Viewing files at commit fcd8a785ee1aCommit detailsView latest

Browse files
tests/test_agent_campaign_cli.py 243 lines · 12.8 KB · python
1
"""Independent CLI integration: real Git, mock agent, durable campaign records."""
2
from pathlib import Path
3
import json
4
import shutil
5
import subprocess
6
import sys
7
import tempfile
8
import unittest
9
10
11
ROOT = Path(__file__).resolve().parents[1]
12
CLI = ROOT / "scripts" / "agent_campaign.py"
13
14
15
@unittest.skipUnless(shutil.which("git"), "Git is required for the campaign integration")
16
class CampaignCLIIntegrationTests(unittest.TestCase):
17
    def setUp(self):
18
        self.temp = tempfile.TemporaryDirectory(prefix="oakbench-campaign-cli-")
19
        self.addCleanup(self.temp.cleanup)
20
        self.root = Path(self.temp.name).resolve()
21
        self.campaign = self.root / "campaign"
22
        self.agents = self.root / "agents.toml"
23
        self.agents.write_text('[agents.mock]\nenabled = true\nadapter = "deterministic_mock"\nprovider = "local"\n')
24
        self.subjects = self.root / "subjects.toml"
25
        self.subjects.write_text('[subjects.git]\nenabled = true\nkind = "git"\nlabel = "Git"\nbin = '
26
                                 + json.dumps(shutil.which("git")) + '\n')
27
        self.manifest = self.root / "manifest.json"
28
        self.manifest.write_text(json.dumps({
29
            "schema_version": 1, "campaign_id": "cli-integration",
30
            "agents_config": str(self.agents), "subjects_config": str(self.subjects),
31
            "agents": [{"name": "mock", "model": None}], "subjects": ["git"],
32
            "workflows": ["wide_config_refactor"], "instruction_levels": ["cheat-sheet"],
33
            "repetitions": 2, "seed": 194, "max_trials": 2, "max_wall_seconds": 180,
34
            "trial_timeout_seconds": 30, "track": "agent-default", "agent_environment": "minimal",
35
        }))
36
37
    def invoke(self, command, *args):
38
        return subprocess.run([sys.executable, str(CLI), command,
39
                               "--campaign-dir", str(self.campaign), *args],
40
                              cwd=ROOT, text=True, capture_output=True, timeout=120)
41
42
    def plan(self):
43
        result = self.invoke("plan", "--manifest", str(self.manifest))
44
        self.assertEqual(result.returncode, 0, result.stdout + result.stderr)
45
        return json.loads((self.campaign / "plan.json").read_text())
46
47
    def fixture_hash(self, plan):
48
        sys.path.insert(0, str(ROOT / "scripts"))
49
        from oakbench import agent_campaign
50
        import workflow_ab
51
        agent_campaign.cached_fixture(self.campaign, plan, workflow_ab.WORKFLOWS["wide_config_refactor"])
52
        record = json.loads((self.campaign / "fixture-integrity/wide_config_refactor.json").read_text())
53
        return agent_campaign.digest(record["manifest"])
54
55
    def events(self):
56
        return [json.loads(line) for line in (self.campaign / "events.jsonl").read_text().splitlines()]
57
58
    def test_real_git_mock_completion_and_noop_resume_preserve_evidence(self):
59
        plan = self.plan()
60
        self.assertEqual(len(plan["schedule"]), 2)
61
        result = self.invoke("run")
62
        self.assertEqual(result.returncode, 0, result.stdout + result.stderr)
63
        self.assertEqual(json.loads(result.stdout)["finished"], 2)
64
        self.assertIn("[trial]", result.stderr)
65
        events = self.events()
66
        self.assertEqual([event["event"] for event in events], ["started", "finished"] * 2)
67
        finished = [event for event in events if event["event"] == "finished"]
68
        for event in finished:
69
            self.assertTrue(event["row"]["success"], event)
70
            self.assertTrue(event["row"]["success_criteria"]["checks"]["required_checks_passed"])
71
            self.assertEqual(event["row"]["task_id"], event["trial_id"])
72
        journal = (self.campaign / "events.jsonl").read_bytes()
