| 1 | """Oak agent-native JSON capability probe rows."""
|
| 2 |
|
| 3 | from __future__ import annotations
|
| 4 |
|
| 5 | import sys
|
| 6 | import unittest
|
| 7 | from pathlib import Path
|
| 8 | from unittest import mock
|
| 9 |
|
| 10 | sys.path.insert(0, str(Path(__file__).resolve().parents[1] / "scripts"))
|
| 11 |
|
| 12 | import bench # noqa: E402
|
| 13 | from oakbench.execution import TimedCapture # noqa: E402
|
| 14 | from oakbench.rows import SKIP_RETURNCODE, validate_rows # noqa: E402
|
| 15 | from oakbench.subjects import Subject # noqa: E402
|
| 16 |
|
| 17 |
|
| 18 | SUBJECT = Subject(name="oak_installed", kind="oak", label="Oak", bin=Path("oak"))
|
| 19 | SCENARIO = bench.Scenario("tiny_text", 1, 1, 1, False, 1)
|
| 20 | METADATA = {"bench_id": "test", "profile": "smoke"}
|
| 21 |
|
| 22 |
|
| 23 | class JsonFieldOracleTest(unittest.TestCase):
|
| 24 | def test_parse_success_and_required_fields(self) -> None:
|
| 25 | fields = bench.json_field_oracle_fields(
|
| 26 | '{"schema_version":1,"branch":"agent-task","changes":[]}',
|
| 27 | ["schema_version", "branch", "changes"],
|
| 28 | )
|
| 29 |
|
| 30 | self.assertTrue(fields["json_parse_success"])
|
| 31 | self.assertTrue(fields["json_oracle_passed"])
|
| 32 | self.assertEqual(fields["json_field_completeness"], 1.0)
|
| 33 | self.assertEqual(fields["json_missing_required_fields"], [])
|
| 34 |
|
| 35 | def test_malformed_json_records_failed_oracle(self) -> None:
|
| 36 | fields = bench.json_field_oracle_fields("{not json", ["schema_version"])
|
| 37 |
|
| 38 | self.assertFalse(fields["json_parse_success"])
|
| 39 | self.assertFalse(fields["json_oracle_passed"])
|
| 40 | self.assertEqual(fields["json_missing_required_fields"], ["schema_version"])
|
| 41 | self.assertIn("json_error", fields)
|
| 42 |
|
| 43 | def test_missing_fields_records_completeness(self) -> None:
|
| 44 | fields = bench.json_field_oracle_fields('{"schema_version":1}', ["schema_version", "branch"])
|
| 45 |
|
| 46 | self.assertTrue(fields["json_parse_success"])
|
| 47 | self.assertFalse(fields["json_oracle_passed"])
|
| 48 | self.assertEqual(fields["json_present_required_field_count"], 1)
|
| 49 | self.assertEqual(fields["json_missing_required_fields"], ["branch"])
|
| 50 | self.assertEqual(fields["json_field_completeness"], 0.5)
|
| 51 |
|
| 52 |
|
| 53 | class AgentNativeJsonProbeRowTest(unittest.TestCase):
|
| 54 | def test_capability_probe_uses_isolated_benchmark_environment(self) -> None:
|
| 55 | completed = mock.Mock(returncode=0, stdout="Usage: oak status --json\n", stderr="")
|
| 56 | with mock.patch.object(bench.subprocess, "run", return_value=completed) as run:
|
| 57 | reason = bench.oak_json_probe_skip_reason("oak", ["status", "--help"], ["status", "--json"])
|
| 58 |
|
| 59 | self.assertIsNone(reason)
|
| 60 | self.assertEqual(run.call_args.kwargs["env"]["GIT_CONFIG_GLOBAL"], bench.os.devnull)
|
| 61 | self.assertEqual(run.call_args.kwargs["env"]["LC_ALL"], "C")
|
| 62 |
|
| 63 | def test_missing_command_emits_capability_skip_row(self) -> None:
|
| 64 | spec = {
|
| 65 | "args": ["agent", "state", "--json"],
|
| 66 | "help_args": ["agent", "state", "--help"],
|
| 67 | "required_fields": ["schema_version"],
|
| 68 | }
|
| 69 |
|
| 70 | with mock.patch.object(
|
| 71 | bench,
|
| 72 | "oak_json_probe_skip_reason",
|
| 73 | return_value="capability_gap:oak_json_probe_help_failed:agent state --help",
|
| 74 | ):
|
| 75 | row = bench.run_agent_native_json_probe(
|
| 76 | SUBJECT, SCENARIO, "oak.agent.state.json", Path.cwd(), spec, 0, METADATA, 20000
|
| 77 | )
|
| 78 |
|
| 79 | self.assertEqual(row["returncode"], SKIP_RETURNCODE)
