File size: 66,513 Bytes
0b65952
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
[
  {
    "name": "inspect_evals/assistant_bench_closed_book_one_shot - Sample fb9ba3ab6a13d0adc677f993e90d54914a5cdf211305a1bba6bf16ec4ccb9b7c",
    "description": "Task from inspect_evals/assistant_bench_closed_book_one_shot dataset, sample fb9ba3ab6a13d0adc677f993e90d54914a5cdf211305a1bba6bf16ec4ccb9b7c",
    "prompt": "Which member of the famous Beckham couple has the most Instagram followers?\n\nOn which social media platform does Andrew Ng have the most followers?",
    "mcp_config": {
      "inspect": {
        "command": "docker",
        "args": [
          "run",
          "--rm",
          "-i",
          "--init",
          "-v",
          "/var/run/docker.sock:/var/run/docker.sock",
          "-v",
          "/home/genteki/.cache/inspect_evals:/root/.cache/inspect_evals",
          "hud-inspect:latest"
        ]
      }
    },
    "setup_tool": {
      "name": "load_task",
      "arguments": {
        "task_name": "inspect_evals/assistant_bench_closed_book_one_shot",
        "sample_id": "fb9ba3ab6a13d0adc677f993e90d54914a5cdf211305a1bba6bf16ec4ccb9b7c",
        "model": "claude-sonnet-4-20250514"
      }
    },
    "evaluate_tool": {
      "name": "score_task",
      "arguments": {}
    },
    "agent_config": {
      "agent_class": "server.agents.InspectAgent",
      "disallowed_tools": [
        "_store_agent_message",
        "load_task",
        "score_task"
      ],
      "allowed_tools": [],
      "initial_screenshot": false
    },
    "system_prompt": "You are a knowledgeable AI assistant tasked with providing concise answers to a wide range of questions.  Always provide an answer, even if you're not entirely certain. If you're unsure, make an educated guess based on your knowledge.\n\nFollow these guidelines:\n1. Don't ask the user any questions.\n2. Structure your response as a series of intermediate questions and answers, followed by a final answer.\n3. Do not generate the same intermediate question twice.\n4. Keep the final answer concise. It should be either: a number, a string, a list of strings, or a list of JSONs.\n5. Ensure the answer can be parsed with the Python method: json.loads(input_str).\n6. If you truly can't provide any answer, use an empty string as the final answer.\n\nThis is a closed-book task, so rely on your existing knowledge without asking for additional information.\n\nUse the format:\nIntermediate question: [Your question]\nIntermediate answer: [Your answer]\n(Repeat as needed)\nThe final answer is: [Your concise final answer]\n\nRemember, it's better to provide a potentially outdated or slightly inaccurate answer than no answer at all. Your goal is to maximize the answer rate and accuracy to the best of your ability. Always provide at least one intermediate question and answer before giving the final answer."
  },
  {
    "name": "inspect_evals/assistant_bench_closed_book_one_shot - Sample 929b45f34805280d77c61d1e093e3d4e551d77ddb6ecd73552b12b1af286388d",
    "description": "Task from inspect_evals/assistant_bench_closed_book_one_shot dataset, sample 929b45f34805280d77c61d1e093e3d4e551d77ddb6ecd73552b12b1af286388d",
    "prompt": "Which member of the famous Beckham couple has the most Instagram followers?\n\nThe dog genome was first mapped in 2004 and has been updated several times since. What is the link to the files that were most relevant in May 2020?",
    "mcp_config": {
      "inspect": {
        "command": "docker",
        "args": [
          "run",
          "--rm",
          "-i",
          "--init",
          "-v",
          "/var/run/docker.sock:/var/run/docker.sock",
          "-v",
          "/home/genteki/.cache/inspect_evals:/root/.cache/inspect_evals",
          "hud-inspect:latest"
        ]
      }
    },
    "setup_tool": {
      "name": "load_task",
      "arguments": {
        "task_name": "inspect_evals/assistant_bench_closed_book_one_shot",
        "sample_id": "929b45f34805280d77c61d1e093e3d4e551d77ddb6ecd73552b12b1af286388d",
        "model": "claude-sonnet-4-20250514"
      }
    },
    "evaluate_tool": {
      "name": "score_task",
      "arguments": {}
    },
    "agent_config": {
      "agent_class": "server.agents.InspectAgent",
      "disallowed_tools": [
        "_store_agent_message",
        "load_task",
        "score_task"
      ],
      "allowed_tools": [],
      "initial_screenshot": false
    },
    "system_prompt": "You are a knowledgeable AI assistant tasked with providing concise answers to a wide range of questions.  Always provide an answer, even if you're not entirely certain. If you're unsure, make an educated guess based on your knowledge.\n\nFollow these guidelines:\n1. Don't ask the user any questions.\n2. Structure your response as a series of intermediate questions and answers, followed by a final answer.\n3. Do not generate the same intermediate question twice.\n4. Keep the final answer concise. It should be either: a number, a string, a list of strings, or a list of JSONs.\n5. Ensure the answer can be parsed with the Python method: json.loads(input_str).\n6. If you truly can't provide any answer, use an empty string as the final answer.\n\nThis is a closed-book task, so rely on your existing knowledge without asking for additional information.\n\nUse the format:\nIntermediate question: [Your question]\nIntermediate answer: [Your answer]\n(Repeat as needed)\nThe final answer is: [Your concise final answer]\n\nRemember, it's better to provide a potentially outdated or slightly inaccurate answer than no answer at all. Your goal is to maximize the answer rate and accuracy to the best of your ability. Always provide at least one intermediate question and answer before giving the final answer."
  },
  {
    "name": "inspect_evals/assistant_bench_closed_book_one_shot - Sample 9baaa267c95f9d8b75741ee9169c50563d297cfa592c20deaffd30dbc5984c74",
    "description": "Task from inspect_evals/assistant_bench_closed_book_one_shot dataset, sample 9baaa267c95f9d8b75741ee9169c50563d297cfa592c20deaffd30dbc5984c74",
    "prompt": "Which member of the famous Beckham couple has the most Instagram followers?\n\nBased on recent years (2020-2023), how likely am I to hit a day with a max temperature (in Fahrenheit) over 95 if I travel to Houston, Texas during June? (provide the answer in percentage)",
    "mcp_config": {
      "inspect": {
        "command": "docker",
        "args": [
          "run",
          "--rm",
          "-i",
          "--init",
          "-v",
          "/var/run/docker.sock:/var/run/docker.sock",
          "-v",
          "/home/genteki/.cache/inspect_evals:/root/.cache/inspect_evals",
          "hud-inspect:latest"
        ]
      }
    },
    "setup_tool": {
      "name": "load_task",
      "arguments": {
        "task_name": "inspect_evals/assistant_bench_closed_book_one_shot",
        "sample_id": "9baaa267c95f9d8b75741ee9169c50563d297cfa592c20deaffd30dbc5984c74",
        "model": "claude-sonnet-4-20250514"
      }
    },
    "evaluate_tool": {
      "name": "score_task",
      "arguments": {}
    },
    "agent_config": {
      "agent_class": "server.agents.InspectAgent",
      "disallowed_tools": [
        "_store_agent_message",
        "load_task",
        "score_task"
      ],
      "allowed_tools": [],
      "initial_screenshot": false
    },
    "system_prompt": "You are a knowledgeable AI assistant tasked with providing concise answers to a wide range of questions.  Always provide an answer, even if you're not entirely certain. If you're unsure, make an educated guess based on your knowledge.\n\nFollow these guidelines:\n1. Don't ask the user any questions.\n2. Structure your response as a series of intermediate questions and answers, followed by a final answer.\n3. Do not generate the same intermediate question twice.\n4. Keep the final answer concise. It should be either: a number, a string, a list of strings, or a list of JSONs.\n5. Ensure the answer can be parsed with the Python method: json.loads(input_str).\n6. If you truly can't provide any answer, use an empty string as the final answer.\n\nThis is a closed-book task, so rely on your existing knowledge without asking for additional information.\n\nUse the format:\nIntermediate question: [Your question]\nIntermediate answer: [Your answer]\n(Repeat as needed)\nThe final answer is: [Your concise final answer]\n\nRemember, it's better to provide a potentially outdated or slightly inaccurate answer than no answer at all. Your goal is to maximize the answer rate and accuracy to the best of your ability. Always provide at least one intermediate question and answer before giving the final answer."
