| 1 | from __future__ import annotations |
| 2 | |
| 3 | import importlib |
| 4 | import json |
| 5 | import sys |
| 6 | import types |
| 7 | import unittest |
| 8 | from dataclasses import dataclass |
| 9 | from pathlib import Path |
| 10 | from types import SimpleNamespace |
| 11 | |
| 12 | |
| 13 | def install_verifiers_stubs() -> None: |
| 14 | class HarnessConfig: |
| 15 | env: dict[str, str] = {} |
| 16 | disabled_tools: list[str] | None = None |
| 17 | |
| 18 | def __init__(self, **values): |
| 19 | self.env = dict(values.pop("env", {})) |
| 20 | self.disabled_tools = values.pop("disabled_tools", None) |
| 21 | for cls in reversed(type(self).mro()): |
| 22 | for name in getattr(cls, "__annotations__", {}): |
| 23 | if name not in self.__dict__ and hasattr(cls, name): |
| 24 | setattr(self, name, getattr(cls, name)) |
| 25 | for name, value in values.items(): |
| 26 | setattr(self, name, value) |
| 27 | |
| 28 | @property |
| 29 | def resolved_env(self): |
| 30 | return dict(self.env) |
| 31 | |
| 32 | class Harness: |
| 33 | @classmethod |
| 34 | def __class_getitem__(cls, _item): |
| 35 | return cls |
| 36 | |
| 37 | def __init__(self, config): |
| 38 | self.config = config |
| 39 | |
| 40 | def resolve_prompt(self, task): |
| 41 | return task.system_prompt, task.prompt |
| 42 | |
| 43 | @dataclass(frozen=True) |
| 44 | class ProgramResult: |
| 45 | exit_code: int |
| 46 | stdout: str |
| 47 | stderr: str |
| 48 | |
| 49 | packages = { |
| 50 | "verifiers": types.ModuleType("verifiers"), |
| 51 | "verifiers.v1": types.ModuleType("verifiers.v1"), |
| 52 | "verifiers.v1.clients": types.ModuleType("verifiers.v1.clients"), |
| 53 | "verifiers.v1.harness": types.ModuleType("verifiers.v1.harness"), |
| 54 | "verifiers.v1.runtimes": types.ModuleType("verifiers.v1.runtimes"), |
| 55 | "verifiers.v1.trace": types.ModuleType("verifiers.v1.trace"), |
| 56 | } |
| 57 | packages["verifiers.v1.clients"].ModelContext = type("ModelContext", (), {}) |
| 58 | packages["verifiers.v1.harness"].Harness = Harness |
| 59 | packages["verifiers.v1.harness"].HarnessConfig = HarnessConfig |
| 60 | packages["verifiers.v1.runtimes"].ProgramResult = ProgramResult |
| 61 | packages["verifiers.v1.runtimes"].Runtime = type("Runtime", (), {}) |
| 62 | packages["verifiers.v1.trace"].Trace = type("Trace", (), {}) |
| 63 | sys.modules.update(packages) |
| 64 | |
| 65 | |
| 66 | ROOT = Path(__file__).resolve().parents[1] |
| 67 | sys.path.insert(0, str(ROOT)) |
| 68 | install_verifiers_stubs() |
| 69 | harness_module = importlib.import_module("codewhale_harness.harness") |
| 70 | |
| 71 | CodewhaleHarness = harness_module.CodewhaleHarness |
| 72 | CodewhaleHarnessConfig = harness_module.CodewhaleHarnessConfig |
| 73 | ProgramResult = sys.modules["verifiers.v1.runtimes"].ProgramResult |
| 74 | |
| 75 | |
| 76 | def event(event_type: str, **fields) -> str: |
| 77 | return json.dumps( |
| 78 | { |
| 79 | "schema": "codewhale.exec-stream", |
| 80 | "schema_version": 1, |
| 81 | "type": event_type, |
| 82 | **fields, |
| 83 | } |
| 84 | ) |
| 85 | |
| 86 | |
