File size: 15,901 Bytes
cd7807c
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
"""
Unit Tests for inference.py
=============================
Tests LLM interaction, parsing, retry logic, context window, and output formatting.

Run: python test_inference.py [--verbose]
"""

import json
import unittest
from io import StringIO
import sys
from unittest.mock import Mock, patch, MagicMock

# Import functions to test
from inference import (
    parse_llm_response,
    create_fallback_action,
    ContextWindow,
    emit_start,
    emit_step,
    emit_end,
    call_llm,
    run_episode,
    MAX_TOKENS,
    MAX_CONTEXT_TURNS,
)


class TestLLMParsing(unittest.TestCase):
    """Test LLM response parsing with various formats"""
    
    def test_parse_clean_json(self):
        """Test parsing clean JSON response"""
        response = json.dumps({
            "thought": "Check alerts",
            "tool": "aws.soc.get_alerts",
            "args": {}
        })
        
        parsed, error = parse_llm_response(response)
        
        self.assertIsNotNone(parsed)
        self.assertIsNone(error)
        self.assertEqual(parsed["tool"], "aws.soc.get_alerts")
        self.assertEqual(parsed["thought"], "Check alerts")
    
    def test_parse_json_with_markdown_block(self):
        """Test parsing JSON inside markdown code block"""
        response = '''Here's my action:
```json
{
  "thought": "Query logs",
  "tool": "aws.cloudwatch.query_deep",
  "args": {"log_group": "/aws/ec2"}
}
```
Now I'll wait for results.'''
        
        parsed, error = parse_llm_response(response)
        
        self.assertIsNotNone(parsed)
        self.assertIsNone(error)
        self.assertEqual(parsed["tool"], "aws.cloudwatch.query_deep")
    
    def test_parse_json_embedded_in_text(self):
        """Test parsing JSON embedded in explanation text"""
        response = '''I need to check the alerts. My action is:
{
  "thought": "Get current alerts",
  "tool": "aws.soc.get_alerts",
  "args": {}
}
This will show me what's happening.'''
        
        parsed, error = parse_llm_response(response)
        
        self.assertIsNotNone(parsed)
        self.assertIsNone(error)
        self.assertEqual(parsed["tool"], "aws.soc.get_alerts")
    
    def test_parse_json_with_single_quotes(self):
        """Test parsing JSON with single quotes (malformed but recoverable)"""
        response = "{'thought': 'Check alerts', 'tool': 'aws.soc.get_alerts', 'args': {}}"
        
        parsed, error = parse_llm_response(response)
        
        self.assertIsNotNone(parsed)
        self.assertIsNone(error)
        self.assertEqual(parsed["tool"], "aws.soc.get_alerts")
    
    def test_parse_json_with_trailing_comma(self):
        """Test parsing JSON with trailing commas (malformed but recoverable)"""
        response = '''{
  "thought": "Check alerts",
  "tool": "aws.soc.get_alerts",
  "args": {},
}'''
        
        parsed, error = parse_llm_response(response)
        
        self.assertIsNotNone(parsed)
        self.assertIsNone(error)
        self.assertEqual(parsed["tool"], "aws.soc.get_alerts")
    
    def test_parse_missing_tool_field(self):
        """Test handling of JSON missing tool field"""
        response = json.dumps({
            "thought": "Something",
            "args": {}
        })
        
        parsed, error = parse_llm_response(response)
        
        self.assertIsNone(parsed)
        self.assertIsNotNone(error)
        self.assertIn("tool", error.lower())
    
    def test_parse_empty_response(self):
        """Test handling of empty response"""
        parsed, error = parse_llm_response("")
        
        self.assertIsNone(parsed)
        self.assertIsNotNone(error)
    
    def test_parse_completely_invalid_json(self):
        """Test handling of completely invalid JSON"""
        response = "this is not json at all gdfshjkl"
        
        parsed, error = parse_llm_response(response)
        
        self.assertIsNone(parsed)
        self.assertIsNotNone(error)
    
    def test_parse_adds_missing_args(self):
        """Test that missing args dict is added"""
        response = json.dumps({
            "thought": "Act",
            "tool": "aws.soc.get_alerts"
        })
        
        parsed, error = parse_llm_response(response)
        
        self.assertIsNotNone(parsed)
        self.assertIn("args", parsed)
        self.assertEqual(parsed["args"], {})
    
    def test_parse_adds_default_thought(self):
        """Test that missing thought is added"""
        response = json.dumps({
            "tool": "aws.soc.get_alerts",
            "args": {}
        })
        
        parsed, error = parse_llm_response(response)
        
        self.assertIsNotNone(parsed)
        self.assertIn("thought", parsed)


class TestFallbackAction(unittest.TestCase):
    """Test fallback action generation"""
    
    def test_fallback_action_structure(self):
        """Test that fallback action has required fields"""
        action = create_fallback_action(step=1)
        
        self.assertIn("tool", action)
        self.assertIn("args", action)
        self.assertIn("thought", action)
    