  },
  {
    "name": "inspect_evals/assistant_bench_closed_book_one_shot - Sample e6bc98089608217e45b6956a46518fe3cce64a799b3ac43c6974c449ae14c408",
    "description": "Task from inspect_evals/assistant_bench_closed_book_one_shot dataset, sample e6bc98089608217e45b6956a46518fe3cce64a799b3ac43c6974c449ae14c408",
    "prompt": "Which member of the famous Beckham couple has the most Instagram followers?\n\nWhat's the highest price a high-rise apartment was sold for in Mission Bay, San Francisco, in 2021?",
    "mcp_config": {
      "inspect": {
        "command": "docker",
        "args": [
          "run",
          "--rm",
          "-i",
          "--init",
          "-v",
          "/var/run/docker.sock:/var/run/docker.sock",
          "-v",
          "/home/genteki/.cache/inspect_evals:/root/.cache/inspect_evals",
          "hud-inspect:latest"
        ]
      }
    },
    "setup_tool": {
      "name": "load_task",
      "arguments": {
        "task_name": "inspect_evals/assistant_bench_closed_book_one_shot",
        "sample_id": "e6bc98089608217e45b6956a46518fe3cce64a799b3ac43c6974c449ae14c408",
        "model": "claude-sonnet-4-20250514"
      }
    },
    "evaluate_tool": {
      "name": "score_task",
      "arguments": {}
    },
    "agent_config": {
      "agent_class": "server.agents.InspectAgent",
      "disallowed_tools": [
        "_store_agent_message",
        "load_task",
        "score_task"
      ],
      "allowed_tools": [],
      "initial_screenshot": false
    },
    "system_prompt": "You are a knowledgeable AI assistant tasked with providing concise answers to a wide range of questions.  Always provide an answer, even if you're not entirely certain. If you're unsure, make an educated guess based on your knowledge.\n\nFollow these guidelines:\n1. Don't ask the user any questions.\n2. Structure your response as a series of intermediate questions and answers, followed by a final answer.\n3. Do not generate the same intermediate question twice.\n4. Keep the final answer concise. It should be either: a number, a string, a list of strings, or a list of JSONs.\n5. Ensure the answer can be parsed with the Python method: json.loads(input_str).\n6. If you truly can't provide any answer, use an empty string as the final answer.\n\nThis is a closed-book task, so rely on your existing knowledge without asking for additional information.\n\nUse the format:\nIntermediate question: [Your question]\nIntermediate answer: [Your answer]\n(Repeat as needed)\nThe final answer is: [Your concise final answer]\n\nRemember, it's better to provide a potentially outdated or slightly inaccurate answer than no answer at all. Your goal is to maximize the answer rate and accuracy to the best of your ability. Always provide at least one intermediate question and answer before giving the final answer."
  },
  {
    "name": "inspect_evals/assistant_bench_closed_book_one_shot - Sample 291b53e665b4dd4365cde995042db4a6f6fecef3fe3a6f4482f23d61bd673918",
    "description": "Task from inspect_evals/assistant_bench_closed_book_one_shot dataset, sample 291b53e665b4dd4365cde995042db4a6f6fecef3fe3a6f4482f23d61bd673918",
    "prompt": "Which member of the famous Beckham couple has the most Instagram followers?\n\nWhat is the link to the GFF3 file for beluga whales that was the most recent one on 20/10/2020?",
    "mcp_config": {
      "inspect": {
        "command": "docker",
        "args": [
          "run",
          "--rm",
          "-i",
          "--init",
          "-v",
          "/var/run/docker.sock:/var/run/docker.sock",
          "-v",
          "/home/genteki/.cache/inspect_evals:/root/.cache/inspect_evals",
          "hud-inspect:latest"
        ]
      }
    },
    "setup_tool": {
      "name": "load_task",
      "arguments": {
        "task_name": "inspect_evals/assistant_bench_closed_book_one_shot",
        "sample_id": "291b53e665b4dd4365cde995042db4a6f6fecef3fe3a6f4482f23d61bd673918",
        "model": "claude-sonnet-4-20250514"
      }
    },
    "evaluate_tool": {
      "name": "score_task",
      "arguments": {}
    },
    "agent_config": {
      "agent_class": "server.agents.InspectAgent",
      "disallowed_tools": [
        "_store_agent_message",
        "load_task",
        "score_task"
      ],
      "allowed_tools": [],
      "initial_screenshot": false
    },
    "system_prompt": "You are a knowledgeable AI assistant tasked with providing concise answers to a wide range of questions.  Always provide an answer, even if you're not entirely certain. If you're unsure, make an educated guess based on your knowledge.\n\nFollow these guidelines:\n1. Don't ask the user any questions.\n2. Structure your response as a series of intermediate questions and answers, followed by a final answer.\n3. Do not generate the same intermediate question twice.\n4. Keep the final answer concise. It should be either: a number, a string, a list of strings, or a list of JSONs.\n5. Ensure the answer can be parsed with the Python method: json.loads(input_str).\n6. If you truly can't provide any answer, use an empty string as the final answer.\n\nThis is a closed-book task, so rely on your existing knowledge without asking for additional information.\n\nUse the format:\nIntermediate question: [Your question]\nIntermediate answer: [Your answer]\n(Repeat as needed)\nThe final answer is: [Your concise final answer]\n\nRemember, it's better to provide a potentially outdated or slightly inaccurate answer than no answer at all. Your goal is to maximize the answer rate and accuracy to the best of your ability. Always provide at least one intermediate question and answer before giving the final answer."
  },
  {
    "name": "inspect_evals/assistant_bench_closed_book_one_shot - Sample 4dbedc5e1a0205e14b7ff3ba89bce3060dab15d0ada3b7e1351a6f2aa8287aec",
    "description": "Task from inspect_evals/assistant_bench_closed_book_one_shot dataset, sample 4dbedc5e1a0205e14b7ff3ba89bce3060dab15d0ada3b7e1351a6f2aa8287aec",
    "prompt": "Which member of the famous Beckham couple has the most Instagram followers?\n\nHow much did I save by purchasing a season pass instead of daily tickets for California's Great America in San Jose, if I planned to visit once a month in June, July, August, and September during the summer of 2024?",
    "mcp_config": {
      "inspect": {
        "command": "docker",
        "args": [
          "run",
          "--rm",
          "-i",
          "--init",
          "-v",
          "/var/run/docker.sock:/var/run/docker.sock",
          "-v",
          "/home/genteki/.cache/inspect_evals:/root/.cache/inspect_evals",
          "hud-inspect:latest"
        ]
      }
    },
    "setup_tool": {
      "name": "load_task",
      "arguments": {
        "task_name": "inspect_evals/assistant_bench_closed_book_one_shot",
        "sample_id": "4dbedc5e1a0205e14b7ff3ba89bce3060dab15d0ada3b7e1351a6f2aa8287aec",
        "model": "claude-sonnet-4-20250514"
      }
    },
    "evaluate_tool": {
      "name": "score_task",
      "arguments": {}
    },
    "agent_config": {
      "agent_class": "server.agents.InspectAgent",
      "disallowed_tools": [
        "_store_agent_message",
        "load_task",
        "score_task"
      ],
      "allowed_tools": [],
      "initial_screenshot": false
    },
    "system_prompt": "You are a knowledgeable AI assistant tasked with providing concise answers to a wide range of questions.  Always provide an answer, even if you're not entirely certain. If you're unsure, make an educated guess based on your knowledge.\n\nFollow these guidelines:\n1. Don't ask the user any questions.\n2. Structure your response as a series of intermediate questions and answers, followed by a final answer.\n3. Do not generate the same intermediate question twice.\n4. Keep the final answer concise. It should be either: a number, a string, a list of strings, or a list of JSONs.\n5. Ensure the answer can be parsed with the Python method: json.loads(input_str).\n6. If you truly can't provide any answer, use an empty string as the final answer.\n\nThis is a closed-book task, so rely on your existing knowledge without asking for additional information.\n\nUse the format:\nIntermediate question: [Your question]\nIntermediate answer: [Your answer]\n(Repeat as needed)\nThe final answer is: [Your concise final answer]\n\nRemember, it's better to provide a potentially outdated or slightly inaccurate answer than no answer at all. Your goal is to maximize the answer rate and accuracy to the best of your ability. Always provide at least one intermediate question and answer before giving the final answer."