| 87 | def successful_stream(model: str = "test/model", sandbox: str = "workspace-write") -> str: |
| 88 | sha = "sha256:" + "a" * 64 |
| 89 | return "\n".join( |
| 90 | [ |
| 91 | event("content", content="sensitive model output"), |
| 92 | event( |
| 93 | "metadata", |
| 94 | meta={ |
| 95 | "receipt_kind": "terminal", |
| 96 | "provider": "openai", |
| 97 | "model": model, |
| 98 | "approval_posture": "auto_tools", |
| 99 | "sandbox_posture": sandbox, |
| 100 | "binary_sha256": sha, |
| 101 | "prompt_sha256": sha, |
| 102 | "status": "completed", |
| 103 | "termination_reason": "resolved", |
| 104 | "workspace": "/private/runtime/path", |
| 105 | "session_id": "private-session-id", |
| 106 | }, |
| 107 | ), |
| 108 | event("done"), |
| 109 | ] |
| 110 | ) |
| 111 | |
| 112 | |
| 113 | class FakeRuntime: |
| 114 | def __init__(self, runtime_type="subprocess", result=None): |
| 115 | self.type = runtime_type |
| 116 | self.result = result or ProgramResult(0, successful_stream(), "") |
| 117 | self.runs = [] |
| 118 | self.programs = [] |
| 119 | self.writes = [] |
| 120 | |
| 121 | async def run(self, argv, env): |
| 122 | self.runs.append((argv, env)) |
| 123 | if argv[-1:] == ["--version"]: |
| 124 | return ProgramResult(0, "codewhale 0.9.1 (candidate)", "") |
| 125 | return ProgramResult(0, "", "") |
| 126 | |
| 127 | async def run_program(self, argv, env): |
| 128 | self.programs.append((argv, env)) |
| 129 | return self.result |
| 130 | |
| 131 | async def write(self, path, data): |
| 132 | self.writes.append((path, data)) |
| 133 | |
| 134 | |
| 135 | def trace(prompt="Do the task", system_prompt="Use evidence"): |
| 136 | return SimpleNamespace( |
| 137 | id="0123456789abcdef", |
| 138 | task=SimpleNamespace( |
| 139 | data=SimpleNamespace(prompt=prompt, system_prompt=system_prompt) |
| 140 | ), |
| 141 | info={}, |
| 142 | ) |
| 143 | |
| 144 | |
| 145 | class HarnessTests(unittest.IsolatedAsyncioTestCase): |
| 146 | async def test_preinstalled_candidate_requires_exact_version(self): |
| 147 | config = CodewhaleHarnessConfig( |
| 148 | version="0.9.1", binary_path="/candidate/codewhale" |
| 149 | ) |
| 150 | runtime = FakeRuntime() |
| 151 | |
| 152 | await CodewhaleHarness(config).setup(runtime) |
| 153 | |
| 154 | self.assertEqual(runtime.runs, [(["/candidate/codewhale", "--version"], {})]) |
| 155 | |
| 156 | async def test_launch_isolates_config_routes_interception_and_keeps_safe_receipt(self): |
| 157 | config = CodewhaleHarnessConfig( |
| 158 | version="0.9.1", |
| 159 | binary_path="/candidate/codewhale", |
| 160 | max_turns=17, |
| 161 | disabled_tools=["web_run"], |
| 162 | env={"CALLER_SETTING": "kept", "OPENAI_API_KEY": "must-not-win"}, |
| 163 | ) |
| 164 | runtime = FakeRuntime() |
| 165 | rollout = trace() |
| 166 | ctx = SimpleNamespace(model="test/model") |
| 167 | |
| 168 | await CodewhaleHarness(config).launch( |
| 169 | ctx, |
| 170 | rollout, |
| 171 | runtime, |
| 172 | "http://127.0.0.1:9000/v1/", |
| 173 | "vf-session-secret", |
| 174 | {"grader": "http://127.0.0.1:8123/mcp"}, |
| 175 | ) |
| 176 | |
| 177 | self.assertEqual(len(runtime.programs), 1) |
| 178 | argv, env = runtime.programs[0] |
