Viewing files at commit 085949193d78Commit detailsView latest

Browse files
tests/test_agent_campaign_cli.py 246 lines · 13.0 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
        shutil.rmtree(self.campaign / "fixtures")
77
        shutil.rmtree(self.campaign / "fixture-integrity")
78
        result = self.invoke("resume")
79
        self.assertEqual(result.returncode, 0, result.stdout + result.stderr)
80
        self.assertEqual((self.campaign / "events.jsonl").read_bytes(), journal)
81
        self.assertFalse((self.campaign / "fixtures").exists())
82
        recovered = [json.loads(line) for line in (self.campaign / "rows.jsonl").read_text().splitlines()]
83
        self.assertEqual(recovered, [event["row"] for event in finished])
84
        result = self.invoke("report")
85
        self.assertEqual(result.returncode, 0, result.stdout + result.stderr)
86
        self.assertIn('"claim_eligible": false', result.stdout.lower())
87
88
    def test_corrupted_cache_resume_is_preflight_error_preserving_pending(self):
89
        self.plan()
90
        first = self.invoke("run", "--limit", "1")
91
        self.assertEqual(first.returncode, 3, first.stderr)
92
        journal = (self.campaign / "events.jsonl").read_bytes()
93
        payload = next((self.campaign / "fixtures/wide_config_refactor").glob("services/*/config.toml"))
94
        payload.chmod(0o600)
95
        payload.write_text("changed cached initial state")
96
        result = self.invoke("resume")
97
        self.assertEqual(result.returncode, 2, result.stdout + result.stderr)
98
        self.assertIn("fixture content drift", result.stderr)
99
        self.assertEqual((self.campaign / "events.jsonl").read_bytes(), journal)
100
        status = self.invoke("status")
101
        self.assertEqual(len(json.loads(status.stdout)["pending_trial_ids"]), 1)
102
103
    def test_cache_and_record_rewrite_cannot_rebind_prior_trial_input(self):
104
        import hashlib
105
        self.plan()
106
        first = self.invoke("run", "--limit", "1")
107
        self.assertEqual(first.returncode, 3, first.stderr)
108
        journal = (self.campaign / "events.jsonl").read_bytes()
109
        fixture = self.campaign / "fixtures/wide_config_refactor"
110
        payload = next(fixture.glob("services/*/config.toml"))
111
        payload.chmod(0o600)
112
        payload.write_bytes(b"changed baseline plus forged cache record")
113
        record_path = self.campaign / "fixture-integrity/wide_config_refactor.json"
114
        record = json.loads(record_path.read_text())
115
        entry = record["manifest"][payload.relative_to(fixture).as_posix()]
116
        entry.update(size=len(payload.read_bytes()), sha256=hashlib.sha256(payload.read_bytes()).hexdigest())
117
        record_path.write_text(json.dumps(record))
118
        result = self.invoke("resume")
119
        self.assertEqual(result.returncode, 2, result.stdout + result.stderr)
120
        self.assertIn("journal fixture binding", result.stderr)
121
        self.assertEqual((self.campaign / "events.jsonl").read_bytes(), journal)
122
        status = self.invoke("status")
123
        self.assertEqual(len(json.loads(status.stdout)["pending_trial_ids"]), 1)
124
125
    def test_incomplete_generation_fails_before_any_admission(self):
126
        self.plan()
127
        fixture = self.campaign / "fixtures/wide_config_refactor"
128
        fixture.mkdir(parents=True)
129
        (fixture / "partial").write_text("crashed generation")
130
        result = self.invoke("run")
131
        self.assertEqual(result.returncode, 2, result.stdout + result.stderr)
132
        self.assertIn("fixture cache is incomplete", result.stderr)
133
        self.assertFalse((self.campaign / "events.jsonl").exists())
134
        status = self.invoke("status")
135
        self.assertEqual(len(json.loads(status.stdout)["pending_trial_ids"]), 2)
136
137
    def test_existing_plan_and_changed_config_fail_before_agent_execution(self):
138
        self.plan()
139
        frozen = (self.campaign / "plan.json").read_bytes()
140
        result = self.invoke("plan", "--manifest", str(self.manifest))
141
        self.assertNotEqual(result.returncode, 0)
142
        self.assertEqual((self.campaign / "plan.json").read_bytes(), frozen)
143
        self.agents.write_text(self.agents.read_text() + "\n# changed configuration\n")
144
        result = self.invoke("run")
145
        self.assertNotEqual(result.returncode, 0)
146
        self.assertFalse((self.campaign / "events.jsonl").exists())
147
148
    def test_torn_journal_cannot_be_discarded_to_retry_a_trial(self):
149
        plan = self.plan()
150
        started = {"event": "started", "trial_id": plan["schedule"][0]["trial_id"],
151
                   "plan_sha256": plan["plan_sha256"], "timeout_seconds": 30,
152
                   "ordinal": 0, "timestamp_utc": "2026-09-10T00:00:00Z",
153
                   "fixture_manifest_sha256": self.fixture_hash(plan)}
154
        journal = json.dumps(started).encode() + b"\n" + b'{"event":"finished"'
155
        (self.campaign / "events.jsonl").write_bytes(journal)
156
