-
Notifications
You must be signed in to change notification settings - Fork 73
Expand file tree
/
Copy pathindex.html
More file actions
1274 lines (1211 loc) · 83.7 KB
/
Copy pathindex.html
File metadata and controls
1274 lines (1211 loc) · 83.7 KB
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
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
<!DOCTYPE html>
<html lang="en">
<head>
<meta charset="UTF-8">
<meta name="viewport" content="width=device-width, initial-scale=1.0">
<title>API Relay Audit - AI API Relay Security Audit for LLM Proxies</title>
<meta name="description" content="API Relay Audit runs locally for AI API relays and LLM proxies, with separate query families for relay audits, prompt injection audits, model substitution signals, and Web3 relay audits.">
<link rel="canonical" href="https://toby-bridges.github.io/api-relay-audit/">
<link rel="alternate" hreflang="en" href="https://toby-bridges.github.io/api-relay-audit/">
<link rel="alternate" hreflang="zh-CN" href="https://toby-bridges.github.io/api-relay-audit/zh/">
<link rel="alternate" hreflang="x-default" href="https://toby-bridges.github.io/api-relay-audit/">
<meta property="og:type" content="website">
<meta property="og:site_name" content="API Relay Audit">
<meta property="og:title" content="API Relay Audit - AI API Relay Security Audit">
<meta property="og:description" content="Local security audit for AI API relays and LLM proxies with clean evidence boundaries across relay, prompt injection, model substitution, and Web3 profiles.">
<meta property="og:url" content="https://toby-bridges.github.io/api-relay-audit/">
<meta property="og:image" content="https://toby-bridges.github.io/api-relay-audit/assets/social-preview.png">
<meta property="og:image:width" content="1280">
<meta property="og:image:height" content="640">
<meta property="og:image:alt" content="API Relay Audit - AI API Relay Security Audit. Runs locally; your API key is sent only to the relay URL you choose.">
<meta name="twitter:card" content="summary_large_image">
<meta name="twitter:title" content="API Relay Audit - AI API Relay Security Audit">
<meta name="twitter:description" content="Local security audit for AI API relays and LLM proxies with clean query-family and evidence boundaries.">
<meta name="twitter:image" content="https://toby-bridges.github.io/api-relay-audit/assets/social-preview.png">
<script type="application/ld+json">
{
"@context": "https://schema.org",
"@graph": [
{
"@type": "WebSite",
"@id": "https://toby-bridges.github.io/api-relay-audit/#website",
"name": "API Relay Audit",
"url": "https://toby-bridges.github.io/api-relay-audit/",
"description": "Local security audit for AI API relays and LLM proxies."
},
{
"@type": "SoftwareApplication",
"@id": "https://toby-bridges.github.io/api-relay-audit/#software",
"name": "API Relay Audit",
"applicationCategory": "SecurityApplication",
"operatingSystem": "macOS, Windows, Linux",
"url": "https://github.com/toby-bridges/api-relay-audit",
"codeRepository": "https://github.com/toby-bridges/api-relay-audit",
"license": "https://github.com/toby-bridges/api-relay-audit/blob/master/LICENSE",
"description": "Local security audit tool for AI API relays and LLM proxies. It preserves separate query families for relay audits, prompt injection audits, model substitution signals, and Web3 relay audits. Your API key is sent only to the relay URL you choose."
},
{
"@type": "FAQPage",
"@id": "https://toby-bridges.github.io/api-relay-audit/#faq",
"mainEntity": [
{
"@type": "Question",
"name": "What is an API relay or LLM proxy?",
"acceptedAnswer": {
"@type": "Answer",
"text": "An API relay or LLM proxy is a third-party service between you and an AI provider such as Anthropic or OpenAI. It forwards your requests upstream, but it can also inject hidden instructions, swap models, truncate context, rewrite tool output, or leak credentials in error responses."
}
},
{
"@type": "Question",
"name": "Is it safe to enter my API key?",
"acceptedAnswer": {
"@type": "Answer",
"text": "API Relay Audit runs locally, so your API key is sent only to the relay URL you specify. No data is sent to API Relay Audit servers, and the standalone version has zero Python package dependencies."
}
},
{
"@type": "Question",
"name": "What does prompt injection mean here?",
"acceptedAnswer": {
"@type": "Answer",
"text": "Prompt injection means the relay may prepend or insert hidden instructions into your request. API Relay Audit compares token usage, runs extraction probes, and records evidence when hidden prompt content appears."
}
},
{
"@type": "Question",
"name": "What is model substitution?",
"acceptedAnswer": {
"@type": "Answer",
"text": "Model substitution means the relay may expose signals for a different model identity, route, or upstream channel than expected. API Relay Audit treats self-ID, stream model identity, latency, and channel fingerprints as evidence signals that require corroboration."
}
},
{
"@type": "Question",
"name": "What is tool-call rewriting?",
"acceptedAnswer": {
"@type": "Answer",
"text": "Tool-call rewriting means the relay modifies package-install commands or tool-like output in the model response. API Relay Audit compares pinned package commands against returned text to detect proxy-layer supply-chain tampering."
}
},
{
"@type": "Question",
"name": "What are SSE anomalies?",
"acceptedAnswer": {
"@type": "Answer",
"text": "SSE anomalies are stream-level integrity issues in Anthropic-style streaming responses. API Relay Audit checks event types, usage monotonicity, thinking signatures, and stream model identity when supported."
}
},
{
"@type": "Question",
"name": "What Web3 wallet risks does it check?",
"acceptedAnswer": {
"@type": "Answer",
"text": "With the web3 or full profile, API Relay Audit checks transfer guidance, signed-transaction refusal, and private-key refusal behavior before wallet-related traffic is trusted."
}
},
{
"@type": "Question",
"name": "What does inconclusive mean?",
"acceptedAnswer": {
"@type": "Answer",
"text": "Inconclusive means the tool could not determine a clean or anomalous result for that step. Blocked probes, unsupported formats, and ambiguous responses are not treated as clean."