| 179 | self.assertEqual(argv[0], "/candidate/codewhale") |
| 180 | self.assertIn("--no-project-config", argv) |
| 181 | self.assertIn("--skip-onboarding", argv) |
| 182 | self.assertEqual(argv[argv.index("--sandbox") + 1], "workspace-write") |
| 183 | self.assertEqual(argv[argv.index("--max-turns") + 1], "17") |
| 184 | self.assertEqual(argv[argv.index("--disallowed-tools") + 1], "web_run") |
| 185 | self.assertEqual(argv[argv.index("--append-system-prompt") + 1], "Use evidence") |
| 186 | self.assertNotIn("vf-session-secret", argv) |
| 187 | self.assertEqual(env["OPENAI_API_KEY"], "vf-session-secret") |
| 188 | self.assertEqual(env["OPENAI_BASE_URL"], "http://127.0.0.1:9000/v1") |
| 189 | self.assertEqual(env["CODEWHALE_PROVIDER"], "openai") |
| 190 | self.assertEqual(env["CODEWHALE_MODEL"], "test/model") |
| 191 | self.assertEqual(env["CODEWHALE_TELEMETRY"], "false") |
| 192 | self.assertEqual(env["CODEWHALE_ALLOW_INSECURE_HTTP"], "1") |
| 193 | self.assertEqual(env["CALLER_SETTING"], "kept") |
| 194 | self.assertEqual(len(runtime.writes), 1) |
| 195 | mcp_path, mcp_bytes = runtime.writes[0] |
| 196 | self.assertEqual(mcp_path, env["CODEWHALE_MCP_CONFIG"]) |
| 197 | self.assertTrue(mcp_path.startswith(".vf-codewhale/")) |
| 198 | self.assertNotIn(rollout.id, mcp_path) |
| 199 | self.assertEqual( |
| 200 | json.loads(mcp_bytes), |
| 201 | {"servers": {"grader": {"url": "http://127.0.0.1:8123/mcp"}}}, |
| 202 | ) |
| 203 | |
| 204 | receipt = rollout.info["codewhale"] |
| 205 | self.assertEqual(receipt["events"], {"content": 1, "done": 1, "metadata": 1}) |
| 206 | self.assertNotIn("sensitive model output", json.dumps(receipt)) |
| 207 | self.assertNotIn("private-session-id", json.dumps(receipt)) |
| 208 | self.assertNotIn("/private/runtime/path", json.dumps(receipt)) |
| 209 | self.assertNotIn("vf-session-secret", json.dumps(receipt)) |
| 210 | |
| 211 | async def test_isolated_runtime_uses_external_sandbox_without_elevation_flag(self): |
| 212 | config = CodewhaleHarnessConfig(binary_path="/candidate/codewhale") |
| 213 | runtime = FakeRuntime( |
| 214 | runtime_type="prime", |
| 215 | result=ProgramResult( |
| 216 | 0, successful_stream(sandbox="external-sandbox"), "" |
| 217 | ), |
| 218 | ) |
| 219 | |
| 220 | await CodewhaleHarness(config).launch( |
| 221 | SimpleNamespace(model="test/model"), |
| 222 | trace(), |
| 223 | runtime, |
| 224 | "https://intercept.example/v1", |
| 225 | "secret", |
| 226 | {}, |
| 227 | ) |
| 228 | |
| 229 | argv, env = runtime.programs[0] |
| 230 | self.assertEqual(argv[argv.index("--sandbox") + 1], "external-sandbox") |
| 231 | self.assertNotIn("--allow-sandbox-elevation", argv) |
| 232 | self.assertNotIn("--max-turns", argv) |
| 233 | self.assertNotIn("CODEWHALE_ALLOW_INSECURE_HTTP", env) |
| 234 | |
| 235 | async def test_success_without_exact_terminal_receipt_fails_closed(self): |
| 236 | runtime = FakeRuntime(result=ProgramResult(0, event("content", content="ok"), "")) |
| 237 | config = CodewhaleHarnessConfig(binary_path="/candidate/codewhale") |
| 238 | |
| 239 | with self.assertRaisesRegex(RuntimeError, "omitted terminal metadata"): |