  },
  {
    "name": "inspect_evals/assistant_bench_closed_book_one_shot - Sample 55f4258484c5b398956133128a50462a767da211f8f72aa5ac5bbffb9bcbba1a",
    "description": "Task from inspect_evals/assistant_bench_closed_book_one_shot dataset, sample 55f4258484c5b398956133128a50462a767da211f8f72aa5ac5bbffb9bcbba1a",
    "prompt": "Which member of the famous Beckham couple has the most Instagram followers?\n\nWhat is the worst rated series (according to Rotten Tomatoes) with more than 1 season that Ted Danson has starred in and is available on Amazon Prime Video (US)?",
    "mcp_config": {
      "inspect": {
        "command": "docker",
        "args": [
          "run",
          "--rm",
          "-i",
          "--init",
          "-v",
          "/var/run/docker.sock:/var/run/docker.sock",
          "-v",
          "/home/genteki/.cache/inspect_evals:/root/.cache/inspect_evals",
          "hud-inspect:latest"
        ]
      }
    },
    "setup_tool": {
      "name": "load_task",
      "arguments": {
        "task_name": "inspect_evals/assistant_bench_closed_book_one_shot",
        "sample_id": "55f4258484c5b398956133128a50462a767da211f8f72aa5ac5bbffb9bcbba1a",
        "model": "claude-sonnet-4-20250514"
      }
    },
    "evaluate_tool": {
      "name": "score_task",
      "arguments": {}
    },
    "agent_config": {
      "agent_class": "server.agents.InspectAgent",
      "disallowed_tools": [
        "_store_agent_message",
        "load_task",
        "score_task"
      ],
      "allowed_tools": [],
      "initial_screenshot": false
    },
    "system_prompt": "You are a knowledgeable AI assistant tasked with providing concise answers to a wide range of questions.  Always provide an answer, even if you're not entirely certain. If you're unsure, make an educated guess based on your knowledge.\n\nFollow these guidelines:\n1. Don't ask the user any questions.\n2. Structure your response as a series of intermediate questions and answers, followed by a final answer.\n3. Do not generate the same intermediate question twice.\n4. Keep the final answer concise. It should be either: a number, a string, a list of strings, or a list of JSONs.\n5. Ensure the answer can be parsed with the Python method: json.loads(input_str).\n6. If you truly can't provide any answer, use an empty string as the final answer.\n\nThis is a closed-book task, so rely on your existing knowledge without asking for additional information.\n\nUse the format:\nIntermediate question: [Your question]\nIntermediate answer: [Your answer]\n(Repeat as needed)\nThe final answer is: [Your concise final answer]\n\nRemember, it's better to provide a potentially outdated or slightly inaccurate answer than no answer at all. Your goal is to maximize the answer rate and accuracy to the best of your ability. Always provide at least one intermediate question and answer before giving the final answer."
  },
  {
    "name": "inspect_evals/assistant_bench_closed_book_one_shot - Sample 52f7224e9c79431e7926afe317782711a0028750693e7456cde22ef6f4bd8bd5",
    "description": "Task from inspect_evals/assistant_bench_closed_book_one_shot dataset, sample 52f7224e9c79431e7926afe317782711a0028750693e7456cde22ef6f4bd8bd5",
    "prompt": "Which member of the famous Beckham couple has the most Instagram followers?\n\nWhat is the highest rated (according to IMDB) Isabelle Adjani feature film that is less than 2 hours and is available on Vudu (now called Fandango at Home) to buy or rent?",
    "mcp_config": {
      "inspect": {
        "command": "docker",
        "args": [
          "run",
          "--rm",
          "-i",
          "--init",
          "-v",
          "/var/run/docker.sock:/var/run/docker.sock",
          "-v",
          "/home/genteki/.cache/inspect_evals:/root/.cache/inspect_evals",
          "hud-inspect:latest"
        ]
      }
    },
    "setup_tool": {
      "name": "load_task",
      "arguments": {
        "task_name": "inspect_evals/assistant_bench_closed_book_one_shot",
        "sample_id": "52f7224e9c79431e7926afe317782711a0028750693e7456cde22ef6f4bd8bd5",
        "model": "claude-sonnet-4-20250514"
      }
    },
    "evaluate_tool": {
      "name": "score_task",
      "arguments": {}
    },
    "agent_config": {
      "agent_class": "server.agents.InspectAgent",
      "disallowed_tools": [
        "_store_agent_message",
        "load_task",
        "score_task"
      ],
      "allowed_tools": [],
      "initial_screenshot": false
    },
    "system_prompt": "You are a knowledgeable AI assistant tasked with providing concise answers to a wide range of questions.  Always provide an answer, even if you're not entirely certain. If you're unsure, make an educated guess based on your knowledge.\n\nFollow these guidelines:\n1. Don't ask the user any questions.\n2. Structure your response as a series of intermediate questions and answers, followed by a final answer.\n3. Do not generate the same intermediate question twice.\n4. Keep the final answer concise. It should be either: a number, a string, a list of strings, or a list of JSONs.\n5. Ensure the answer can be parsed with the Python method: json.loads(input_str).\n6. If you truly can't provide any answer, use an empty string as the final answer.\n\nThis is a closed-book task, so rely on your existing knowledge without asking for additional information.\n\nUse the format:\nIntermediate question: [Your question]\nIntermediate answer: [Your answer]\n(Repeat as needed)\nThe final answer is: [Your concise final answer]\n\nRemember, it's better to provide a potentially outdated or slightly inaccurate answer than no answer at all. Your goal is to maximize the answer rate and accuracy to the best of your ability. Always provide at least one intermediate question and answer before giving the final answer."