|
| 80 | self.assertTrue(row["skipped"])
|
| 81 | self.assertEqual(row["command"], [])
|
| 82 | self.assertEqual(row["intended_command"], ["oak", "agent", "state", "--json"])
|
| 83 | self.assertEqual(validate_rows([row], "core"), [])
|
| 84 |
|
| 85 | def test_minimum_oak_version_emits_capability_skip_row(self) -> None:
|
| 86 | spec = {
|
| 87 | "args": ["agent", "state", "--json"],
|
| 88 | "help_args": ["agent", "state", "--help"],
|
| 89 | "required_fields": ["schema_version", "recommended_action"],
|
| 90 | "minimum_oak_version": "0.98.0",
|
| 91 | }
|
| 92 |
|
| 93 | with mock.patch.object(bench, "oak_cli_semver", return_value=(0, 97, 0)):
|
| 94 | row = bench.run_agent_native_json_probe(
|
| 95 | SUBJECT, SCENARIO, "oak.agent.state.json", Path.cwd(), spec, 0, METADATA, 20000
|
| 96 | )
|
| 97 |
|
| 98 | self.assertEqual(row["returncode"], SKIP_RETURNCODE)
|
| 99 | self.assertTrue(row["skipped"])
|
| 100 | self.assertEqual(row["command"], [])
|
| 101 | self.assertEqual(row["intended_command"], ["oak", "agent", "state", "--json"])
|
| 102 | self.assertEqual(row["minimum_oak_version"], "0.98.0")
|
| 103 | self.assertIn("requires_oak>=0.98.0", row["skip_reason"])
|
| 104 | self.assertEqual(validate_rows([row], "core"), [])
|
| 105 |
|
| 106 | def test_present_command_records_tokens_output_and_json_oracle(self) -> None:
|
| 107 | spec = {
|
| 108 | "args": ["status", "--json"],
|
| 109 | "help_args": ["status", "--help"],
|
| 110 | "required_fields": ["schema_version", "branch", "changes"],
|
| 111 | }
|
| 112 | stdout = '{"schema_version":1,"branch":"agent-task","changes":[]}'
|
| 113 | capture = TimedCapture(
|
| 114 | elapsed_ms=12.34,
|
| 115 | returncode=0,
|
| 116 | stdout_text=stdout,
|
| 117 | stdout_bytes=len(stdout),
|
| 118 | stdout_truncated=False,
|
| 119 | stderr_text="",
|
| 120 | stderr_bytes=0,
|
| 121 | stderr_truncated=False,
|
| 122 | peak_rss_bytes=1234,
|
| 123 | )
|
| 124 |
|
| 125 | with mock.patch.object(bench, "oak_json_probe_skip_reason", return_value=None), mock.patch.object(
|
| 126 | bench, "run_timed", return_value=capture
|
| 127 | ):
|
| 128 | row = bench.run_agent_native_json_probe(
|
| 129 | SUBJECT, SCENARIO, "oak.status.json", Path.cwd(), spec, 0, METADATA, 20000
|
| 130 | )
|
| 131 |
|
| 132 | self.assertEqual(row["returncode"], 0)
|
| 133 | self.assertTrue(row["json_parse_success"])
|
| 134 | self.assertTrue(row["json_oracle_passed"])
|
| 135 | self.assertEqual(row["probe_output_bytes"], len(stdout))
|
| 136 | self.assertGreater(row["estimated_tokens_total"], 0)
|
| 137 | self.assertEqual(row["tool_call_count"], 1)
|
| 138 | self.assertEqual(validate_rows([row], "core"), [])
|
| 139 |
|
| 140 | def test_json_oracle_parses_full_stdout_when_admitted_output_is_truncated(self) -> None:
|
| 141 | spec = {
|
| 142 | "args": ["branch", "diff", "agent-task", "--json"],
|
| 143 | "help_args": ["branch", "diff", "--help"],
|
| 144 | "required_fields": ["schema_version", "branch", "changes"],
|
| 145 | }
|
| 146 | stdout = '{"schema_version":1,"branch":"agent-task","changes":["' + ("x" * 100) + '"]}'
|
| 147 | admitted_stdout = stdout[:32]
|
| 148 | capture = TimedCapture(
|
| 149 | elapsed_ms=12.34,
|
| 150 | returncode=0,
|
| 151 | stdout_text=admitted_stdout,
|
| 152 | stdout_bytes=len(stdout),
|
| 153 | stdout_truncated=True,
|
| 154 | stderr_text="",
|
| 155 | stderr_bytes=0,
|
| 156 | stderr_truncated=False,
|
| 157 | peak_rss_bytes=1234,
|
| 158 | stdout_full_text=stdout,
|
| 159 | stdout_full_text_truncated=False,
|
| 160 | )
|
| 161 |
|
| 162 | with mock.patch.object(bench, "oak_json_probe_skip_reason", return_value=None), mock.patch.object(