}
}
]
}
]
}
</script>
<style>
*{margin:0;padding:0;box-sizing:border-box}
:root{
--bg:#0a0e17;--bg2:#0d1220;--card:#111827;--card-hover:#1a2332;
--border:#1e293b;--border2:#283548;
--text:#e2e8f0;--text2:#94a3b8;--text3:#64748b;
--accent:#3b82f6;--accent2:#60a5fa;
--green:#10b981;--yellow:#f59e0b;--red:#ef4444;--gray:#6b7280;
--radius:10px;
}
html{scroll-behavior:smooth}
body{background:var(--bg);color:var(--text);font-family:-apple-system,BlinkMacSystemFont,'Segoe UI','Noto Sans SC',sans-serif;line-height:1.6;-webkit-font-smoothing:antialiased}
/* ── Nav ── */
nav{position:fixed;top:0;left:0;right:0;z-index:100;background:rgba(10,14,23,.85);backdrop-filter:blur(12px);border-bottom:1px solid var(--border)}
.nav-inner{max-width:960px;margin:0 auto;padding:12px 24px;display:flex;justify-content:space-between;align-items:center}
.nav-logo{font-weight:700;font-size:1rem;color:var(--text);text-decoration:none;display:flex;align-items:center;gap:8px}
.nav-logo span{background:linear-gradient(135deg,var(--accent2),#a78bfa);-webkit-background-clip:text;-webkit-text-fill-color:transparent}
.nav-links{display:flex;gap:20px;align-items:center}
.nav-links a{color:var(--text2);font-size:.85rem;text-decoration:none;transition:color .2s}
.nav-links a:hover{color:var(--text)}
.lang-switch{display:flex;gap:2px;background:var(--card);border:1px solid var(--border);border-radius:6px;overflow:hidden}
.lang-btn{padding:4px 8px;font-size:.7rem;background:none;border:none;color:var(--text3);cursor:pointer;transition:all .2s}
.lang-btn.active{background:var(--accent);color:#fff}
.lang-btn:hover:not(.active){color:var(--text)}
/* ── Sections ── */
.container{max-width:960px;margin:0 auto;padding:0 24px}
section{padding:80px 0 50px}
section:first-of-type{padding-top:100px}
/* ── Hero ── */
.hero{text-align:center}
.hero h1{font-size:2.2rem;font-weight:800;margin-bottom:16px;letter-spacing:-.02em}
.hero h1 .gradient{background:linear-gradient(135deg,var(--accent2),#a78bfa,#f472b6);-webkit-background-clip:text;-webkit-text-fill-color:transparent}
.hero-sub{color:var(--text2);font-size:1.05rem;max-width:600px;margin:0 auto 12px;line-height:1.7}
.hero-stats{display:flex;justify-content:center;gap:32px;margin:20px 0 28px;flex-wrap:wrap}
.stat{text-align:center}
.stat-num{font-size:1.6rem;font-weight:700;color:var(--accent2)}
.stat-label{font-size:.75rem;color:var(--text3);margin-top:2px}
.hero-cta{display:flex;gap:12px;justify-content:center;flex-wrap:wrap;margin-bottom:0}
.btn{padding:10px 24px;border-radius:8px;font-size:.9rem;font-weight:600;text-decoration:none;transition:all .2s;cursor:pointer;border:none;display:inline-flex;align-items:center;gap:8px}
.btn-primary{background:var(--accent);color:#fff}.btn-primary:hover{background:#2563eb}
.btn-secondary{background:transparent;border:1px solid var(--border2);color:var(--text2)}.btn-secondary:hover{border-color:var(--accent);color:var(--text)}
.security-badge{display:inline-flex;align-items:center;gap:6px;background:rgba(16,185,129,.08);border:1px solid rgba(16,185,129,.2);color:var(--green);padding:8px 16px;border-radius:20px;font-size:.8rem;margin-bottom:8px}
/* ── Search summary ── */
.summary-section{padding-top:28px}
.summary-grid{display:grid;grid-template-columns:repeat(auto-fit,minmax(240px,1fr));gap:16px;margin-top:24px}
.summary-block{background:var(--card);border:1px solid var(--border);border-radius:var(--radius);padding:22px 24px}
.summary-block h2,.summary-block h3{font-size:1rem;font-weight:700;margin-bottom:8px}
.summary-block p,.summary-block li{color:var(--text2);font-size:.86rem;line-height:1.65}
.summary-block ul{padding-left:18px;margin-top:8px}
/* ── Demo Report ── */
.demo-section h2{font-size:1.4rem;font-weight:700;margin-bottom:8px;text-align:center}
.demo-subtitle{text-align:center;color:var(--text2);font-size:.9rem;margin-bottom:32px}
.demo-tabs{display:flex;gap:8px;margin-bottom:20px;justify-content:center;flex-wrap:wrap}
.demo-tab{padding:8px 16px;border-radius:8px;font-size:.82rem;background:var(--card);border:1px solid var(--border);color:var(--text2);cursor:pointer;transition:all .2s}
.demo-tab.active{border-color:var(--accent);color:var(--accent);background:rgba(59,130,246,.08)}
.demo-tab:hover:not(.active){border-color:var(--border2);color:var(--text)}
.demo-card{background:var(--card);border:1px solid var(--border);border-radius:var(--radius);overflow:hidden;display:none}
.demo-card.active{display:block}
.demo-bar{height:3px}
.demo-bar.red{background:var(--red)}.demo-bar.yellow{background:var(--yellow)}.demo-bar.green{background:var(--green)}
.demo-head{padding:24px 28px 16px;display:flex;justify-content:space-between;align-items:flex-start;flex-wrap:wrap;gap:12px}
.demo-domain{font-size:1.15rem;font-weight:700;font-family:'SF Mono',Monaco,'Cascadia Code',monospace}
.demo-badge{padding:4px 12px;border-radius:20px;font-size:.78rem;font-weight:600}
.demo-badge.red{background:rgba(239,68,68,.12);color:#fca5a5}
.demo-badge.yellow{background:rgba(245,158,11,.12);color:#fcd34d}
.demo-badge.green{background:rgba(16,185,129,.12);color:#6ee7b7}
.demo-meta{padding:0 28px 16px;display:flex;gap:16px;color:var(--text3);font-size:.8rem;flex-wrap:wrap}
.demo-grid{padding:0 28px 24px;display:grid;grid-template-columns:repeat(auto-fit,minmax(200px,1fr));gap:12px}
.demo-metric{background:rgba(255,255,255,.025);border-radius:8px;padding:14px 16px}
.demo-metric-label{font-size:.7rem;color:var(--text3);text-transform:uppercase;letter-spacing:.5px;margin-bottom:6px}
.demo-metric-value{font-size:.95rem;font-weight:600}
.demo-metric-value.ok{color:var(--green)}.demo-metric-value.warn{color:var(--yellow)}.demo-metric-value.danger{color:var(--red)}
.demo-detail{padding:0 28px 24px}
.demo-table{width:100%;border-collapse:collapse;font-size:.82rem}
.demo-table th{text-align:left;padding:8px 12px;color:var(--text3);font-weight:500;border-bottom:1px solid var(--border);font-size:.72rem;text-transform:uppercase;letter-spacing:.5px}
.demo-table td{padding:8px 12px;border-bottom:1px solid rgba(30,41,59,.4)}
.pass{color:var(--green)}.fail{color:var(--red)}.warn-text{color:var(--yellow)}
.demo-flags{padding:0 28px 24px;display:flex;flex-wrap:wrap;gap:8px}
.flag{background:rgba(239,68,68,.08);border:1px solid rgba(239,68,68,.18);color:#fca5a5;padding:5px 12px;border-radius:6px;font-size:.78rem}
.demo-note{text-align:center;color:var(--text3);font-size:.78rem;margin-top:16px}
/* ── How it works ── */
.steps{display:grid;grid-template-columns:repeat(auto-fit,minmax(260px,1fr));gap:16px;margin-top:32px}
.step{background:var(--card);border:1px solid var(--border);border-radius:var(--radius);padding:24px}
.step-num{font-size:.7rem;font-weight:700;color:var(--accent);text-transform:uppercase;letter-spacing:1px;margin-bottom:8px}
.step h3{font-size:1rem;font-weight:600;margin-bottom:8px}
.step p{color:var(--text2);font-size:.85rem;line-height:1.6}
/* ── Install / CTA ── */
.install-section{padding-top:40px}
.install-section h2{font-size:1.4rem;font-weight:700;margin-bottom:8px;text-align:center}
.install-sub{color:var(--text2);font-size:.9rem;margin-bottom:28px;text-align:center}
/* Form */
.audit-form-wrapper{max-width:560px;margin:0 auto 28px}
.audit-form{background:var(--card);border:1px solid var(--border);border-radius:var(--radius);padding:28px;position:relative}