  },
  {
    "name": "inspect_evals/assistant_bench_closed_book_one_shot - Sample 557e78eceec08ca8b0da5f9fdaca6e1c7ec6140a8ce600983ee716327dab005e",
    "description": "Task from inspect_evals/assistant_bench_closed_book_one_shot dataset, sample 557e78eceec08ca8b0da5f9fdaca6e1c7ec6140a8ce600983ee716327dab005e",
    "prompt": "Which member of the famous Beckham couple has the most Instagram followers?\n\nWhich bar is closest to Mummers Museum in Philadelphia and is wheel chair accessible?",
    "mcp_config": {
      "inspect": {
        "command": "docker",
        "args": [
          "run",
          "--rm",
          "-i",
          "--init",
          "-v",
          "/var/run/docker.sock:/var/run/docker.sock",
          "-v",
          "/home/genteki/.cache/inspect_evals:/root/.cache/inspect_evals",
          "hud-inspect:latest"
        ]
      }
    },
    "setup_tool": {
      "name": "load_task",
      "arguments": {
        "task_name": "inspect_evals/assistant_bench_closed_book_one_shot",
        "sample_id": "557e78eceec08ca8b0da5f9fdaca6e1c7ec6140a8ce600983ee716327dab005e",
        "model": "claude-sonnet-4-20250514"
      }
    },
    "evaluate_tool": {
      "name": "score_task",
      "arguments": {}
    },
    "agent_config": {
      "agent_class": "server.agents.InspectAgent",
      "disallowed_tools": [
        "_store_agent_message",
        "load_task",
        "score_task"
      ],
      "allowed_tools": [],
      "initial_screenshot": false
    },
    "system_prompt": "You are a knowledgeable AI assistant tasked with providing concise answers to a wide range of questions.  Always provide an answer, even if you're not entirely certain. If you're unsure, make an educated guess based on your knowledge.\n\nFollow these guidelines:\n1. Don't ask the user any questions.\n2. Structure your response as a series of intermediate questions and answers, followed by a final answer.\n3. Do not generate the same intermediate question twice.\n4. Keep the final answer concise. It should be either: a number, a string, a list of strings, or a list of JSONs.\n5. Ensure the answer can be parsed with the Python method: json.loads(input_str).\n6. If you truly can't provide any answer, use an empty string as the final answer.\n\nThis is a closed-book task, so rely on your existing knowledge without asking for additional information.\n\nUse the format:\nIntermediate question: [Your question]\nIntermediate answer: [Your answer]\n(Repeat as needed)\nThe final answer is: [Your concise final answer]\n\nRemember, it's better to provide a potentially outdated or slightly inaccurate answer than no answer at all. Your goal is to maximize the answer rate and accuracy to the best of your ability. Always provide at least one intermediate question and answer before giving the final answer."
  },
  {
    "name": "inspect_evals/assistant_bench_closed_book_one_shot - Sample 797f7a5b65ca28b7e7156e7db1e9f117bd4a021de0cd512bfdbb0be897d89eab",
    "description": "Task from inspect_evals/assistant_bench_closed_book_one_shot dataset, sample 797f7a5b65ca28b7e7156e7db1e9f117bd4a021de0cd512bfdbb0be897d89eab",
    "prompt": "Which member of the famous Beckham couple has the most Instagram followers?\n\nWhich restaurants (not takeaway only) within 1 blocks of Washington Square Park have vegan mains for under $15?",
    "mcp_config": {
      "inspect": {
        "command": "docker",
        "args": [
          "run",
          "--rm",
          "-i",
          "--init",
          "-v",
          "/var/run/docker.sock:/var/run/docker.sock",
          "-v",
          "/home/genteki/.cache/inspect_evals:/root/.cache/inspect_evals",
          "hud-inspect:latest"
        ]
      }
    },
    "setup_tool": {
      "name": "load_task",
      "arguments": {
        "task_name": "inspect_evals/assistant_bench_closed_book_one_shot",
        "sample_id": "797f7a5b65ca28b7e7156e7db1e9f117bd4a021de0cd512bfdbb0be897d89eab",
        "model": "claude-sonnet-4-20250514"
      }
    },
    "evaluate_tool": {
      "name": "score_task",
      "arguments": {}
    },
    "agent_config": {
      "agent_class": "server.agents.InspectAgent",
      "disallowed_tools": [
        "_store_agent_message",
        "load_task",
        "score_task"
      ],
      "allowed_tools": [],
      "initial_screenshot": false
    },
    "system_prompt": "You are a knowledgeable AI assistant tasked with providing concise answers to a wide range of questions.  Always provide an answer, even if you're not entirely certain. If you're unsure, make an educated guess based on your knowledge.\n\nFollow these guidelines:\n1. Don't ask the user any questions.\n2. Structure your response as a series of intermediate questions and answers, followed by a final answer.\n3. Do not generate the same intermediate question twice.\n4. Keep the final answer concise. It should be either: a number, a string, a list of strings, or a list of JSONs.\n5. Ensure the answer can be parsed with the Python method: json.loads(input_str).\n6. If you truly can't provide any answer, use an empty string as the final answer.\n\nThis is a closed-book task, so rely on your existing knowledge without asking for additional information.\n\nUse the format:\nIntermediate question: [Your question]\nIntermediate answer: [Your answer]\n(Repeat as needed)\nThe final answer is: [Your concise final answer]\n\nRemember, it's better to provide a potentially outdated or slightly inaccurate answer than no answer at all. Your goal is to maximize the answer rate and accuracy to the best of your ability. Always provide at least one intermediate question and answer before giving the final answer."
  },
  {
    "name": "inspect_evals/assistant_bench_web_search_zero_shot - Sample 3af8028c2a59e28ca88baff0e6d91f2a9f170c5ef91003f1c8406755a2760ad4",
    "description": "Task from inspect_evals/assistant_bench_web_search_zero_shot dataset, sample 3af8028c2a59e28ca88baff0e6d91f2a9f170c5ef91003f1c8406755a2760ad4",
    "prompt": "Which Magic: The Gathering Standard card (non-foil paper version released in its original set) that was banned at the same time as Oko, Thief of Crowns (including Oko, Crown of Thorns) had the highest price decrease from its all-time high to its all-time low?",
    "mcp_config": {
      "inspect": {
        "command": "docker",
        "args": [
          "run",
          "--rm",
          "-i",
          "--init",
          "-v",
          "/var/run/docker.sock:/var/run/docker.sock",
          "-v",
          "/home/genteki/.cache/inspect_evals:/root/.cache/inspect_evals",
          "hud-inspect:latest"
        ]
      }
    },
    "setup_tool": {
      "name": "load_task",
      "arguments": {
        "task_name": "inspect_evals/assistant_bench_web_search_zero_shot",
        "sample_id": "3af8028c2a59e28ca88baff0e6d91f2a9f170c5ef91003f1c8406755a2760ad4",
        "model": "claude-sonnet-4-20250514"
      }
    },
    "evaluate_tool": {
      "name": "score_task",
      "arguments": {}
    },
    "agent_config": {
      "agent_class": "server.agents.InspectAgent",
      "disallowed_tools": [
        "_store_agent_message",
        "load_task",
        "score_task"
      ],
      "allowed_tools": [],
      "initial_screenshot": false
    },
    "system_prompt": "You are a helpful agent that has to provide answers to complex tasks.\n\nTasks can require accessing the web. To access the web, you are equipped with a search engine.\nTo access the search engine, simply generate an intermediate question. The intermediate question will be sent to a search engine (Google) and the result from the search engine will be provided by the user in the next turn.\n\nIt is best to access the search engine with simple queries, as they are most likely to return the correct answers. For example, if you require to find the number of Instagram followers for a group, generate a query for each member of the group.\n\nWhen the user message ends with an intermediate question, always generate an intermediate answer.\n\nIf the search results do not answer your intermediate question, begin the intermediate answer with \"Intermediate answer: The intermediate answer cannot be inferred from the search results\".\nIn that case, you can use your own knowledge to answer the intermediate question or generate a new intermediate question.\n\nYou can generate additional intermediate questions and answers until arriving at the final answer. Do not generate the same intermediate question twice.\n\nWhen arriving at the final answer, start your prediction with: \"The final answer is:\" followed by the answer to the task.\nIf you are unsure what the final answer is, generate the empty string (\"\").\n\nKeep the answer as concise as possible. The answer should either be: a number, a string, a list of strings, or a list of jsons.\nThe answer should be parsed with the python method: json.loads(input_str). If no answer is found, generate an empty string.\n\nThe format will be:\nAssistant: Intermediate question: [Your search engine query]\nUser:      Context0: [Search engine results]\n           Task: [Original task]\n           Intermediate question: [Your most recent search engine query]\nAssistant: Intermediate answer: [Your answer as inferred from the search results or your own knowledge]\n           Intermediate question: [Your next search engine query]\n(Repeat as needed)\nAssistant: Intermediate answer: [Your answer as inferred from the search results or your own knowledge]\n           The final answer is: [Your next search engine query]"
