",
+    "chmod", "unzip", "tar",
+})
 # Shell builtins cannot be invoked via subprocess.run; cwd is already
 # controlled by the cwd= keyword. Scenarios that include these in
 # setup_commands (a common shell-style convention) must be tolerated.
@@ -106,6 +111,9 @@ def _setup_sandbox(sandbox_dir: Path, scenario: Scenario) -> None:
         if not parts or parts[0] in SHELL_BUILTINS:
             # Shell builtins (cd/pushd/popd) cannot run as subprocess; skip.
             continue
+        if parts[0] not in ALLOWED_SETUP_EXECUTABLES:
+            # Restrict to known-safe executables to prevent arbitrary code execution.
+            continue
         try:
             subprocess.run(parts, cwd=sandbox_dir, capture_output=True)
         except FileNotFoundError:

tests/test_invariant_runner.py

@@ -0,0 +1,64 @@
+import os
+import sys
+import pytest
+from pathlib import Path
+
+_SKILL_COMPLY_ROOT = Path(__file__).resolve().parent.parent / "skills" / "skill-comply"
+if str(_SKILL_COMPLY_ROOT) not in sys.path:
+    sys.path.insert(0, str(_SKILL_COMPLY_ROOT))
+
+from scripts.runner import _setup_sandbox          # noqa: E402
+from scripts.scenario_generator import Scenario    # noqa: E402
+
+_GLOBAL_MARKER = "/tmp/runner_test_pwned_marker"
+
+
+@pytest.fixture(autouse=True)
+def _remove_marker():
+    if os.path.exists(_GLOBAL_MARKER):
+        os.remove(_GLOBAL_MARKER)
+    yield
+    if os.path.exists(_GLOBAL_MARKER):
+        os.remove(_GLOBAL_MARKER)
+
+
+@pytest.mark.parametrize(
+    "setup_commands,test_id",
+    [
+        (
+            ("python -c \"import os; os.system('touch /tmp/runner_test_pwned_marker')\"",),
+            "python_interpreter",
+        ),
+        (
+            ("../../../../../../bin/sh -c 'touch /tmp/runner_test_pwned_marker'",),
+            "path_traversal",
+        ),
+        (
+            ("bash -c 'touch /tmp/runner_test_pwned_marker'",),
+            "non_allowlisted_binary",
+        ),
+        (
+            ("echo hello",),
+            "benign_echo",
+        ),
+    ],
+    ids=["python_interpreter", "path_traversal", "non_allowlisted_binary", "benign_echo"],
+)
+def test_setup_sandbox_blocks_dangerous_commands(setup_commands, test_id, tmp_path):
+    """Invariant: _setup_sandbox must not execute disallowed commands."""
+    scenario = Scenario(
+        id=f"test-{test_id}",
+        level=1,
+        level_name="basic",
+        description="security test scenario",
+        prompt="",
+        setup_commands=setup_commands,
+    )
+    sandbox_dir = tmp_path / "sandbox"
+
+    _setup_sandbox(sandbox_dir, scenario)
+
+    assert not os.path.exists(_GLOBAL_MARKER), (
+        f"Arbitrary command execution detected for '{test_id}': "
+        f"marker file created at {_GLOBAL_MARKER}"
+    )