.form-row{margin-bottom:16px}
.form-row:last-of-type{margin-bottom:0}
.form-label{display:block;font-size:.78rem;color:var(--text2);margin-bottom:6px;font-weight:500}
.form-input{width:100%;padding:10px 14px;background:var(--bg);border:1px solid var(--border);border-radius:8px;color:var(--text);font-size:.88rem;font-family:inherit;outline:none;transition:border-color .2s}
.form-input:focus{border-color:var(--accent)}
.form-input::placeholder{color:var(--text3)}
.form-select{width:100%;padding:10px 14px;background:var(--bg);border:1px solid var(--border);border-radius:8px;color:var(--text);font-size:.88rem;font-family:inherit;outline:none;-webkit-appearance:none;appearance:none;background-image:url("data:image/svg+xml,%3Csvg xmlns='http://www.w3.org/2000/svg' width='12' height='12' viewBox='0 0 24 24' fill='none' stroke='%2394a3b8' stroke-width='2'%3E%3Cpath d='M6 9l6 6 6-6'/%3E%3C/svg%3E");background-repeat:no-repeat;background-position:right 12px center;cursor:pointer}
.form-select:focus{border-color:var(--accent)}
.form-hint{font-size:.72rem;color:var(--text3);margin-top:4px}
.generate-btn{width:100%;padding:12px;background:var(--accent);color:#fff;border:none;border-radius:8px;font-size:.92rem;font-weight:600;cursor:pointer;transition:background .2s;margin-top:20px}
.generate-btn:hover{background:#2563eb}
/* Generated command */
.generated-cmd{display:none;max-width:560px;margin:0 auto 20px}
.generated-cmd.visible{display:block}
.cmd-steps{margin-bottom:16px}
.cmd-step{display:flex;gap:12px;align-items:flex-start;margin-bottom:12px}
.cmd-step-num{width:24px;height:24px;background:var(--accent);color:#fff;border-radius:50%;display:flex;align-items:center;justify-content:center;font-size:.72rem;font-weight:700;flex-shrink:0;margin-top:2px}
.cmd-step-text{font-size:.85rem;color:var(--text2);line-height:1.5}
.cmd-block{background:#0d1117;border:1px solid var(--border);border-radius:var(--radius);padding:16px 20px;font-family:'SF Mono',Monaco,'Cascadia Code',monospace;font-size:.82rem;color:var(--accent2);position:relative;overflow-x:auto;line-height:1.7;word-break:break-all}
.cmd-block .comment{color:var(--text3)}
.copy-btn{position:absolute;top:10px;right:10px;background:var(--card);border:1px solid var(--border);color:var(--text3);padding:4px 10px;border-radius:6px;font-size:.72rem;cursor:pointer;transition:all .2s}
.copy-btn:hover{color:var(--text);border-color:var(--border2)}
/* Security flow */
.security-flow{max-width:560px;margin:0 auto 20px;background:var(--card);border:1px solid var(--border);border-radius:var(--radius);padding:20px 24px}
.flow-title{font-size:.82rem;font-weight:600;color:var(--green);margin-bottom:14px;display:flex;align-items:center;gap:8px}
.flow-diagram{display:flex;align-items:center;justify-content:center;gap:0;flex-wrap:wrap;font-size:.78rem;padding:4px 0}
.flow-node{padding:8px 14px;border-radius:8px;text-align:center;white-space:nowrap}
.flow-node.you{background:rgba(59,130,246,.12);border:1px solid rgba(59,130,246,.25);color:var(--accent2)}
.flow-node.local{background:rgba(16,185,129,.1);border:1px solid rgba(16,185,129,.2);color:var(--green)}
.flow-node.relay{background:rgba(245,158,11,.1);border:1px solid rgba(245,158,11,.2);color:var(--yellow)}
.flow-node.danger{background:rgba(239,68,68,.08);border:1px solid rgba(239,68,68,.18);color:#fca5a5}
.flow-arrow{color:var(--text3);font-size:.9rem;padding:0 6px;flex-shrink:0}
.flow-vs{display:flex;align-items:center;gap:12px;margin-top:14px;padding-top:14px;border-top:1px solid var(--border);font-size:.75rem;color:var(--text3)}
.flow-vs-bad{color:var(--red);display:flex;align-items:center;gap:4px}
.flow-vs-good{color:var(--green);display:flex;align-items:center;gap:4px}
.cost-badge{display:inline-flex;align-items:center;gap:4px;background:rgba(245,158,11,.08);border:1px solid rgba(245,158,11,.15);color:var(--yellow);padding:4px 12px;border-radius:12px;font-size:.78rem;margin-top:8px}
.install-note{color:var(--text3);font-size:.8rem;margin-top:16px;display:flex;align-items:center;justify-content:center;gap:6px;text-align:center}
/* ── Comparison table ── */
.compare-section h2{font-size:1.4rem;font-weight:700;margin-bottom:8px;text-align:center}
.compare-sub{text-align:center;color:var(--text2);font-size:.9rem;margin-bottom:28px}
.compare-table{width:100%;border-collapse:collapse;font-size:.82rem;background:var(--card);border:1px solid var(--border);border-radius:var(--radius);overflow:hidden}
.compare-table th{padding:12px 16px;text-align:left;font-weight:600;border-bottom:1px solid var(--border);font-size:.78rem}
.compare-table th:first-child{color:var(--text2)}
.compare-table th.ours{color:var(--accent2)}
.compare-table td{padding:10px 16px;border-bottom:1px solid rgba(30,41,59,.3)}
.compare-table td:first-child{color:var(--text2);font-size:.8rem}
.compare-table tr:last-child td{border-bottom:none}
.compare-table .check{color:var(--green)}.compare-table .cross{color:var(--text3)}.compare-table .soon{color:var(--yellow)}
/* ── FAQ ── */
.faq-section h2{font-size:1.4rem;font-weight:700;margin-bottom:28px;text-align:center}
.faq-item{background:var(--card);border:1px solid var(--border);border-radius:var(--radius);margin-bottom:10px;overflow:hidden}
.faq-q{padding:16px 20px;cursor:pointer;font-weight:600;font-size:.9rem;display:flex;justify-content:space-between;align-items:center;transition:background .2s}
.faq-q:hover{background:var(--card-hover)}
.faq-q::after{content:'+';color:var(--text3);font-size:1.1rem;transition:transform .2s}
.faq-item.open .faq-q::after{content:'-'}
.faq-a{padding:0 20px;max-height:0;overflow:hidden;transition:max-height .3s ease,padding .3s ease;color:var(--text2);font-size:.85rem;line-height:1.7}
.faq-item.open .faq-a{padding:0 20px 16px;max-height:600px}
.faq-a a{color:var(--accent2);text-decoration:none}
.faq-a a:hover{text-decoration:underline}
/* ── Guides ── */
.guides-section h2{font-size:1.4rem;font-weight:700;margin-bottom:8px;text-align:center}
.guides-sub{text-align:center;color:var(--text2);font-size:.9rem;margin-bottom:28px}
.guide-grid{display:grid;grid-template-columns:repeat(auto-fit,minmax(220px,1fr));gap:14px}
.guide-card{background:var(--card);border:1px solid var(--border);border-radius:var(--radius);padding:18px 20px;text-decoration:none;display:block;transition:border-color .2s,background .2s}
.guide-card:hover{border-color:var(--accent);background:var(--card-hover);text-decoration:none}
.guide-card h3{font-size:.95rem;margin-bottom:8px;color:var(--text)}
.guide-card p{font-size:.8rem;color:var(--text2);line-height:1.55}
/* ── Footer ── */
footer{text-align:center;padding:40px 24px;border-top:1px solid var(--border);color:var(--text3);font-size:.78rem}
footer a{color:var(--text2);text-decoration:none}
footer a:hover{color:var(--text)}
/* ── Responsive ── */
@media(max-width:640px){
.hero h1{font-size:1.6rem}
.hero-stats{gap:20px}
.stat-num{font-size:1.3rem}
.demo-head{padding:16px 16px 12px}
.demo-meta,.demo-grid,.demo-detail,.demo-flags{padding-left:16px;padding-right:16px}
.nav-links a.hide-mobile{display:none}
.compare-table{font-size:.72rem}
.compare-table th,.compare-table td{padding:8px 10px}
}
/* ── Audit Tabs ── */
.audit-tabs{display:flex;gap:8px;margin-bottom:24px;justify-content:center}