        result = self.invoke("resume")
157
        self.assertEqual(result.returncode, 2)
158
        self.assertIn("--acknowledge-torn-journal", result.stderr)
159
        self.assertEqual((self.campaign / "events.jsonl").read_bytes(), journal)
160
        self.assertFalse((self.campaign / "trials").exists())
161
162
    def test_insufficient_budget_stops_without_shortening_timeout(self):
163
        manifest = json.loads(self.manifest.read_text())
164
        manifest["max_wall_seconds"] = 0.15
165
        self.manifest.write_text(json.dumps(manifest))
166
        self.plan()
167
        result = self.invoke("run")
168
        self.assertEqual(result.returncode, 3, result.stdout + result.stderr)
169
        state = json.loads(result.stdout)
170
        self.assertEqual(state["stopped_reason"], "insufficient_budget_for_frozen_trial_timeout")
171
        self.assertEqual(len(state["pending_trial_ids"]), 2)
172
        self.assertFalse((self.campaign / "events.jsonl").exists())
173
174
    def test_torn_start_readonly_and_acknowledged_recovery_never_retries(self):
175
        plan = self.plan()
176
        interrupted = plan["schedule"][0]["trial_id"]
177
        raw = b'{"event":"started","trial_id":'
178
        journal = self.campaign / "events.jsonl"
179
        journal.write_bytes(raw)
180
        for command in ("status", "report"):
181
            result = self.invoke(command)
182
            self.assertEqual(result.returncode, 0, result.stdout + result.stderr)
183
            evidence = json.loads(result.stdout)
184
            self.assertFalse(evidence["journal_recoveries"][0]["acknowledged"])
185
            self.assertEqual(journal.read_bytes(), raw)
186
        result = self.invoke("resume", "--acknowledge-torn-journal")
187
        self.assertEqual(result.returncode, 3, result.stdout + result.stderr)
188
        events = self.events()
189
        self.assertEqual(events[0]["trial_id"], interrupted)
190
        self.assertTrue(events[0]["uncertain_torn_start"])
191
        marker = events[1]
192
        self.assertTrue(marker["acknowledged"])
193
        self.assertEqual((self.campaign / marker["archive"]).read_bytes(), raw)
194
        self.assertFalse((self.campaign / "trials" / interrupted).exists())
195
        self.assertEqual(len([e for e in events if e["event"] == "finished"]), 1)
196
        result = self.invoke("resume")
197
        self.assertEqual(result.returncode, 3, result.stdout + result.stderr)
198
        self.assertEqual(self.events(), events)
199
200
    def test_valid_digest_malformed_plan_exits_contract_error(self):
201
        import hashlib
202
        self.campaign.mkdir()
203
        malformed = {"schema_version": 1}
204
        malformed["plan_sha256"] = hashlib.sha256(b'{"schema_version":1}').hexdigest()
205
        (self.campaign / "plan.json").write_text(json.dumps(malformed))
206
        result = self.invoke("status")
207
        self.assertEqual(result.returncode, 2)
208
        self.assertIn("invalid frozen plan", result.stderr)
209
        self.assertNotIn("Traceback", result.stderr)
210
211
    def test_bounded_invocation_resumes_only_remaining_trial(self):
212
        self.plan()
213
        result = self.invoke("run", "--limit", "1")
214
        self.assertEqual(result.returncode, 3, result.stdout + result.stderr)
215
        first = self.events()
216
        self.assertEqual(len(first), 2)
217
        self.assertTrue(first[1]["row"]["success"], first[1])
218
        result = self.invoke("resume", "--limit", "1")
219
        self.assertEqual(result.returncode, 0, result.stdout + result.stderr)
220
        self.assertEqual(self.events()[:2], first)
221
        self.assertEqual(len(self.events()), 4)
222
223
    def test_interrupted_attempt_stays_visible_after_remaining_trial_runs(self):
224
        plan = self.plan()
225
        interrupted = plan["schedule"][0]["trial_id"]
226
        started = {"event": "started", "trial_id": interrupted,
227
                   "plan_sha256": plan["plan_sha256"], "timeout_seconds": 30,
228
                   "ordinal": 0, "timestamp_utc": "2026-09-10T00:00:00Z",
229
                   "fixture_manifest_sha256": self.fixture_hash(plan)}
230
        (self.campaign / "events.jsonl").write_text(json.dumps(started) + "\n")
231
        result = self.invoke("resume")
232
        self.assertEqual(result.returncode, 3, result.stdout + result.stderr)
233
        events = self.events()
234
        self.assertEqual(len(events), 3)
235
        self.assertEqual(events[0], started)
236
        self.assertNotEqual(events[-1]["trial_id"], interrupted)
237
        self.assertTrue(events[-1]["row"]["success"], events[-1])
238
        self.assertFalse((self.campaign / "trials" / interrupted).exists())
239
        result = self.invoke("report", "--format", "json")
240
        self.assertEqual(result.returncode, 0, result.stdout + result.stderr)
241
        counts = json.loads(result.stdout)["counts"]
242
        self.assertEqual((counts["started"], counts["success"], counts["interrupted"]), (2, 1, 1))
243
244
245
if __name__ == "__main__":
246
    unittest.main()