  },
  {
    "name": "inspect_evals/assistant_bench_web_search_zero_shot - Sample 4dbedc5e1a0205e14b7ff3ba89bce3060dab15d0ada3b7e1351a6f2aa8287aec",
    "description": "Task from inspect_evals/assistant_bench_web_search_zero_shot dataset, sample 4dbedc5e1a0205e14b7ff3ba89bce3060dab15d0ada3b7e1351a6f2aa8287aec",
    "prompt": "How much did I save by purchasing a season pass instead of daily tickets for California's Great America in San Jose, if I planned to visit once a month in June, July, August, and September during the summer of 2024?",
    "mcp_config": {
      "inspect": {
        "command": "docker",
        "args": [
          "run",
          "--rm",
          "-i",
          "--init",
          "-v",
          "/var/run/docker.sock:/var/run/docker.sock",
          "-v",
          "/home/genteki/.cache/inspect_evals:/root/.cache/inspect_evals",
          "hud-inspect:latest"
        ]
      }
    },
    "setup_tool": {
      "name": "load_task",
      "arguments": {
        "task_name": "inspect_evals/assistant_bench_web_search_zero_shot",
        "sample_id": "4dbedc5e1a0205e14b7ff3ba89bce3060dab15d0ada3b7e1351a6f2aa8287aec",
        "model": "claude-sonnet-4-20250514"
      }
    },
    "evaluate_tool": {
      "name": "score_task",
      "arguments": {}
    },
    "agent_config": {
      "agent_class": "server.agents.InspectAgent",
      "disallowed_tools": [
        "_store_agent_message",
        "load_task",
        "score_task"
      ],
      "allowed_tools": [],
      "initial_screenshot": false
    },
    "system_prompt": "You are a helpful agent that has to provide answers to complex tasks.\n\nTasks can require accessing the web. To access the web, you are equipped with a search engine.\nTo access the search engine, simply generate an intermediate question. The intermediate question will be sent to a search engine (Google) and the result from the search engine will be provided by the user in the next turn.\n\nIt is best to access the search engine with simple queries, as they are most likely to return the correct answers. For example, if you require to find the number of Instagram followers for a group, generate a query for each member of the group.\n\nWhen the user message ends with an intermediate question, always generate an intermediate answer.\n\nIf the search results do not answer your intermediate question, begin the intermediate answer with \"Intermediate answer: The intermediate answer cannot be inferred from the search results\".\nIn that case, you can use your own knowledge to answer the intermediate question or generate a new intermediate question.\n\nYou can generate additional intermediate questions and answers until arriving at the final answer. Do not generate the same intermediate question twice.\n\nWhen arriving at the final answer, start your prediction with: \"The final answer is:\" followed by the answer to the task.\nIf you are unsure what the final answer is, generate the empty string (\"\").\n\nKeep the answer as concise as possible. The answer should either be: a number, a string, a list of strings, or a list of jsons.\nThe answer should be parsed with the python method: json.loads(input_str). If no answer is found, generate an empty string.\n\nThe format will be:\nAssistant: Intermediate question: [Your search engine query]\nUser:      Context0: [Search engine results]\n           Task: [Original task]\n           Intermediate question: [Your most recent search engine query]\nAssistant: Intermediate answer: [Your answer as inferred from the search results or your own knowledge]\n           Intermediate question: [Your next search engine query]\n(Repeat as needed)\nAssistant: Intermediate answer: [Your answer as inferred from the search results or your own knowledge]\n           The final answer is: [Your next search engine query]"
  },
  {
    "name": "inspect_evals/assistant_bench_web_search_zero_shot - Sample 55f4258484c5b398956133128a50462a767da211f8f72aa5ac5bbffb9bcbba1a",
    "description": "Task from inspect_evals/assistant_bench_web_search_zero_shot dataset, sample 55f4258484c5b398956133128a50462a767da211f8f72aa5ac5bbffb9bcbba1a",
    "prompt": "What is the worst rated series (according to Rotten Tomatoes) with more than 1 season that Ted Danson has starred in and is available on Amazon Prime Video (US)?",
    "mcp_config": {
      "inspect": {
        "command": "docker",
        "args": [
          "run",
          "--rm",
          "-i",
          "--init",
          "-v",
          "/var/run/docker.sock:/var/run/docker.sock",
          "-v",
          "/home/genteki/.cache/inspect_evals:/root/.cache/inspect_evals",
          "hud-inspect:latest"
        ]
      }
    },
    "setup_tool": {
      "name": "load_task",
      "arguments": {
        "task_name": "inspect_evals/assistant_bench_web_search_zero_shot",
        "sample_id": "55f4258484c5b398956133128a50462a767da211f8f72aa5ac5bbffb9bcbba1a",
        "model": "claude-sonnet-4-20250514"
      }
    },
    "evaluate_tool": {
      "name": "score_task",
      "arguments": {}
    },
    "agent_config": {
      "agent_class": "server.agents.InspectAgent",
      "disallowed_tools": [
        "_store_agent_message",
        "load_task",
        "score_task"
      ],
      "allowed_tools": [],
      "initial_screenshot": false
    },
    "system_prompt": "You are a helpful agent that has to provide answers to complex tasks.\n\nTasks can require accessing the web. To access the web, you are equipped with a search engine.\nTo access the search engine, simply generate an intermediate question. The intermediate question will be sent to a search engine (Google) and the result from the search engine will be provided by the user in the next turn.\n\nIt is best to access the search engine with simple queries, as they are most likely to return the correct answers. For example, if you require to find the number of Instagram followers for a group, generate a query for each member of the group.\n\nWhen the user message ends with an intermediate question, always generate an intermediate answer.\n\nIf the search results do not answer your intermediate question, begin the intermediate answer with \"Intermediate answer: The intermediate answer cannot be inferred from the search results\".\nIn that case, you can use your own knowledge to answer the intermediate question or generate a new intermediate question.\n\nYou can generate additional intermediate questions and answers until arriving at the final answer. Do not generate the same intermediate question twice.\n\nWhen arriving at the final answer, start your prediction with: \"The final answer is:\" followed by the answer to the task.\nIf you are unsure what the final answer is, generate the empty string (\"\").\n\nKeep the answer as concise as possible. The answer should either be: a number, a string, a list of strings, or a list of jsons.\nThe answer should be parsed with the python method: json.loads(input_str). If no answer is found, generate an empty string.\n\nThe format will be:\nAssistant: Intermediate question: [Your search engine query]\nUser:      Context0: [Search engine results]\n           Task: [Original task]\n           Intermediate question: [Your most recent search engine query]\nAssistant: Intermediate answer: [Your answer as inferred from the search results or your own knowledge]\n           Intermediate question: [Your next search engine query]\n(Repeat as needed)\nAssistant: Intermediate answer: [Your answer as inferred from the search results or your own knowledge]\n           The final answer is: [Your next search engine query]"
  },
  {
    "name": "inspect_evals/assistant_bench_web_search_zero_shot - Sample 2ddae3b7a208e3c25f14d82d7a1faaaa1832fbf950b4dac345e755c4c361f294",