.audit-tab{padding:10px 24px;border-radius:8px;font-size:.88rem;font-weight:600;background:var(--card);border:1px solid var(--border);color:var(--text2);cursor:pointer;transition:all .2s}
.audit-tab.active{border-color:var(--accent);color:var(--accent);background:rgba(59,130,246,.08)}
.audit-tab:hover:not(.active){border-color:var(--border2);color:var(--text)}
.tab-panel{display:none}
.tab-panel.active{display:block}
/* Agent tab */
.agent-pitch{text-align:center;color:var(--text2);font-size:.95rem;margin-bottom:24px;line-height:1.6}
.agent-steps{max-width:560px;margin:0 auto}
.agent-step{background:var(--card);border:1px solid var(--border);border-radius:var(--radius);padding:20px 24px;margin-bottom:14px}
.agent-step-head{display:flex;align-items:center;gap:10px;margin-bottom:10px}
.agent-step-num{width:26px;height:26px;background:var(--accent);color:#fff;border-radius:50%;display:flex;align-items:center;justify-content:center;font-size:.75rem;font-weight:700;flex-shrink:0}
.agent-step-title{font-size:.9rem;font-weight:600}
.agent-step p{color:var(--text2);font-size:.83rem;line-height:1.6;margin-bottom:10px}
.agent-cmd{background:#0d1117;border:1px solid var(--border);border-radius:8px;padding:14px 18px;font-family:'SF Mono',Monaco,'Cascadia Code',monospace;font-size:.8rem;color:var(--accent2);position:relative;overflow-x:auto;line-height:1.6;word-break:break-all}
.agent-trigger{background:rgba(59,130,246,.06);border:1px solid rgba(59,130,246,.15);border-radius:8px;padding:14px 18px;font-size:.88rem;color:var(--text);font-style:italic;text-align:center}
.agent-note{text-align:center;color:var(--text3);font-size:.78rem;margin-top:16px;display:flex;align-items:center;justify-content:center;gap:6px}
/* ── Terminal Demo Animation ── */
.terminal-demo{max-width:560px;margin:24px auto 0;background:#0d1117;border:1px solid var(--border);border-radius:var(--radius);overflow:hidden;display:none}
.terminal-demo.visible{display:block}
.term-bar{background:#161b22;padding:8px 14px;display:flex;align-items:center;gap:6px}
.term-dot{width:10px;height:10px;border-radius:50%}
.term-dot.r{background:#ff5f57}.term-dot.y{background:#febc2e}.term-dot.g{background:#28c840}
.term-bar-title{color:var(--text3);font-size:.7rem;margin-left:8px;font-family:monospace}
.term-body{padding:16px 18px;font-family:'SF Mono',Monaco,'Cascadia Code',monospace;font-size:.78rem;line-height:1.8;color:var(--text2);min-height:160px}
.term-line{opacity:0;transform:translateY(4px);animation:termIn .3s forwards}
.term-line .prompt{color:var(--green)}
.term-line .out{color:var(--text3)}
.term-line .highlight{color:var(--accent2)}
.term-line .warn{color:var(--yellow)}
.term-line .ok{color:var(--green)}
@keyframes termIn{to{opacity:1;transform:translateY(0)}}
/* ── Hidden i18n ── */
[data-i18n]{transition:opacity .15s}
</style>
</head>
<body>
<!-- ── Nav ── -->
<nav>
<div class="nav-inner">
<a class="nav-logo" href="#">
<span>API Relay Audit</span>
</a>
<div class="nav-links">
<a href="#demo" class="hide-mobile" data-i18n="nav_demo">Demo</a>
<a href="#install" class="hide-mobile" data-i18n="nav_install">Quick Start</a>
<a href="#guides" class="hide-mobile">Guides</a>
<a href="#faq" class="hide-mobile" data-i18n="nav_faq">FAQ</a>
<a href="https://github.com/toby-bridges/api-relay-audit" target="_blank" rel="noopener noreferrer">GitHub</a>
<a href="https://x.com/li9292" target="_blank" rel="noopener noreferrer">X</a>
<div class="lang-switch">
<button class="lang-btn" onclick="window.location.href='zh/'">中</button>
<button class="lang-btn active" onclick="setLang('en')">EN</button>
</div>
</div>
</div>
</nav>
<!-- ── Hero ── -->
<section class="hero container">
<div class="security-badge">
<svg width="14" height="14" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2"><path d="M12 22s8-4 8-10V5l-8-3-8 3v7c0 6 8 10 8 10z"/></svg>
<span data-i18n="hero_badge">Local execution — your API key is sent only to the relay URL you choose</span>
</div>
<h1 data-i18n="hero_title"><span class="gradient">AI API Relay</span> Security Audit</h1>
<p class="hero-sub" data-i18n="hero_sub">API Relay Audit is a local, evidence-first audit for AI API relays and LLM proxies. It keeps relay audits, prompt injection audits, model substitution signals, and Web3 relay audits as separate query families.</p>
<div class="hero-stats">
<div class="stat"><div class="stat-num">14</div><div class="stat-label" data-i18n="stat_steps">Audit Steps</div></div>
<div class="stat"><div class="stat-num">6D</div><div class="stat-label" data-i18n="stat_matrix">Risk Matrix</div></div>
<div class="stat"><div class="stat-num">778</div><div class="stat-label" data-i18n="stat_tests">Unit Tests</div></div>
<div class="stat"><div class="stat-num">0</div><div class="stat-label" data-i18n="stat_deps">Dependencies (standalone)</div></div>
</div>
<div class="hero-cta">
<a class="btn btn-primary" href="#install" data-i18n="cta_install">Start Audit</a>
<a class="btn btn-secondary" href="#demo" data-i18n="cta_demo">See Demo Report</a>
</div>
</section>
<!-- ── Search / AI Summary ── -->
<section class="summary-section container" id="overview">
<div class="summary-grid">
<div class="summary-block">
<h2 data-i18n="what_title">What is API Relay Audit?</h2>
<p data-i18n="what_desc">API Relay Audit is a local security audit tool for AI API relays and LLM proxies. It checks whether a third-party relay injects prompts, exposes model identity signals, rewrites tool output, leaks credentials in error responses, or produces stream integrity anomalies.</p>
</div>
<div class="summary-block">
<h3 data-i18n="use_title">When to use it</h3>
<ul>
<li data-i18n="use_1">Audit third-party AI API relays, mirrors, gateways, and LLM proxies.</li>
<li data-i18n="use_2">Check Claude-compatible or OpenAI-compatible proxies before production traffic.</li>
<li data-i18n="use_3">Test relay behavior before coding-agent automation or wallet-related actions.</li>
</ul>
</div>
<div class="summary-block">
<h3 data-i18n="claim_title">What it does not claim</h3>
<ul>
<li data-i18n="claim_1">It does not certify that a relay is safe.</li>
<li data-i18n="claim_2">It does not replace manual security review.</li>
<li data-i18n="claim_3">It does not treat inconclusive as clean.</li>
</ul>
</div>
</div>
</section>
<!-- ── Audit Families ── -->
<section class="summary-section container" id="audit-families">
<h2 style="font-size:1.4rem;font-weight:700;text-align:center" data-i18n="family_title">Audit Families</h2>
<p style="text-align:center;color:var(--text2);font-size:.9rem;margin:8px auto 24px;max-width:680px" data-i18n="family_sub">Each family maps to a distinct search intent, runtime profile, and evidence boundary.</p>
<div class="summary-grid">
<div class="summary-block">
<h3 data-i18n="family_api_title">API relay audit</h3>
<p data-i18n="family_api_desc">Audit third-party relays, mirrors, gateways, LLM proxies, or resale APIs with the general profile before trusting production or agent traffic.</p>
</div>
<div class="summary-block">
<h3 data-i18n="family_prompt_title">Prompt injection audit</h3>
<p data-i18n="family_prompt_desc">Detect hidden prompt injection, prompt leakage, instruction override, and extraction behavior through Steps 3-6.</p>
</div>
<div class="summary-block">