|
| 163 | bench, "run_timed", return_value=capture
|
| 164 | ) as run_timed:
|
| 165 | row = bench.run_agent_native_json_probe(
|
| 166 | SUBJECT, SCENARIO, "oak.branch.diff.json", Path.cwd(), spec, 0, METADATA, 32
|
| 167 | )
|
| 168 |
|
| 169 | self.assertEqual(run_timed.call_args.args[3], 32)
|
| 170 | self.assertEqual(run_timed.call_args.kwargs["full_output_bytes"], bench.JSON_PROBE_MAX_BYTES)
|
| 171 | self.assertEqual(row["returncode"], 0)
|
| 172 | self.assertTrue(row["stdout_truncated"])
|
| 173 | self.assertEqual(row["token_estimate"]["stdout_admitted_chars"], len(admitted_stdout))
|
| 174 | self.assertTrue(row["json_parse_success"])
|
| 175 | self.assertTrue(row["json_oracle_passed"])
|
| 176 | self.assertEqual(row["json_validation_source"], "full_stdout_capture")
|
| 177 | self.assertEqual(validate_rows([row], "core"), [])
|
| 178 |
|
| 179 | def test_json_oracle_is_unmeasured_when_full_stdout_capture_is_truncated(self) -> None:
|
| 180 | spec = {
|
| 181 | "args": ["branch", "review", "agent-task", "--json"],
|
| 182 | "help_args": ["branch", "review", "--help"],
|
| 183 | "required_fields": ["schema_version", "branch", "findings"],
|
| 184 | }
|
| 185 | capture = TimedCapture(
|
| 186 | elapsed_ms=12.34,
|
| 187 | returncode=0,
|
| 188 | stdout_text='{"schema_version":1,"branch":"agent',
|
| 189 | stdout_bytes=200_000_000,
|
| 190 | stdout_truncated=True,
|
| 191 | stderr_text="",
|
| 192 | stderr_bytes=0,
|
| 193 | stderr_truncated=False,
|
| 194 | peak_rss_bytes=1234,
|
| 195 | stdout_full_text=None,
|
| 196 | stdout_full_text_truncated=True,
|
| 197 | )
|
| 198 |
|
| 199 | with mock.patch.object(bench, "oak_json_probe_skip_reason", return_value=None), mock.patch.object(
|
| 200 | bench, "run_timed", return_value=capture
|
| 201 | ):
|
| 202 | row = bench.run_agent_native_json_probe(
|
| 203 | SUBJECT, SCENARIO, "oak.branch.review.json", Path.cwd(), spec, 0, METADATA, 32
|
| 204 | )
|
| 205 |
|
| 206 | self.assertEqual(row["returncode"], 0)
|
| 207 | self.assertIsNone(row["json_validation_source"])
|
| 208 | self.assertIsNone(row["json_parse_success"])
|
| 209 | self.assertIsNone(row["json_oracle_passed"])
|
| 210 | self.assertEqual(
|
| 211 | row["json_validation_unmeasured_reason"],
|
| 212 | "stdout_exceeded_json_validation_capture_limit",
|
| 213 | )
|
| 214 | self.assertEqual(validate_rows([row], "core"), [])
|
| 215 |
|
| 216 | def test_present_command_with_incomplete_json_fails_row(self) -> None:
|
| 217 | spec = {
|
| 218 | "args": ["status", "--json"],
|
| 219 | "help_args": ["status", "--help"],
|
| 220 | "required_fields": ["schema_version", "branch"],
|
| 221 | }
|
| 222 | capture = TimedCapture(
|
| 223 | elapsed_ms=1.0,
|
| 224 | returncode=0,
|
| 225 | stdout_text='{"schema_version":1}',
|
| 226 | stdout_bytes=len('{"schema_version":1}'),
|
| 227 | stdout_truncated=False,
|
| 228 | stderr_text="",
|
| 229 | stderr_bytes=0,
|
| 230 | stderr_truncated=False,
|
| 231 | )
|
| 232 |
|
| 233 | with mock.patch.object(bench, "oak_json_probe_skip_reason", return_value=None), mock.patch.object(
|
| 234 | bench, "run_timed", return_value=capture
|
| 235 | ):
|
| 236 | row = bench.run_agent_native_json_probe(
|
| 237 | SUBJECT, SCENARIO, "oak.status.json", Path.cwd(), spec, 0, METADATA, 20000
|
| 238 | )
|
| 239 |
|
| 240 | self.assertEqual(row["returncode"], 1)
|
| 241 | self.assertFalse(row["json_oracle_passed"])
|
| 242 | self.assertEqual(row["json_missing_required_fields"], ["branch"])
|
| 243 | self.assertEqual(validate_rows([row], "core"), [])
|
| 244 |
|
| 245 |
|
| 246 | if __name__ == "__main__":
|
| 247 | unittest.main()
|