    "description": "Task from inspect_evals/assistant_bench_web_search_zero_shot dataset, sample 2ddae3b7a208e3c25f14d82d7a1faaaa1832fbf950b4dac345e755c4c361f294",
    "prompt": "What's the lowest price a Single Family house was sold in Queen Anne in January 2023?",
    "mcp_config": {
      "inspect": {
        "command": "docker",
        "args": [
          "run",
          "--rm",
          "-i",
          "--init",
          "-v",
          "/var/run/docker.sock:/var/run/docker.sock",
          "-v",
          "/home/genteki/.cache/inspect_evals:/root/.cache/inspect_evals",
          "hud-inspect:latest"
        ]
      }
    },
    "setup_tool": {
      "name": "load_task",
      "arguments": {
        "task_name": "inspect_evals/assistant_bench_web_search_zero_shot",
        "sample_id": "2ddae3b7a208e3c25f14d82d7a1faaaa1832fbf950b4dac345e755c4c361f294",
        "model": "claude-sonnet-4-20250514"
      }
    },
    "evaluate_tool": {
      "name": "score_task",
      "arguments": {}
    },
    "agent_config": {
      "agent_class": "server.agents.InspectAgent",
      "disallowed_tools": [
        "_store_agent_message",
        "load_task",
        "score_task"
      ],
      "allowed_tools": [],
      "initial_screenshot": false
    },
    "system_prompt": "You are a helpful agent that has to provide answers to complex tasks.\n\nTasks can require accessing the web. To access the web, you are equipped with a search engine.\nTo access the search engine, simply generate an intermediate question. The intermediate question will be sent to a search engine (Google) and the result from the search engine will be provided by the user in the next turn.\n\nIt is best to access the search engine with simple queries, as they are most likely to return the correct answers. For example, if you require to find the number of Instagram followers for a group, generate a query for each member of the group.\n\nWhen the user message ends with an intermediate question, always generate an intermediate answer.\n\nIf the search results do not answer your intermediate question, begin the intermediate answer with \"Intermediate answer: The intermediate answer cannot be inferred from the search results\".\nIn that case, you can use your own knowledge to answer the intermediate question or generate a new intermediate question.\n\nYou can generate additional intermediate questions and answers until arriving at the final answer. Do not generate the same intermediate question twice.\n\nWhen arriving at the final answer, start your prediction with: \"The final answer is:\" followed by the answer to the task.\nIf you are unsure what the final answer is, generate the empty string (\"\").\n\nKeep the answer as concise as possible. The answer should either be: a number, a string, a list of strings, or a list of jsons.\nThe answer should be parsed with the python method: json.loads(input_str). If no answer is found, generate an empty string.\n\nThe format will be:\nAssistant: Intermediate question: [Your search engine query]\nUser:      Context0: [Search engine results]\n           Task: [Original task]\n           Intermediate question: [Your most recent search engine query]\nAssistant: Intermediate answer: [Your answer as inferred from the search results or your own knowledge]\n           Intermediate question: [Your next search engine query]\n(Repeat as needed)\nAssistant: Intermediate answer: [Your answer as inferred from the search results or your own knowledge]\n           The final answer is: [Your next search engine query]"
  },
  {
    "name": "inspect_evals/assistant_bench_web_search_zero_shot - Sample 748899d9d70c09beb3bd48ac8a3658bdcfd2f9114fe6dc4c4b8d2f9541ef4607",
    "description": "Task from inspect_evals/assistant_bench_web_search_zero_shot dataset, sample 748899d9d70c09beb3bd48ac8a3658bdcfd2f9114fe6dc4c4b8d2f9541ef4607",
    "prompt": "How much does it cost to send an envelope with 1-week delivery from Rio de Janeiro to NYC with DHL, USPS, or Fedex? (Format your answers as a list of jsons with the keys \"sender\" and \"price (usd)\" for each option you find)",
    "mcp_config": {
      "inspect": {
        "command": "docker",
        "args": [
          "run",
          "--rm",
          "-i",
          "--init",
          "-v",
          "/var/run/docker.sock:/var/run/docker.sock",
          "-v",
          "/home/genteki/.cache/inspect_evals:/root/.cache/inspect_evals",
          "hud-inspect:latest"
        ]
      }
    },
    "setup_tool": {
      "name": "load_task",
      "arguments": {
        "task_name": "inspect_evals/assistant_bench_web_search_zero_shot",
        "sample_id": "748899d9d70c09beb3bd48ac8a3658bdcfd2f9114fe6dc4c4b8d2f9541ef4607",
        "model": "claude-sonnet-4-20250514"
      }
    },
    "evaluate_tool": {
      "name": "score_task",
      "arguments": {}
    },
    "agent_config": {
      "agent_class": "server.agents.InspectAgent",
      "disallowed_tools": [
        "_store_agent_message",
        "load_task",
        "score_task"
      ],
      "allowed_tools": [],
      "initial_screenshot": false
    },
    "system_prompt": "You are a helpful agent that has to provide answers to complex tasks.\n\nTasks can require accessing the web. To access the web, you are equipped with a search engine.\nTo access the search engine, simply generate an intermediate question. The intermediate question will be sent to a search engine (Google) and the result from the search engine will be provided by the user in the next turn.\n\nIt is best to access the search engine with simple queries, as they are most likely to return the correct answers. For example, if you require to find the number of Instagram followers for a group, generate a query for each member of the group.\n\nWhen the user message ends with an intermediate question, always generate an intermediate answer.\n\nIf the search results do not answer your intermediate question, begin the intermediate answer with \"Intermediate answer: The intermediate answer cannot be inferred from the search results\".\nIn that case, you can use your own knowledge to answer the intermediate question or generate a new intermediate question.\n\nYou can generate additional intermediate questions and answers until arriving at the final answer. Do not generate the same intermediate question twice.\n\nWhen arriving at the final answer, start your prediction with: \"The final answer is:\" followed by the answer to the task.\nIf you are unsure what the final answer is, generate the empty string (\"\").\n\nKeep the answer as concise as possible. The answer should either be: a number, a string, a list of strings, or a list of jsons.\nThe answer should be parsed with the python method: json.loads(input_str). If no answer is found, generate an empty string.\n\nThe format will be:\nAssistant: Intermediate question: [Your search engine query]\nUser:      Context0: [Search engine results]\n           Task: [Original task]\n           Intermediate question: [Your most recent search engine query]\nAssistant: Intermediate answer: [Your answer as inferred from the search results or your own knowledge]\n           Intermediate question: [Your next search engine query]\n(Repeat as needed)\nAssistant: Intermediate answer: [Your answer as inferred from the search results or your own knowledge]\n           The final answer is: [Your next search engine query]"
  },
  {
    "name": "inspect_evals/assistant_bench_web_search_zero_shot - Sample 291b53e665b4dd4365cde995042db4a6f6fecef3fe3a6f4482f23d61bd673918",
    "description": "Task from inspect_evals/assistant_bench_web_search_zero_shot dataset, sample 291b53e665b4dd4365cde995042db4a6f6fecef3fe3a6f4482f23d61bd673918",
    "prompt": "What is the link to the GFF3 file for beluga whales that was the most recent one on 20/10/2020?",
    "mcp_config": {
      "inspect": {
        "command": "docker",
        "args": [
          "run",
          "--rm",
          "-i",
          "--init",
          "-v",
          "/var/run/docker.sock:/var/run/docker.sock",
          "-v",
          "/home/genteki/.cache/inspect_evals:/root/.cache/inspect_evals",
          "hud-inspect:latest"
        ]
      }
    },
    "setup_tool": {
      "name": "load_task",
      "arguments": {
        "task_name": "inspect_evals/assistant_bench_web_search_zero_shot",
        "sample_id": "291b53e665b4dd4365cde995042db4a6f6fecef3fe3a6f4482f23d61bd673918",
        "model": "claude-sonnet-4-20250514"
      }
    },
    "evaluate_tool": {
      "name": "score_task",
      "arguments": {}
    },
    "agent_config": {
      "agent_class": "server.agents.InspectAgent",
      "disallowed_tools": [
        "_store_agent_message",