    def test_fallback_varies_by_step(self):
        """Test that fallback varies by step to avoid loops"""
        action1 = create_fallback_action(step=1)
        action2 = create_fallback_action(step=2)
        action3 = create_fallback_action(step=3)
        
        tools = [action1["tool"], action2["tool"], action3["tool"]]
        # Should not all be the same
        self.assertGreater(len(set(tools)), 1)
    
    def test_fallback_has_valid_tool(self):
        """Test that fallback action uses valid tools"""
        valid_tools = [
            "aws.soc.get_alerts",
            "aws.guardduty.get_findings",
            "aws.cloudtrail.lookup_events",
            "aws.ec2.describe",
            "aws.iam.describe_role",
        ]
        
        action = create_fallback_action(step=1)
        self.assertIn(action["tool"], valid_tools)


class TestContextWindow(unittest.TestCase):
    """Test context window sliding mechanism"""
    
    def test_context_window_init(self):
        """Test context window initialization"""
        context = ContextWindow(system_prompt="Test prompt", max_turns=3)
        
        self.assertEqual(context.system_prompt, "Test prompt")
        self.assertEqual(context.max_turns, 3)
        self.assertEqual(len(context.turns), 0)
    
    def test_context_window_add_turn(self):
        """Test adding turns to context"""
        context = ContextWindow(max_turns=3)
        
        context.add_turn(
            observation="Obs 1",
            action="Action 1",
            result="Result 1"
        )
        
        self.assertEqual(len(context.turns), 1)
        self.assertEqual(context.turns[0]["observation"], "Obs 1")
    
    def test_context_window_sliding(self):
        """Test that context window slides when exceeding max"""
        context = ContextWindow(max_turns=2)
        
        for i in range(5):
            context.add_turn(
                observation=f"Obs {i}",
                action=f"Action {i}",
                result=f"Result {i}"
            )
        
        # Should only keep last 2 turns
        self.assertEqual(len(context.turns), 2)
        self.assertEqual(context.turns[0]["observation"], "Obs 3")
        self.assertEqual(context.turns[1]["observation"], "Obs 4")
    
    def test_context_get_messages_format(self):
        """Test that get_messages returns proper OpenAI format"""
        context = ContextWindow(system_prompt="System", max_turns=2)
        context.add_turn(
            observation="Obs",
            action='{"tool": "test"}',
            result="Result"
        )
        
        messages = context.get_messages()
        
        # Should have system + user + assistant
        self.assertGreaterEqual(len(messages), 1)
        self.assertEqual(messages[0]["role"], "system")
        self.assertEqual(messages[0]["content"], "System")
    
    def test_context_clear(self):
        """Test clearing context"""
        context = ContextWindow()
        context.add_turn("O", "A", "R")
        
        context.clear()
        
        self.assertEqual(len(context.turns), 0)


class TestOutputFormatting(unittest.TestCase):
    """Test stdout output formatting for hackathon compliance"""
    
    def test_emit_start_format(self):
        """Test [START] line format"""
        output = StringIO()
        with patch('sys.stdout', output):
            emit_start(task_name="easy", env_name="cloudsoc", model_name="TestModel")
        
        result = output.getvalue()
        self.assertIn("[START]", result)
        self.assertIn("task=easy", result)
        self.assertIn("env=cloudsoc", result)
        self.assertIn("model=TestModel", result)
    
    def test_emit_step_format(self):
        """Test [STEP] line format"""
        output = StringIO()
        with patch('sys.stdout', output):
            emit_step(
                step=1,
                action='aws.soc.get_alerts({})',
                reward=0.05,
                done=False,
                error=None
            )
        
        result = output.getvalue()
        self.assertIn("[STEP]", result)
        self.assertIn("step=1", result)
        self.assertIn("reward=0.05", result)
        self.assertIn("done=false", result)
        self.assertIn("error=null", result)
    
    def test_emit_step_with_error(self):
        """Test [STEP] line with error"""
        output = StringIO()
        with patch('sys.stdout', output):
            emit_step(
                step=1,
                action='aws.invalid()',
                reward=-0.02,
                done=False,
                error="VALIDATION_ERROR"
            )
        
        result = output.getvalue()
        self.assertIn("error=VALIDATION_ERROR", result)
    
    def test_emit_end_format(self):
        """Test [END] line format"""
        output = StringIO()
        with patch('sys.stdout', output):
            emit_end(success=True, steps=5, score=0.75, rewards=[0.1, 0.2, 0.15, 0.3, 0.0])
        
        result = output.getvalue()
        self.assertIn("[END]", result)
        self.assertIn("success=true", result)
        self.assertIn("steps=5", result)
        # Score should be between 0.1 and 0.9
        self.assertIn("score=0.750", result)
        score_val = float(result.split('score=')[1].split()[0])
        self.assertGreaterEqual(score_val, 0.1)
        self.assertLessEqual(score_val, 0.9)
        self.assertIn("rewards=0.10,0.20,0.15,0.30,0.00", result)
    
    def test_emit_end_newlines(self):
        """Test that output lines have no embedded newlines"""
        output = StringIO()
        with patch('sys.stdout', output):
            emit_step(
                step=1,
                action='tool({"param": "value"})',
                reward=0.0,
                done=False,
                error=None
            )
        
        result = output.getvalue().strip()
        # Should be single line
        self.assertEqual(result.count('\n'), 0)


class TestCallLLMFallback(unittest.TestCase):
    """Test LLM calling with error handling"""
    