<h3 data-i18n="family_model_title">Model substitution signals</h3>
<p data-i18n="family_model_desc">Collect identity, stream, latency, and channel signals without treating self-ID or fingerprints as standalone provider proof.</p>
</div>
<div class="summary-block">
<h3 data-i18n="family_web3_title">Web3 relay audit</h3>
<p data-i18n="family_web3_desc">Use the profile-gated Step 11 checks with --profile web3 or --profile full before wallet-sensitive agent workflows.</p>
</div>
</div>
</section>
<!-- ── Audit Form (moved to first position for immediate user value) ── -->
<section class="install-section container" id="install">
<!-- Security flow diagram -->
<div class="security-flow">
<div class="flow-title">
<svg width="14" height="14" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2"><path d="M12 22s8-4 8-10V5l-8-3-8 3v7c0 6 8 10 8 10z"/></svg>
<span data-i18n="flow_title">How your Key flows — the critical difference</span>
</div>
<div class="flow-diagram">
<div class="flow-node you" data-i18n="flow_you">You</div>
<div class="flow-arrow">→</div>
<div class="flow-node local" data-i18n="flow_local">Your Terminal</div>
<div class="flow-arrow">→</div>
<div class="flow-node relay" data-i18n="flow_relay">Your Relay</div>
<div class="flow-arrow">→</div>
<div class="flow-node you">Anthropic / OpenAI</div>
</div>
<div class="flow-vs">
<span class="flow-vs-good">
<svg width="12" height="12" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2"><path d="M20 6L9 17l-5-5"/></svg>
<span data-i18n="flow_good">api-relay-audit: no extra checker server sees your key</span>
</span>
<span style="color:var(--text3)">vs</span>
<span class="flow-vs-bad">
<svg width="12" height="12" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2"><path d="M18 6L6 18M6 6l12 12"/></svg>
<span data-i18n="flow_bad">Web tools: Key goes to a 3rd-party server first</span>
</span>
</div>
</div>
<!-- Tab switcher -->
<div class="audit-tabs">
<button class="audit-tab active" onclick="switchAuditTab('human')" data-i18n="tab_human">For Human</button>
<button class="audit-tab" onclick="switchAuditTab('agent')" data-i18n="tab_agent">For Agent</button>
</div>
<!-- Tab: Human -->
<div class="tab-panel active" id="tab-human">
<!-- Interactive form -->
<div class="audit-form-wrapper">
<div class="audit-form">
<div class="form-row">
<label class="form-label" data-i18n="form_url_label">API Endpoint URL</label>
<input class="form-input" type="url" id="f-url" placeholder="https://your-relay.com/v1" spellcheck="false" autocomplete="off">
<div class="form-hint" data-i18n="form_url_hint">The base URL of the relay you want to audit</div>
</div>
<div class="form-row">
<label class="form-label" data-i18n="form_key_label">API Key</label>
<input class="form-input" type="password" id="f-key" placeholder="sk-..." spellcheck="false" autocomplete="off">
<div class="form-hint">
<svg width="10" height="10" viewBox="0 0 24 24" fill="none" stroke="var(--green)" stroke-width="2" style="vertical-align:middle"><path d="M12 22s8-4 8-10V5l-8-3-8 3v7c0 6 8 10 8 10z"/></svg>
<span data-i18n="form_key_hint">Stays in your browser to generate the command. When you run it, the key is sent only to your chosen relay URL.</span>
</div>
</div>
<div class="form-row">
<label class="form-label" data-i18n="form_model_label">Model</label>
<select class="form-select" id="f-model">
<option value="claude-opus-4-6">Claude Opus 4.6</option>
<option value="claude-sonnet-4-6">Claude Sonnet 4.6</option>
<option value="claude-haiku-4-5-20251001">Claude Haiku 4.5</option>
</select>
</div>
<button class="generate-btn" onclick="generateCmd()" data-i18n="form_generate">Generate Audit Command</button>
</div>
</div>
<!-- Generated command output -->
<div class="generated-cmd" id="gen-cmd">
<div class="cmd-steps">
<div class="cmd-step">
<div class="cmd-step-num">1</div>
<div class="cmd-step-text" data-i18n="cmd_step1">Open your terminal (macOS: Terminal.app / Windows: PowerShell)</div>
</div>
<div class="cmd-step">
<div class="cmd-step-num">2</div>
<div class="cmd-step-text" data-i18n="cmd_step2">Paste this command and press Enter — takes ~2 minutes:</div>
</div>
</div>
<div class="cmd-block" id="cmd-output">
<button class="copy-btn" onclick="copyCmdOutput()" data-i18n="copy_btn">Copy</button>
</div>
<div class="cmd-steps" style="margin-top:16px">
<div class="cmd-step">
<div class="cmd-step-num">3</div>
<div class="cmd-step-text" data-i18n="cmd_step3">Done! A Markdown audit report will be saved to your current directory.</div>
</div>
</div>
</div>
<!-- Terminal animation: shows what happens after paste -->
<div class="terminal-demo" id="term-demo">
<div class="term-bar">
<div class="term-dot r"></div><div class="term-dot y"></div><div class="term-dot g"></div>
<span class="term-bar-title">Terminal</span>
</div>
<div class="term-body" id="term-body"></div>
</div>
</div><!-- /tab-human -->
<!-- Tab: Agent -->
<div class="tab-panel" id="tab-agent">
<p class="agent-pitch" data-i18n="agent_pitch">Install as an OpenClaw or Hermes skill — let an agent run the local 14-step audit from a secure environment.</p>
<div class="agent-steps">
<div class="agent-step">
<div class="agent-step-head">
<div class="agent-step-num">1</div>
<div class="agent-step-title" data-i18n="agent_step1_title">Install the Skill</div>
</div>
<p data-i18n="agent_step1_desc">Use Hermes direct install now, or OpenClaw after ClawHub publication:</p>
<div class="agent-cmd" data-cmd="hermes skills install toby-bridges/api-relay-audit/skills/api-relay-audit">
<button class="copy-btn" onclick="copyAgentCmd(this)" data-i18n="copy_btn">Copy</button>
hermes skills install toby-bridges/api-relay-audit/skills/api-relay-audit
</div>
</div>
<div class="agent-step">
<div class="agent-step-head">
<div class="agent-step-num">2</div>
<div class="agent-step-title" data-i18n="agent_step2_title">Trigger with one sentence</div>
</div>
<p data-i18n="agent_step2_desc">Set the key through the agent's secure environment, then say:</p>
<div class="agent-trigger" data-i18n="agent_trigger">"Audit this relay: https://your-relay.com/v1 using API_RELAY_AUDIT_KEY"</div>
</div>
<div class="agent-step">
<div class="agent-step-head">
<div class="agent-step-num">3</div>
<div class="agent-step-title" data-i18n="agent_step3_title">Agent does everything</div>
</div>
<p data-i18n="agent_step3_desc">The agent downloads the pinned script, runs the 14-step audit locally, and presents the findings — you read the report.</p>
</div>
</div>
<p class="agent-note">
<svg width="10" height="10" viewBox="0 0 24 24" fill="none" stroke="var(--green)" stroke-width="2" style="vertical-align:middle"><path d="M12 22s8-4 8-10V5l-8-3-8 3v7c0 6 8 10 8 10z"/></svg>
<span data-i18n="agent_security">Same security model — the agent runs locally on your machine, and the key is sent only to your chosen relay URL.</span>
</p>
</div><!-- /tab-agent -->
<div style="text-align:center">
<div class="cost-badge">
<svg width="12" height="12" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2"><circle cx="12" cy="12" r="10"/><path d="M12 6v12M8 10h8M9 14h6"/></svg>
<span data-i18n="cost_note">Each audit costs ~$0.2-0.5 in API usage (billed by your provider)</span>
</div>
<p class="install-note">