        "load_task",
        "score_task"
      ],
      "allowed_tools": [],
      "initial_screenshot": false
    },
    "system_prompt": "You are a helpful agent that has to provide answers to complex tasks.\n\nTasks can require accessing the web. To access the web, you are equipped with a search engine.\nTo access the search engine, simply generate an intermediate question. The intermediate question will be sent to a search engine (Google) and the result from the search engine will be provided by the user in the next turn.\n\nIt is best to access the search engine with simple queries, as they are most likely to return the correct answers. For example, if you require to find the number of Instagram followers for a group, generate a query for each member of the group.\n\nWhen the user message ends with an intermediate question, always generate an intermediate answer.\n\nIf the search results do not answer your intermediate question, begin the intermediate answer with \"Intermediate answer: The intermediate answer cannot be inferred from the search results\".\nIn that case, you can use your own knowledge to answer the intermediate question or generate a new intermediate question.\n\nYou can generate additional intermediate questions and answers until arriving at the final answer. Do not generate the same intermediate question twice.\n\nWhen arriving at the final answer, start your prediction with: \"The final answer is:\" followed by the answer to the task.\nIf you are unsure what the final answer is, generate the empty string (\"\").\n\nKeep the answer as concise as possible. The answer should either be: a number, a string, a list of strings, or a list of jsons.\nThe answer should be parsed with the python method: json.loads(input_str). If no answer is found, generate an empty string.\n\nThe format will be:\nAssistant: Intermediate question: [Your search engine query]\nUser:      Context0: [Search engine results]\n           Task: [Original task]\n           Intermediate question: [Your most recent search engine query]\nAssistant: Intermediate answer: [Your answer as inferred from the search results or your own knowledge]\n           Intermediate question: [Your next search engine query]\n(Repeat as needed)\nAssistant: Intermediate answer: [Your answer as inferred from the search results or your own knowledge]\n           The final answer is: [Your next search engine query]"
  },
  {
    "name": "inspect_evals/assistant_bench_web_search_zero_shot - Sample cca4776df3c73e7f9430a2e624aafad056b14322a0b7ca6c0c22b7e7f3f0890a",
    "description": "Task from inspect_evals/assistant_bench_web_search_zero_shot dataset, sample cca4776df3c73e7f9430a2e624aafad056b14322a0b7ca6c0c22b7e7f3f0890a",
    "prompt": "Which member of Apple\u2019s Board of Directors did not hold C-suite positions at their companies when they joined the board?",
    "mcp_config": {
      "inspect": {
        "command": "docker",
        "args": [
          "run",
          "--rm",
          "-i",
          "--init",
          "-v",
          "/var/run/docker.sock:/var/run/docker.sock",
          "-v",
          "/home/genteki/.cache/inspect_evals:/root/.cache/inspect_evals",
          "hud-inspect:latest"
        ]
      }
    },
    "setup_tool": {
      "name": "load_task",
      "arguments": {
        "task_name": "inspect_evals/assistant_bench_web_search_zero_shot",
        "sample_id": "cca4776df3c73e7f9430a2e624aafad056b14322a0b7ca6c0c22b7e7f3f0890a",
        "model": "claude-sonnet-4-20250514"
      }
    },
    "evaluate_tool": {
      "name": "score_task",
      "arguments": {}
    },
    "agent_config": {
      "agent_class": "server.agents.InspectAgent",
      "disallowed_tools": [
        "_store_agent_message",
        "load_task",
        "score_task"
      ],
      "allowed_tools": [],
      "initial_screenshot": false
    },
    "system_prompt": "You are a helpful agent that has to provide answers to complex tasks.\n\nTasks can require accessing the web. To access the web, you are equipped with a search engine.\nTo access the search engine, simply generate an intermediate question. The intermediate question will be sent to a search engine (Google) and the result from the search engine will be provided by the user in the next turn.\n\nIt is best to access the search engine with simple queries, as they are most likely to return the correct answers. For example, if you require to find the number of Instagram followers for a group, generate a query for each member of the group.\n\nWhen the user message ends with an intermediate question, always generate an intermediate answer.\n\nIf the search results do not answer your intermediate question, begin the intermediate answer with \"Intermediate answer: The intermediate answer cannot be inferred from the search results\".\nIn that case, you can use your own knowledge to answer the intermediate question or generate a new intermediate question.\n\nYou can generate additional intermediate questions and answers until arriving at the final answer. Do not generate the same intermediate question twice.\n\nWhen arriving at the final answer, start your prediction with: \"The final answer is:\" followed by the answer to the task.\nIf you are unsure what the final answer is, generate the empty string (\"\").\n\nKeep the answer as concise as possible. The answer should either be: a number, a string, a list of strings, or a list of jsons.\nThe answer should be parsed with the python method: json.loads(input_str). If no answer is found, generate an empty string.\n\nThe format will be:\nAssistant: Intermediate question: [Your search engine query]\nUser:      Context0: [Search engine results]\n           Task: [Original task]\n           Intermediate question: [Your most recent search engine query]\nAssistant: Intermediate answer: [Your answer as inferred from the search results or your own knowledge]\n           Intermediate question: [Your next search engine query]\n(Repeat as needed)\nAssistant: Intermediate answer: [Your answer as inferred from the search results or your own knowledge]\n           The final answer is: [Your next search engine query]"
  },
  {
    "name": "inspect_evals/assistant_bench_web_search_zero_shot - Sample 9e31099fffa6a3891c94934fd4fc2f3f522d51c1904ff3561f3a10e4bf245821",
    "description": "Task from inspect_evals/assistant_bench_web_search_zero_shot dataset, sample 9e31099fffa6a3891c94934fd4fc2f3f522d51c1904ff3561f3a10e4bf245821",
    "prompt": "Which of the people in the monday.com C-suite did not have a C-suite position during the company's IPO?",
    "mcp_config": {
      "inspect": {
        "command": "docker",
        "args": [
          "run",
          "--rm",
          "-i",
          "--init",
          "-v",
          "/var/run/docker.sock:/var/run/docker.sock",
          "-v",
          "/home/genteki/.cache/inspect_evals:/root/.cache/inspect_evals",
          "hud-inspect:latest"
        ]
      }
    },
    "setup_tool": {
      "name": "load_task",
      "arguments": {
        "task_name": "inspect_evals/assistant_bench_web_search_zero_shot",
        "sample_id": "9e31099fffa6a3891c94934fd4fc2f3f522d51c1904ff3561f3a10e4bf245821",
        "model": "claude-sonnet-4-20250514"
      }
    },
    "evaluate_tool": {
      "name": "score_task",
      "arguments": {}
    },
    "agent_config": {
      "agent_class": "server.agents.InspectAgent",
      "disallowed_tools": [
        "_store_agent_message",
        "load_task",
        "score_task"
      ],
      "allowed_tools": [],
      "initial_screenshot": false
    },