    @patch('inference.client')
    def test_call_llm_success(self, mock_client):
        """Test successful LLM call"""
        mock_response = MagicMock()
        mock_response.choices[0].message.content = '{"tool": "test", "args": {}}'
        mock_client.chat.completions.create.return_value = mock_response
        
        result = call_llm([{"role": "user", "content": "Test"}])
        
        self.assertIsNotNone(result)
        self.assertIn("tool", result)
    
    @patch('inference.client')
    def test_call_llm_timeout_fallback(self, mock_client):
        """Test that LLM timeout returns fallback"""
        mock_client.chat.completions.create.side_effect = TimeoutError("Timeout")
        
        result = call_llm([{"role": "user", "content": "Test"}])
        
        # Should return fallback action string
        self.assertIsNotNone(result)
        self.assertIsInstance(result, str)


class TestMaxConstraints(unittest.TestCase):
    """Test that 2vCPU/8GB constraints are respected"""
    
    def test_max_tokens_set(self):
        """Test that MAX_TOKENS is reasonable for 8GB RAM"""
        # Should be <= 512 for lightweight 3B model
        self.assertLessEqual(MAX_TOKENS, 512)
    
    def test_max_context_turns_set(self):
        """Test that MAX_CONTEXT_TURNS is reasonable for 8GB RAM"""
        # Should be <= 8 to control memory
        self.assertLessEqual(MAX_CONTEXT_TURNS, 8)


class TestEpisodeIntegration(unittest.TestCase):
    """Integration tests for episode execution"""
    
    @patch('inference.CloudSOCEnv')
    @patch('inference.call_llm')
    def test_episode_basic_flow(self, mock_call_llm, mock_env_class):
        """Test basic episode flow with mocked environment"""
        
        # Mock environment
        mock_env = MagicMock()
        mock_env_class.return_value = mock_env
        mock_env.reset.return_value = ("obs", {"action": "reset"})
        mock_env.max_steps = 5
        mock_env.get_system_prompt.return_value = "Test system prompt"
        
        # Mock steps
        mock_env.step.side_effect = [
            ("obs", 0.1, False, False, {"last_action_error": None}),
            ("obs", 0.05, True, False, {"last_action_error": None}),  # Done
        ]
        
        mock_env.get_state_for_next_task.return_value = None
        
        # Mock LLM response
        valid_action = json.dumps({
            "thought": "Test",
            "tool": "aws.soc.get_alerts",
            "args": {}
        })
        mock_call_llm.return_value = valid_action
        
        # Run episode
        output = StringIO()
        with patch('sys.stdout', output):
            success, steps, rewards, _ = run_episode(
                task="easy",
                seed=42,
                verbose=False
            )
        
        # Verify output format
        lines = output.getvalue().strip().split('\n')
        self.assertTrue(lines[0].startswith("[START]"))
        self.assertTrue(any(line.startswith("[STEP]") for line in lines))
        self.assertTrue(lines[-1].startswith("[END]"))
    
    @patch('inference.CloudSOCEnv')
    @patch('inference.call_llm')
    def test_episode_handles_parse_errors(self, mock_call_llm, mock_env_class):
        """Test that episode handles parsing errors gracefully"""
        
        mock_env = MagicMock()
        mock_env_class.return_value = mock_env
        mock_env.reset.return_value = ("obs", {})
        mock_env.max_steps = 3
        mock_env.get_system_prompt.return_value = "Test"
        
        # First call returns invalid JSON (triggers fallback), second succeeds
        mock_call_llm.side_effect = [
            "completely invalid json",
            json.dumps({"thought": "Act", "tool": "aws.soc.get_alerts", "args": {}}),
        ]
        
        mock_env.step.side_effect = [
            ("obs", -0.02, False, False, {"last_action_error": "PARSE_ERROR"}),
            ("obs", 0.1, True, False, {"last_action_error": None}),
        ]
        
        mock_env.get_state_for_next_task.return_value = None
        
        # Episode should complete without crashing
        output = StringIO()
        with patch('sys.stdout', output):
            success, steps, rewards, _ = run_episode(task="easy", verbose=False)
        
        # Should have run 2 steps
        self.assertEqual(steps, 2)


if __name__ == "__main__":
    import sys
    
    # Parse args
    verbose = "--verbose" in sys.argv
    if verbose:
        sys.argv.remove("--verbose")
    
    # Run tests
    suite = unittest.TestLoader().loadTestsFromModule(sys.modules[__name__])
    runner = unittest.TextTestRunner(verbosity=2 if verbose else 1)
    result = runner.run(suite)
    
    sys.exit(0 if result.wasSuccessful() else 1)