<svg width="12" height="12" viewBox="0 0 24 24" fill="none" stroke="var(--green)" stroke-width="2"><path d="M12 22s8-4 8-10V5l-8-3-8 3v7c0 6 8 10 8 10z"/></svg>
<span data-i18n="install_security">Your API key is only sent to the relay URL you specify; it is not sent to API Relay Audit or an extra web checker.</span>
</p>
</div>
</section>
<!-- ── Demo Report ── -->
<section class="demo-section container" id="demo">
<h2 data-i18n="demo_title">Audit Report Demo</h2>
<p class="demo-subtitle" data-i18n="demo_sub">Redacted report examples from relay audits — click tabs to compare</p>
<div class="demo-tabs">
<button class="demo-tab active" onclick="showDemo(0)" data-rating="red"><span data-i18n="demo_tab_high">High Risk</span> — example1.com</button>
<button class="demo-tab" onclick="showDemo(1)" data-rating="yellow"><span data-i18n="demo_tab_mid">Medium Risk</span> — example2.com</button>
<button class="demo-tab" onclick="showDemo(2)" data-rating="green"><span data-i18n="demo_tab_low">Low Risk</span> — example3.com</button>
</div>
<div id="demo-cards"></div>
<p class="demo-note" data-i18n="demo_note">Domain names and sensitive details redacted; examples preserve the report shape without publishing raw relay traffic.</p>
</section>
<!-- ── How It Works ── -->
<section class="container">
<h2 style="font-size:1.4rem;font-weight:700;text-align:center" data-i18n="how_title">Detect Prompt Injection and Model Substitution Signals</h2>
<p style="text-align:center;color:var(--text2);font-size:.9rem;margin-bottom:8px" data-i18n="how_sub">Threat taxonomy based on Liu et al., "Your Agent Is Mine" (arXiv:2604.08407)</p>
<div class="steps">
<div class="step">
<div class="step-num" data-i18n="step_infra_label">Step 1-2</div>
<h3 data-i18n="step_infra_title">Infrastructure Recon</h3>
<p data-i18n="step_infra_desc">DNS, CDN, SSL certificate, management panel fingerprint, model list enumeration — understand what's behind the relay.</p>
</div>
<div class="step">
<div class="step-num" data-i18n="step_inject_label">Step 3</div>
<h3 data-i18n="step_inject_title">Token Injection (AC-1)</h3>
<p data-i18n="step_inject_desc">Compares actual token usage against expected values. Hidden system prompt injection adds extra tokens — the delta reveals it.</p>
</div>
<div class="step">
<div class="step-num" data-i18n="step_extract_label">Step 4 & 6</div>
<h3 data-i18n="step_extract_title">Prompt Extraction</h3>
<p data-i18n="step_extract_desc">3 attack vectors attempt to extract hidden system prompts: verbatim recall, translation trick, JSON continuation. Plus jailbreak resistance tests.</p>
</div>
<div class="step">
<div class="step-num" data-i18n="step_id_label">Step 5</div>
<h3 data-i18n="step_id_title">Identity Consistency Signals</h3>
<p data-i18n="step_id_desc">An identity keyword set checks whether a claimed Claude route leaks GPT, DeepSeek, GLM, Qwen, or other model identities. This is an evidence signal, not standalone provider-level proof.</p>
</div>
<div class="step">
<div class="step-num" data-i18n="step_ctx_label">Step 7</div>
<h3 data-i18n="step_ctx_title">Context Truncation</h3>
<p data-i18n="step_ctx_desc">5 canary markers + binary search pinpoint the real context window boundary. Is your 200K context really 200K?</p>
</div>
<div class="step">
<div class="step-num" data-i18n="step_tool_label">Step 8 (AC-1.a)</div>
<h3 data-i18n="step_tool_title">Tool-Call Rewriting</h3>
<p data-i18n="step_tool_desc">Checks if the relay silently modifies package install commands in responses — typosquatting supply-chain attacks at the proxy layer.</p>
</div>
<div class="step">
<div class="step-num" data-i18n="step_err_label">Step 9 (AC-2)</div>
<h3 data-i18n="step_err_title">Error Response Leakage</h3>
<p data-i18n="step_err_desc">7 deliberately broken requests probe for API key, env vars, file paths, and LiteLLM internals leaking in error responses.</p>
</div>
<div class="step">
<div class="step-num" data-i18n="step_sse_label">Step 10-11</div>
<h3 data-i18n="step_sse_title">Stream Integrity & Web3</h3>
<p data-i18n="step_sse_desc">SSE event whitelist, usage monotonicity, thinking signature validity, model identity check. Plus Web3 signature-isolation probes (profile-gated).</p>
</div>
</div>
</section>
<!-- ── Comparison ── -->
<section class="compare-section container" id="compare">
<h2 data-i18n="cmp_title">How We Compare</h2>
<p class="compare-sub" data-i18n="cmp_sub">Three tools, three approaches — pick the right one for your needs</p>
<div style="overflow-x:auto">
<table class="compare-table">
<thead>
<tr>
<th data-i18n="cmp_col_dim">Dimension</th>
<th class="ours">api-relay-audit</th>
<th>hvoy.ai</th>
<th>cctest.ai</th>
</tr>
</thead>
<tbody>
<tr><td data-i18n="cmp_inject">Token Injection</td><td class="check">✓</td><td class="cross">✕</td><td class="check">✓</td></tr>
<tr><td data-i18n="cmp_prompt">Prompt Extraction</td><td class="check">✓</td><td class="cross">✕</td><td class="cross">✕</td></tr>
<tr><td data-i18n="cmp_identity">Identity Substitution</td><td class="check">✓</td><td class="check">✓</td><td class="cross">✕</td></tr>
<tr><td data-i18n="cmp_jailbreak">Jailbreak Resistance</td><td class="check">✓</td><td class="cross">✕</td><td class="cross">✕</td></tr>
<tr><td data-i18n="cmp_context">Context Truncation</td><td class="check">✓</td><td class="cross">✕</td><td class="cross">✕</td></tr>
<tr><td data-i18n="cmp_tool">Tool-Call Rewriting (AC-1.a)</td><td class="check">✓</td><td class="cross">✕</td><td class="cross">✕</td></tr>
<tr><td data-i18n="cmp_error">Error Response Leakage (AC-2)</td><td class="check">✓</td><td class="cross">✕</td><td class="cross">✕</td></tr>
<tr><td data-i18n="cmp_stream">Stream Integrity (SSE)</td><td class="check">✓</td><td class="check">✓</td><td class="cross">✕</td></tr>
<tr><td data-i18n="cmp_web3">Web3 Injection</td><td class="check">✓</td><td class="cross">✕</td><td class="cross">✕</td></tr>
<tr><td data-i18n="cmp_channel">Upstream Channel Classifier</td><td class="check">✓</td><td class="cross">✕</td><td class="check">✓</td></tr>
<tr><td data-i18n="cmp_local">Local Execution (No extra checker server)</td><td class="check">✓</td><td class="cross">✕</td><td class="cross">✕</td></tr>
<tr><td data-i18n="cmp_oss">Fully Open Source</td><td class="check">✓</td><td class="cross">Partial</td><td class="cross">✕</td></tr>
<tr><td data-i18n="cmp_leaderboard">Public Leaderboard</td><td class="cross">✕</td><td class="check">✓</td><td class="cross">✕</td></tr>
<tr><td data-i18n="cmp_report">Structured Audit Report</td><td class="check">✓</td><td class="cross">✕</td><td class="cross">✕</td></tr>
</tbody>
</table>
</div>
</section>
<!-- ── Guides ── -->
<section class="guides-section container" id="guides">
<h2>Guides for AI API Relay Security</h2>
<p class="guides-sub">Short, citation-friendly pages for GitHub, Google, and AI summaries.</p>
<div class="guide-grid">
<a class="guide-card" href="guides/what-is-ai-api-relay-proxy.html">
<h3>What is an AI API relay or LLM proxy?</h3>
<p>Define the trust boundary and what an intermediary can change.</p>
</a>
<a class="guide-card" href="guides/audit-claude-api-relay-safely.html">
<h3>How to audit a Claude API relay safely</h3>
<p>Run a local audit without adding another API-key trust hop.</p>
</a>
<a class="guide-card" href="guides/claude-code-anthropic-base-url-prompt-steganography.html">