| 240 | await CodewhaleHarness(config).launch( |
| 241 | SimpleNamespace(model="test/model"), |
| 242 | trace(), |
| 243 | runtime, |
| 244 | "https://intercept.example/v1", |
| 245 | "secret-not-in-error", |
| 246 | {}, |
| 247 | ) |
| 248 | |
| 249 | async def test_non_resolved_terminal_receipt_fails_closed(self): |
| 250 | stdout = successful_stream().replace('"resolved"', '"approval_required"') |
| 251 | runtime = FakeRuntime(result=ProgramResult(0, stdout, "")) |
| 252 | config = CodewhaleHarnessConfig(binary_path="/candidate/codewhale") |
| 253 | |
| 254 | with self.assertRaisesRegex(RuntimeError, "was not resolved"): |
| 255 | await CodewhaleHarness(config).launch( |
| 256 | SimpleNamespace(model="test/model"), |
| 257 | trace(), |
| 258 | runtime, |
| 259 | "https://intercept.example/v1", |
| 260 | "secret-not-in-error", |
| 261 | {}, |
| 262 | ) |
| 263 | |
| 264 | async def test_oversized_terminal_receipt_fails_closed_without_copying_content(self): |
| 265 | stdout = successful_stream().replace("test/model", "x" * 513) |
| 266 | runtime = FakeRuntime(result=ProgramResult(0, stdout, "")) |
| 267 | config = CodewhaleHarnessConfig(binary_path="/candidate/codewhale") |
| 268 | |
| 269 | with self.assertRaisesRegex(RuntimeError, "exceeded its string bound"): |
| 270 | await CodewhaleHarness(config).launch( |
| 271 | SimpleNamespace(model="x" * 513), |
| 272 | trace(), |
| 273 | runtime, |
| 274 | "https://intercept.example/v1", |
| 275 | "secret-not-in-error", |
| 276 | {}, |
| 277 | ) |
| 278 | |
| 279 | def test_exact_version_rejects_prerelease_and_longer_tokens(self): |
| 280 | has_version = harness_module._has_version |
| 281 | |
| 282 | self.assertTrue(has_version("codewhale 0.9.1 (abc123)", "0.9.1")) |
| 283 | self.assertFalse(has_version("codewhale 0.9.1-rc.1", "0.9.1")) |
| 284 | self.assertFalse(has_version("codewhale 0.9.10", "0.9.1")) |
| 285 | |
| 286 | async def test_invalid_config_rejects_shell_and_tool_injection(self): |
| 287 | for config in [ |
| 288 | CodewhaleHarnessConfig(version="0.9.1; touch /tmp/pwned"), |
| 289 | CodewhaleHarnessConfig(disabled_tools=["read_file,exec_shell"]), |
| 290 | CodewhaleHarnessConfig(max_turns=0), |
| 291 | ]: |
| 292 | with self.assertRaises(ValueError): |
| 293 | await CodewhaleHarness(config).setup(FakeRuntime()) |
| 294 | |
| 295 | async def test_install_is_pinned_and_checksum_verified(self): |
| 296 | config = CodewhaleHarnessConfig(version="0.9.1") |
| 297 | runtime = FakeRuntime() |
| 298 | |
| 299 | await CodewhaleHarness(config).setup(runtime) |
| 300 | |
| 301 | self.assertEqual(len(runtime.runs), 1) |
| 302 | argv, env = runtime.runs[0] |
| 303 | self.assertEqual(argv[:2], ["sh", "-c"]) |
| 304 | self.assertEqual(env, {}) |
| 305 | command = argv[2] |
| 306 | self.assertIn("releases/download", command) |
| 307 | self.assertIn("codewhale-artifacts-sha256.txt", command) |
| 308 | self.assertIn("sha256sum", command) |
| 309 | self.assertIn("flock", command) |
| 310 | self.assertIn("util-linux", command) |
| 311 | |
| 312 | |
| 313 | if __name__ == "__main__": |
| 314 | unittest.main() |
| 315 |