    "system_prompt": "You are a helpful agent that has to provide answers to complex tasks.\n\nTasks can require accessing the web. To access the web, you are equipped with a search engine.\nTo access the search engine, simply generate an intermediate question. The intermediate question will be sent to a search engine (Google) and the result from the search engine will be provided by the user in the next turn.\n\nIt is best to access the search engine with simple queries, as they are most likely to return the correct answers. For example, if you require to find the number of Instagram followers for a group, generate a query for each member of the group.\n\nWhen the user message ends with an intermediate question, always generate an intermediate answer.\n\nIf the search results do not answer your intermediate question, begin the intermediate answer with \"Intermediate answer: The intermediate answer cannot be inferred from the search results\".\nIn that case, you can use your own knowledge to answer the intermediate question or generate a new intermediate question.\n\nYou can generate additional intermediate questions and answers until arriving at the final answer. Do not generate the same intermediate question twice.\n\nWhen arriving at the final answer, start your prediction with: \"The final answer is:\" followed by the answer to the task.\nIf you are unsure what the final answer is, generate the empty string (\"\").\n\nKeep the answer as concise as possible. The answer should either be: a number, a string, a list of strings, or a list of jsons.\nThe answer should be parsed with the python method: json.loads(input_str). If no answer is found, generate an empty string.\n\nThe format will be:\nAssistant: Intermediate question: [Your search engine query]\nUser:      Context0: [Search engine results]\n           Task: [Original task]\n           Intermediate question: [Your most recent search engine query]\nAssistant: Intermediate answer: [Your answer as inferred from the search results or your own knowledge]\n           Intermediate question: [Your next search engine query]\n(Repeat as needed)\nAssistant: Intermediate answer: [Your answer as inferred from the search results or your own knowledge]\n           The final answer is: [Your next search engine query]"
  },
  {
    "name": "inspect_evals/assistant_bench_web_search_zero_shot - Sample 6f224e7730ed027cbac73aebb1aea7f954053082041b02b19f4ff126a0a8a208",
    "description": "Task from inspect_evals/assistant_bench_web_search_zero_shot dataset, sample 6f224e7730ed027cbac73aebb1aea7f954053082041b02b19f4ff126a0a8a208",
    "prompt": "Which members of Fubo's Management Team joined the company during the same year Fubo's IPO happened?",
    "mcp_config": {
      "inspect": {
        "command": "docker",
        "args": [
          "run",
          "--rm",
          "-i",
          "--init",
          "-v",
          "/var/run/docker.sock:/var/run/docker.sock",
          "-v",
          "/home/genteki/.cache/inspect_evals:/root/.cache/inspect_evals",
          "hud-inspect:latest"
        ]
      }
    },
    "setup_tool": {
      "name": "load_task",
      "arguments": {
        "task_name": "inspect_evals/assistant_bench_web_search_zero_shot",
        "sample_id": "6f224e7730ed027cbac73aebb1aea7f954053082041b02b19f4ff126a0a8a208",
        "model": "claude-sonnet-4-20250514"
      }
    },
    "evaluate_tool": {
      "name": "score_task",
      "arguments": {}
    },
    "agent_config": {
      "agent_class": "server.agents.InspectAgent",
      "disallowed_tools": [
        "_store_agent_message",
        "load_task",
        "score_task"
      ],
      "allowed_tools": [],
      "initial_screenshot": false
    },
    "system_prompt": "You are a helpful agent that has to provide answers to complex tasks.\n\nTasks can require accessing the web. To access the web, you are equipped with a search engine.\nTo access the search engine, simply generate an intermediate question. The intermediate question will be sent to a search engine (Google) and the result from the search engine will be provided by the user in the next turn.\n\nIt is best to access the search engine with simple queries, as they are most likely to return the correct answers. For example, if you require to find the number of Instagram followers for a group, generate a query for each member of the group.\n\nWhen the user message ends with an intermediate question, always generate an intermediate answer.\n\nIf the search results do not answer your intermediate question, begin the intermediate answer with \"Intermediate answer: The intermediate answer cannot be inferred from the search results\".\nIn that case, you can use your own knowledge to answer the intermediate question or generate a new intermediate question.\n\nYou can generate additional intermediate questions and answers until arriving at the final answer. Do not generate the same intermediate question twice.\n\nWhen arriving at the final answer, start your prediction with: \"The final answer is:\" followed by the answer to the task.\nIf you are unsure what the final answer is, generate the empty string (\"\").\n\nKeep the answer as concise as possible. The answer should either be: a number, a string, a list of strings, or a list of jsons.\nThe answer should be parsed with the python method: json.loads(input_str). If no answer is found, generate an empty string.\n\nThe format will be:\nAssistant: Intermediate question: [Your search engine query]\nUser:      Context0: [Search engine results]\n           Task: [Original task]\n           Intermediate question: [Your most recent search engine query]\nAssistant: Intermediate answer: [Your answer as inferred from the search results or your own knowledge]\n           Intermediate question: [Your next search engine query]\n(Repeat as needed)\nAssistant: Intermediate answer: [Your answer as inferred from the search results or your own knowledge]\n           The final answer is: [Your next search engine query]"
  },
  {
    "name": "inspect_evals/assistant_bench_web_search_zero_shot - Sample 8fa42360185068216f2919935148d4e1ad28ddc18da0abd0f4bb0b6b6f84b127",
    "description": "Task from inspect_evals/assistant_bench_web_search_zero_shot dataset, sample 8fa42360185068216f2919935148d4e1ad28ddc18da0abd0f4bb0b6b6f84b127",
    "prompt": "Which Vanguard ETF had the highest pct increase between 2013 and 2023: VGT, MGK, or VONG?",
    "mcp_config": {
      "inspect": {
        "command": "docker",
        "args": [
          "run",
          "--rm",
          "-i",
          "--init",
          "-v",
          "/var/run/docker.sock:/var/run/docker.sock",
          "-v",
          "/home/genteki/.cache/inspect_evals:/root/.cache/inspect_evals",
          "hud-inspect:latest"
        ]
      }
    },
    "setup_tool": {
      "name": "load_task",
      "arguments": {
        "task_name": "inspect_evals/assistant_bench_web_search_zero_shot",
        "sample_id": "8fa42360185068216f2919935148d4e1ad28ddc18da0abd0f4bb0b6b6f84b127",
        "model": "claude-sonnet-4-20250514"
      }
    },
    "evaluate_tool": {
      "name": "score_task",
      "arguments": {}
    },
    "agent_config": {
      "agent_class": "server.agents.InspectAgent",
      "disallowed_tools": [
        "_store_agent_message",
        "load_task",
        "score_task"
      ],
      "allowed_tools": [],
      "initial_screenshot": false
    },
    "system_prompt": "You are a helpful agent that has to provide answers to complex tasks.\n\nTasks can require accessing the web. To access the web, you are equipped with a search engine.\nTo access the search engine, simply generate an intermediate question. The intermediate question will be sent to a search engine (Google) and the result from the search engine will be provided by the user in the next turn.\n\nIt is best to access the search engine with simple queries, as they are most likely to return the correct answers. For example, if you require to find the number of Instagram followers for a group, generate a query for each member of the group.\n\nWhen the user message ends with an intermediate question, always generate an intermediate answer.\n\nIf the search results do not answer your intermediate question, begin the intermediate answer with \"Intermediate answer: The intermediate answer cannot be inferred from the search results\".\nIn that case, you can use your own knowledge to answer the intermediate question or generate a new intermediate question.\n\nYou can generate additional intermediate questions and answers until arriving at the final answer. Do not generate the same intermediate question twice.\n\nWhen arriving at the final answer, start your prediction with: \"The final answer is:\" followed by the answer to the task.\nIf you are unsure what the final answer is, generate the empty string (\"\").\n\nKeep the answer as concise as possible. The answer should either be: a number, a string, a list of strings, or a list of jsons.\nThe answer should be parsed with the python method: json.loads(input_str). If no answer is found, generate an empty string.\n\nThe format will be:\nAssistant: Intermediate question: [Your search engine query]\nUser:      Context0: [Search engine results]\n           Task: [Original task]\n           Intermediate question: [Your most recent search engine query]\nAssistant: Intermediate answer: [Your answer as inferred from the search results or your own knowledge]\n           Intermediate question: [Your next search engine query]\n(Repeat as needed)\nAssistant: Intermediate answer: [Your answer as inferred from the search results or your own knowledge]\n           The final answer is: [Your next search engine query]"
  }
]