<h3>Claude Code, ANTHROPIC_BASE_URL, and prompt steganography</h3>
<p>Separate reported client markers from relay behavior, then audit the gateway trust boundary.</p>
</a>
<a class="guide-card" href="guides/compare-api-relay-audit-hvoy-cctest.html">
<h3>api-relay-audit vs hvoy.ai vs cctest.ai</h3>
<p>Compare local audits, relay lookup, and web-based checks.</p>
</a>
<a class="guide-card" href="guides/detect-prompt-injection-llm-api-proxies.html">
<h3>Detecting prompt injection in LLM API proxies</h3>
<p>Understand token deltas, extraction probes, and identity signals.</p>
</a>
<a class="guide-card" href="guides/web3-wallet-prompt-injection-ai-agents.html">
<h3>Web3 wallet prompt injection through AI agents</h3>
<p>Check transfer guidance, signed-transaction refusal, and private-key refusal.</p>
</a>
<a class="guide-card" href="guides/openclaw-hermes-skill-api-relay-audit.html">
<h3>OpenClaw and Hermes skill for relay audits</h3>
<p>Run local AI API relay audits from agent workflows without adding a web checker.</p>
</a>
</div>
</section>
<!-- ── FAQ ── -->
<section class="faq-section container" id="faq">
<h2 data-i18n="faq_title">FAQ</h2>
<div class="faq-item">
<div class="faq-q" onclick="toggleFaq(this)" data-i18n="faq_q1">What is an API relay or LLM proxy?</div>
<div class="faq-a" data-i18n="faq_a1">An API relay or LLM proxy is a third-party service between you and an AI provider such as Anthropic or OpenAI. It forwards your requests upstream, but it can also inject hidden instructions, swap models, truncate context, rewrite tool output, or leak credentials in error responses.</div>
</div>
<div class="faq-item">
<div class="faq-q" onclick="toggleFaq(this)" data-i18n="faq_q2">Is it safe to enter my API Key?</div>
<div class="faq-a" data-i18n="faq_a2">API Relay Audit runs locally, so your API key is sent only to the relay URL you specify. No data is sent to API Relay Audit servers, and the standalone version has zero Python package dependencies.</div>
</div>
<div class="faq-item">
<div class="faq-q" onclick="toggleFaq(this)" data-i18n="faq_q3">What does prompt injection mean here?</div>
<div class="faq-a" data-i18n="faq_a3">Prompt injection means the relay may prepend or insert hidden instructions into your request. API Relay Audit compares token usage, runs extraction probes, and records evidence when hidden prompt content appears.</div>
</div>
<div class="faq-item">
<div class="faq-q" onclick="toggleFaq(this)" data-i18n="faq_q4">What is model substitution?</div>
<div class="faq-a" data-i18n="faq_a4">Model substitution means the relay may expose signals for a different model identity, route, or upstream channel than expected. API Relay Audit checks self-ID, stream model identity, latency, and channel fingerprints as signals that require corroboration before provider-level claims.</div>
</div>
<div class="faq-item">
<div class="faq-q" onclick="toggleFaq(this)" data-i18n="faq_q5">What is tool-call rewriting?</div>
<div class="faq-a" data-i18n="faq_a5">Tool-call rewriting means the relay modifies package-install commands or tool-like output in the model response. API Relay Audit compares pinned package commands against returned text to detect proxy-layer supply-chain tampering.</div>
</div>
<div class="faq-item">
<div class="faq-q" onclick="toggleFaq(this)" data-i18n="faq_q6">What are SSE anomalies?</div>
<div class="faq-a" data-i18n="faq_a6">SSE anomalies are stream-level integrity issues in Anthropic-style streaming responses. API Relay Audit checks event types, usage monotonicity, thinking signatures, and stream model identity when supported.</div>
</div>
<div class="faq-item">
<div class="faq-q" onclick="toggleFaq(this)" data-i18n="faq_q7">What Web3 wallet risks does it check?</div>
<div class="faq-a" data-i18n="faq_a7">With the web3 or full profile, API Relay Audit checks transfer guidance, signed-transaction refusal, and private-key refusal behavior before wallet-related traffic is trusted.</div>
</div>
<div class="faq-item">
<div class="faq-q" onclick="toggleFaq(this)" data-i18n="faq_q8">What does inconclusive mean?</div>
<div class="faq-a" data-i18n="faq_a8">Inconclusive means the tool could not determine a clean or anomalous result for that step. Blocked probes, unsupported formats, and ambiguous responses are not treated as clean.</div>
</div>
</section>
<footer>
<p>
<a href="https://github.com/toby-bridges/api-relay-audit" target="_blank" rel="noopener noreferrer">GitHub</a> ·
<a href="zh/">中文</a> ·
<a href="sitemap.xml">Sitemap</a> ·
<a href="https://x.com/li9292" target="_blank" rel="noopener noreferrer">X @li9292</a> ·
AGPL-3.0-only License ·
<span data-i18n="footer_threat">Threat model: </span><a href="https://arxiv.org/abs/2604.08407" target="_blank" rel="noopener noreferrer">arXiv:2604.08407</a>
</p>
</footer>
<script>
// ── Demo Data ──
const demos = [
{
domain:"example1.com", rating:"red", ratingLabel:"HIGH RISK",
ratingLabel_zh:"高风险", ratingLabel_ja:"高リスク", ratingLabel_ko:"고위험",
testDate:"2026-02-25",
infra:{cdn:"Unknown",system:"Unknown",ssl:"Unknown"},
tokenInjection:{delta:3200,verdict:"Severe injection",verdict_zh:"严重注入",verdict_ja:"深刻な注入",verdict_ko:"심각한 주입"},
promptExtraction:{leaked:0,total:0},
toolSub:{detected:false,probes:0},
errorLeak:{severity:"none"},
contextLength:{maxTokens:"N/A"},
redFlags:["Injected 3200 tokens of Kiro identity","2-layer proxy architecture","Severe request tampering"],
redFlags_zh:["注入 3200 tokens Kiro 身份信息","2 层代理架构","严重篡改用户请求"],
summary:"Injects 3200 tokens of Kiro identity prompt, routed through 2 proxy layers. Severe request tampering risk.",
summary_zh:"注入 3200 tokens 的 Kiro 身份 prompt，经过 2 层代理转发，对用户请求有严重篡改风险。"
},
{
domain:"example2.com", rating:"yellow", ratingLabel:"MEDIUM RISK",
ratingLabel_zh:"低中风险", ratingLabel_ja:"中リスク", ratingLabel_ko:"중간위험",
testDate:"2026-02-26",
infra:{cdn:"Alibaba Cloud ESA (HK)",system:"New API v0.10.9",ssl:"Let's Encrypt"},
tokenInjection:{delta:80,verdict:"Injection (Claude Code CLI identity)",verdict_zh:"有注入（Claude Code CLI 身份）",verdict_ja:"注入あり",verdict_ko:"주입 감지"},
promptExtraction:{leaked:2,total:6,methods:["Translation","JSON continuation"]},
toolSub:{detected:false,probes:5},
errorLeak:{severity:"none"},
contextLength:{maxTokens:"~200K"},
redFlags:["Hidden system prompt ~80 tokens","Multi-domain shared cert","Custom system prompt causes 422 conflict"],
redFlags_zh:["隐藏 system prompt 注入 ~80 tokens","多域名共享证书","自定义 system prompt 导致 422 冲突"],
summary:"Injects ~80 tokens of Claude Code CLI identity. Translation and JSON continuation methods can extract prompt. Context ~200K tokens intact.",
summary_zh:"注入约 80 tokens 的 Claude Code CLI 身份指令。翻译法和 JSON 接龙法可提取 prompt。上下文 ~200K tokens 完整。",
tests:[
{step:"Token Injection",result:"warn",detail:"+80 tokens"},
{step:"Prompt Extraction",result:"warn",detail:"2/6 extracted"},
{step:"Identity",result:"fail",detail:"422 error"},
{step:"Context Length",result:"pass",detail:"~200K OK"},
{step:"Tool-Call (AC-1.a)",result:"pass",detail:"5/5 intact"},
{step:"Error Leakage (AC-2)",result:"pass",detail:"Clean"},
]
},
{
domain:"example3.com", rating:"green", ratingLabel:"LOW RISK",
ratingLabel_zh:"低风险", ratingLabel_ja:"低リスク", ratingLabel_ko:"저위험",
testDate:"2026-02-26",