73
        # Simulate the crash window after the authoritative terminal fsync but
74
        # before refreshing the derived rows. Resume must repair this view.
75
        (self.campaign / "rows.jsonl").write_text("")
76
        result = self.invoke("resume")
77
        self.assertEqual(result.returncode, 0, result.stdout + result.stderr)
78
        self.assertEqual((self.campaign / "events.jsonl").read_bytes(), journal)
79
        recovered = [json.loads(line) for line in (self.campaign / "rows.jsonl").read_text().splitlines()]
80
        self.assertEqual(recovered, [event["row"] for event in finished])
81
        result = self.invoke("report")
82
        self.assertEqual(result.returncode, 0, result.stdout + result.stderr)
83
        self.assertIn('"claim_eligible": false', result.stdout.lower())
84
85
    def test_corrupted_cache_resume_is_preflight_error_preserving_pending(self):
86
        self.plan()
87
        first = self.invoke("run", "--limit", "1")
88
        self.assertEqual(first.returncode, 3, first.stderr)
89
        journal = (self.campaign / "events.jsonl").read_bytes()
90
        payload = next((self.campaign / "fixtures/wide_config_refactor").glob("services/*/config.toml"))
91
        payload.chmod(0o600)
92
        payload.write_text("changed cached initial state")
93
        result = self.invoke("resume")
94
        self.assertEqual(result.returncode, 2, result.stdout + result.stderr)
95
        self.assertIn("fixture content drift", result.stderr)
96
        self.assertEqual((self.campaign / "events.jsonl").read_bytes(), journal)
97
        status = self.invoke("status")
98
        self.assertEqual(len(json.loads(status.stdout)["pending_trial_ids"]), 1)
99
100
    def test_cache_and_record_rewrite_cannot_rebind_prior_trial_input(self):
101
        import hashlib
102
        self.plan()
103
        first = self.invoke("run", "--limit", "1")
104
        self.assertEqual(first.returncode, 3, first.stderr)
105
        journal = (self.campaign / "events.jsonl").read_bytes()
106
        fixture = self.campaign / "fixtures/wide_config_refactor"
107
        payload = next(fixture.glob("services/*/config.toml"))
108
        payload.chmod(0o600)
109
        payload.write_bytes(b"changed baseline plus forged cache record")
110
        record_path = self.campaign / "fixture-integrity/wide_config_refactor.json"
111
        record = json.loads(record_path.read_text())
112
        entry = record["manifest"][payload.relative_to(fixture).as_posix()]
113
        entry.update(size=len(payload.read_bytes()), sha256=hashlib.sha256(payload.read_bytes()).hexdigest())
114
        record_path.write_text(json.dumps(record))
115
        result = self.invoke("resume")
116
        self.assertEqual(result.returncode, 2, result.stdout + result.stderr)
117
        self.assertIn("journal fixture binding", result.stderr)
118
        self.assertEqual((self.campaign / "events.jsonl").read_bytes(), journal)
119
        status = self.invoke("status")
120
        self.assertEqual(len(json.loads(status.stdout)["pending_trial_ids"]), 1)
121
122
    def test_incomplete_generation_fails_before_any_admission(self):
123
        self.plan()
124
        fixture = self.campaign / "fixtures/wide_config_refactor"
125
        fixture.mkdir(parents=True)
126
        (fixture / "partial").write_text("crashed generation")
127
        result = self.invoke("run")
128
        self.assertEqual(result.returncode, 2, result.stdout + result.stderr)
129
        self.assertIn("fixture cache is incomplete", result.stderr)
130
        self.assertFalse((self.campaign / "events.jsonl").exists())
131
        status = self.invoke("status")
132
        self.assertEqual(len(json.loads(status.stdout)["pending_trial_ids"]), 2)
133
134
    def test_existing_plan_and_changed_config_fail_before_agent_execution(self):
135
        self.plan()
136
        frozen = (self.campaign / "plan.json").read_bytes()
137
        result = self.invoke("plan", "--manifest", str(self.manifest))
138
        self.assertNotEqual(result.returncode, 0)
139
        self.assertEqual((self.campaign / "plan.json").read_bytes(), frozen)
140
        self.agents.write_text(self.agents.read_text() + "\n# changed configuration\n")
141
        result = self.invoke("run")
142
        self.assertNotEqual(result.returncode, 0)
143
        self.assertFalse((self.campaign / "events.jsonl").exists())
144
145
    def test_torn_journal_cannot_be_discarded_to_retry_a_trial(self):
146
        plan = self.plan()
147
        started = {"event": "started", "trial_id": plan["schedule"][0]["trial_id"],
148
                   "plan_sha256": plan["plan_sha256"], "timeout_seconds": 30,
149
                   "ordinal": 0, "timestamp_utc": "2026-09-10T00:00:00Z",
150
                   "fixture_manifest_sha256": self.fixture_hash(plan)}
151
        journal = json.dumps(started).encode() + b"\n" + b'{"event":"finished"'
152
        (self.campaign / "events.jsonl").write_bytes(journal)
153
        result = self.invoke("resume")
154
        self.assertEqual(result.returncode, 2)