infra:{cdn:"Cloudflare (AMS)",system:"New API (self-hosted)",ssl:"Google Trust Services"},
tokenInjection:{delta:25,verdict:"Minor (likely system default)",verdict_zh:"轻微注入（疑似系统默认）",verdict_ja:"軽微（システムデフォルトの可能性）",verdict_ko:"경미(시스템 기본값 추정)"},
promptExtraction:{leaked:1,total:6,methods:["Verbatim recall"]},
toolSub:{detected:false,probes:5},
errorLeak:{severity:"none"},
contextLength:{maxTokens:"~197K"},
redFlags:["Minor ~25 tokens (likely New API default behavior)"],
redFlags_zh:["轻微注入 ~25 tokens（疑似 New API 系统默认行为）"],
summary:"Self-hosted relay, minor ~25 token injection (likely New API default). Cat-test and identity test passed. Context ~197K tokens intact. Overall safe and reliable.",
summary_zh:"自建中转站，轻微注入约 25 tokens（可能是 New API 默认行为）。猫叫和身份测试通过，上下文 ~197K tokens 完整。整体安全可靠。",
tests:[
{step:"Token Injection",result:"pass",detail:"+25 (default)"},
{step:"Prompt Extraction",result:"pass",detail:"1/6 (minor)"},
{step:"Identity",result:"pass",detail:"Claude by Anthropic"},
{step:"Context Length",result:"pass",detail:"~197K OK"},
{step:"Tool-Call (AC-1.a)",result:"pass",detail:"5/5 intact"},
{step:"Error Leakage (AC-2)",result:"pass",detail:"Clean"},
]
}
];
function esc(s){return s==null?'—':String(s).replace(/</g,'<').replace(/>/g,'>')}
function renderDemo(i){
const d=demos[i];
const lang=currentLang;
const rl = d['ratingLabel_'+lang]||d.ratingLabel;
const vd = d.tokenInjection['verdict_'+lang]||d.tokenInjection.verdict;
const sm = d['summary_'+lang]||d.summary;
const rf = d['redFlags_'+lang]||d.redFlags;
const testsHtml = d.tests ? `
<div class="demo-detail">
<table class="demo-table">
<tr><th>${t('demo_th_step')}</th><th>${t('demo_th_result')}</th><th>${t('demo_th_detail')}</th></tr>
${d.tests.map(t=>`<tr>
<td>${esc(t.step)}</td>
<td class="${t.result==='pass'?'pass':t.result==='warn'?'warn-text':'fail'}">${t.result==='pass'?'✓ Clean':t.result==='warn'?'⚠ Warning':'✕ Issue'}</td>
<td>${esc(t.detail)}</td>
</tr>`).join('')}
</table>
</div>` : '';
return `
<div class="demo-card active" id="demo-${i}">
<div class="demo-bar ${d.rating}"></div>
<div class="demo-head">
<div>
<div class="demo-domain">${esc(d.domain)}</div>
<div style="color:var(--text3);font-size:.8rem;margin-top:4px">${esc(d.infra.system)}</div>
</div>
<span class="demo-badge ${d.rating}">${esc(rl)}</span>
</div>
<div class="demo-meta">
<span>${t('demo_date')}: ${esc(d.testDate)}</span>
<span>CDN: ${esc(d.infra.cdn)}</span>
<span>SSL: ${esc(d.infra.ssl)}</span>
</div>
<div class="demo-grid">
<div class="demo-metric">
<div class="demo-metric-label">Token Injection</div>
<div class="demo-metric-value ${d.tokenInjection.delta>500?'danger':d.tokenInjection.delta>50?'warn':'ok'}">${d.tokenInjection.delta>0?'+'+d.tokenInjection.delta+' tokens':t('demo_no_inject')}</div>
<div style="color:var(--text3);font-size:.72rem;margin-top:4px">${esc(vd)}</div>
</div>
<div class="demo-metric">
<div class="demo-metric-label">Prompt Extraction</div>
<div class="demo-metric-value ${d.promptExtraction.leaked>1?'warn':d.promptExtraction.leaked>0?'warn':'ok'}">${d.promptExtraction.leaked}/${d.promptExtraction.total} ${t('demo_extracted')}</div>
</div>
<div class="demo-metric">
<div class="demo-metric-label">Context Length</div>
<div class="demo-metric-value ok">${esc(d.contextLength.maxTokens)}</div>
</div>
<div class="demo-metric">
<div class="demo-metric-label">${t('demo_risk_items')}</div>
<div class="demo-metric-value ${rf.length>2?'danger':rf.length>0?'warn':'ok'}">${rf.length}</div>
</div>
</div>
${testsHtml}
<div class="demo-detail">
<div style="color:var(--text2);font-size:.85rem;line-height:1.7;background:rgba(255,255,255,.02);padding:14px 16px;border-radius:8px;border-left:3px solid var(--accent)">${esc(sm)}</div>
</div>
${rf.length?`<div class="demo-flags">${rf.map(f=>`<span class="flag">${esc(f)}</span>`).join('')}</div>`:''}
</div>`;
}
let currentDemo=0;
function showDemo(i){
currentDemo=i;
document.querySelectorAll('.demo-tab').forEach((t,j)=>t.classList.toggle('active',j===i));
document.getElementById('demo-cards').innerHTML=renderDemo(i);
}
// ── Audit Tab Switching ──
function switchAuditTab(tab){
document.querySelectorAll('.audit-tab').forEach((b,i)=>b.classList.toggle('active',i===(tab==='human'?0:1)));
document.getElementById('tab-human').classList.toggle('active',tab==='human');
document.getElementById('tab-agent').classList.toggle('active',tab==='agent');
}
function copyAgentCmd(btn){
const block=btn.parentElement;
const text=block.getAttribute('data-cmd')||block.innerText.replace(t('copy_btn'),'').replace(t('copied_btn'),'').trim();
navigator.clipboard.writeText(text).then(()=>{
btn.textContent=t('copied_btn');
setTimeout(()=>btn.textContent=t('copy_btn'),1500);
});
}
// ── Terminal Animation ──
function playTerminalDemo(userUrl){
const el=document.getElementById('term-demo');
const body=document.getElementById('term-body');
el.classList.add('visible');
body.innerHTML='';
const domain=userUrl.replace(/^https?:\/\//,'').replace(/\/.*$/,'');
const lines=[
{text:'$ curl -sO https://raw.githubusercontent.com/.../audit.py',cls:'prompt',delay:0},
{text:t('term_downloading'),cls:'out',delay:600},
{text:'$ python3 audit.py --key \'sk-...\' --url \''+esc(domain)+'/v1\'',cls:'prompt',delay:1200},
{text:'',cls:'out',delay:1600},
{text:t('term_step1'),cls:'highlight',delay:2000},
{text:t('term_step1_out'),cls:'out',delay:2600},
{text:t('term_step3'),cls:'highlight',delay:3200},
{text:t('term_step3_out'),cls:'warn',delay:3800},
{text:t('term_step5'),cls:'highlight',delay:4400},
{text:t('term_step5_out'),cls:'ok',delay:5000},
{text:t('term_step8'),cls:'highlight',delay:5600},
{text:t('term_step8_out'),cls:'ok',delay:6200},
{text:'...',cls:'out',delay:6600},
{text:'',cls:'out',delay:7000},
{text:t('term_done').replace('{domain}',domain),cls:'ok',delay:7400},
{text:t('term_open_hint'),cls:'out',delay:8000},
];
lines.forEach(l=>{
setTimeout(()=>{
const div=document.createElement('div');
div.className='term-line';
div.style.animationDelay='0s';
if(l.cls==='prompt') div.innerHTML='<span class="prompt">'+l.text+'</span>';
else if(l.cls==='highlight') div.innerHTML='<span class="highlight">'+l.text+'</span>';
else if(l.cls==='warn') div.innerHTML='<span class="warn">'+l.text+'</span>';
else if(l.cls==='ok') div.innerHTML='<span class="ok">'+l.text+'</span>';
else div.innerHTML='<span class="out">'+l.text+'</span>';
body.appendChild(div);
body.scrollTop=body.scrollHeight;
},l.delay);
});
}
// ── FAQ Toggle ──
function toggleFaq(el){
el.parentElement.classList.toggle('open');
}
// ── Form & Command Generator ──
// Shell-safe escaping: wrap in single quotes, escape inner single quotes
function shellEscape(s){
return "'"+s.replace(/'/g,"'\\''")+"'";
}
function isWindows(){
return /Win/i.test(navigator.platform||navigator.userAgent);
}
function generateCmd(){
const url=document.getElementById('f-url').value.trim();
const key=document.getElementById('f-key').value.trim();
const model=document.getElementById('f-model').value;
if(!url){document.getElementById('f-url').focus();return}
if(!key){document.getElementById('f-key').focus();return}
const win=isWindows();
const scriptRef='v2.3.0';
let cmd;
if(win){
// PowerShell: Invoke-WebRequest, semicolon, single-quote escaping via doubling
const psKey=key.replace(/'/g,"''");
const psUrl=url.replace(/'/g,"''");