155
        self.assertIn("--acknowledge-torn-journal", result.stderr)
156
        self.assertEqual((self.campaign / "events.jsonl").read_bytes(), journal)
157
        self.assertFalse((self.campaign / "trials").exists())
158
159
    def test_insufficient_budget_stops_without_shortening_timeout(self):
160
        manifest = json.loads(self.manifest.read_text())
161
        manifest["max_wall_seconds"] = 0.15
162
        self.manifest.write_text(json.dumps(manifest))
163
        self.plan()
164
        result = self.invoke("run")
165
        self.assertEqual(result.returncode, 3, result.stdout + result.stderr)
166
        state = json.loads(result.stdout)
167
        self.assertEqual(state["stopped_reason"], "insufficient_budget_for_frozen_trial_timeout")
168
        self.assertEqual(len(state["pending_trial_ids"]), 2)
169
        self.assertFalse((self.campaign / "events.jsonl").exists())
170
171
    def test_torn_start_readonly_and_acknowledged_recovery_never_retries(self):
172
        plan = self.plan()
173
        interrupted = plan["schedule"][0]["trial_id"]
174
        raw = b'{"event":"started","trial_id":'
175
        journal = self.campaign / "events.jsonl"
176
        journal.write_bytes(raw)
177
        for command in ("status", "report"):
178
            result = self.invoke(command)
179
            self.assertEqual(result.returncode, 0, result.stdout + result.stderr)
180
            evidence = json.loads(result.stdout)
181
            self.assertFalse(evidence["journal_recoveries"][0]["acknowledged"])
182
            self.assertEqual(journal.read_bytes(), raw)
183
        result = self.invoke("resume", "--acknowledge-torn-journal")
184
        self.assertEqual(result.returncode, 3, result.stdout + result.stderr)
185
        events = self.events()
186
        self.assertEqual(events[0]["trial_id"], interrupted)
187
        self.assertTrue(events[0]["uncertain_torn_start"])
188
        marker = events[1]
189
        self.assertTrue(marker["acknowledged"])
190
        self.assertEqual((self.campaign / marker["archive"]).read_bytes(), raw)
191
        self.assertFalse((self.campaign / "trials" / interrupted).exists())
192
        self.assertEqual(len([e for e in events if e["event"] == "finished"]), 1)
193
        result = self.invoke("resume")
194
        self.assertEqual(result.returncode, 3, result.stdout + result.stderr)
195
        self.assertEqual(self.events(), events)
196
197
    def test_valid_digest_malformed_plan_exits_contract_error(self):
198
        import hashlib
199
        self.campaign.mkdir()
200
        malformed = {"schema_version": 1}
201
        malformed["plan_sha256"] = hashlib.sha256(b'{"schema_version":1}').hexdigest()
202
        (self.campaign / "plan.json").write_text(json.dumps(malformed))
203
        result = self.invoke("status")
204
        self.assertEqual(result.returncode, 2)
205
        self.assertIn("invalid frozen plan", result.stderr)
206
        self.assertNotIn("Traceback", result.stderr)
207
208
    def test_bounded_invocation_resumes_only_remaining_trial(self):
209
        self.plan()
210
        result = self.invoke("run", "--limit", "1")
211
        self.assertEqual(result.returncode, 3, result.stdout + result.stderr)
212
        first = self.events()
213
        self.assertEqual(len(first), 2)
214
        self.assertTrue(first[1]["row"]["success"], first[1])
215
        result = self.invoke("resume", "--limit", "1")
216
        self.assertEqual(result.returncode, 0, result.stdout + result.stderr)
217
        self.assertEqual(self.events()[:2], first)
218
        self.assertEqual(len(self.events()), 4)
219
220
    def test_interrupted_attempt_stays_visible_after_remaining_trial_runs(self):
221
        plan = self.plan()
222
        interrupted = plan["schedule"][0]["trial_id"]
223
        started = {"event": "started", "trial_id": interrupted,
224
                   "plan_sha256": plan["plan_sha256"], "timeout_seconds": 30,
225
                   "ordinal": 0, "timestamp_utc": "2026-09-10T00:00:00Z",
226
                   "fixture_manifest_sha256": self.fixture_hash(plan)}
227
        (self.campaign / "events.jsonl").write_text(json.dumps(started) + "\n")
228
        result = self.invoke("resume")
229
        self.assertEqual(result.returncode, 3, result.stdout + result.stderr)
230
        events = self.events()
231
        self.assertEqual(len(events), 3)
232
        self.assertEqual(events[0], started)
233
        self.assertNotEqual(events[-1]["trial_id"], interrupted)
234
        self.assertTrue(events[-1]["row"]["success"], events[-1])
235
        self.assertFalse((self.campaign / "trials" / interrupted).exists())
236
        result = self.invoke("report", "--format", "json")
237
        self.assertEqual(result.returncode, 0, result.stdout + result.stderr)
238
        counts = json.loads(result.stdout)["counts"]
239
        self.assertEqual((counts["started"], counts["success"], counts["interrupted"]), (2, 1, 1))
240
241
242
if __name__ == "__main__":
243
    unittest.main()