@highflame/policy 2.2.35 → 2.2.37

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
Files changed (78) hide show
  1. package/_schemas/agent_ops/context.json +822 -0
  2. package/_schemas/agent_ops/schema.cedarschema +134 -109
  3. package/_schemas/agent_ops/templates/ported/agent-security/agent_identity_a2a_cross_origin.cedar +6 -4
  4. package/_schemas/agent_ops/templates/ported/agent-security/agent_identity_a2a_escalation.cedar +5 -4
  5. package/_schemas/agent_ops/templates/ported/agent-security/agent_identity_a2a_inter_agent_injection.cedar +9 -8
  6. package/_schemas/agent_ops/templates/ported/agent-security/agent_identity_claude_block_injection.cedar +2 -2
  7. package/_schemas/agent_ops/templates/ported/agent-security/agent_identity_defaults.cedar +3 -3
  8. package/_schemas/agent_ops/templates/ported/agent-security/agent_identity_multi_agent_trust.cedar +6 -6
  9. package/_schemas/agent_ops/templates/ported/content-safety/trust_safety_chat_assistant.cedar +5 -4
  10. package/_schemas/agent_ops/templates/ported/content-safety/trust_safety_defaults.cedar +18 -12
  11. package/_schemas/agent_ops/templates/ported/content-safety/trust_safety_semantic.cedar +6 -6
  12. package/_schemas/agent_ops/templates/ported/content-safety/trust_safety_toxicity.cedar +4 -4
  13. package/_schemas/agent_ops/templates/ported/data-protection/data_protection_advanced_secrets.cedar +4 -2
  14. package/_schemas/agent_ops/templates/ported/data-protection/data_protection_data_pipeline.cedar +2 -1
  15. package/_schemas/agent_ops/templates/ported/data-protection/data_protection_defaults.cedar +4 -2
  16. package/_schemas/agent_ops/templates/ported/data-protection/privacy_advanced_pii.cedar +42 -21
  17. package/_schemas/agent_ops/templates/ported/data-protection/privacy_defaults.cedar +14 -7
  18. package/_schemas/agent_ops/templates/ported/data-protection/privacy_pii_model.cedar +10 -5
  19. package/_schemas/agent_ops/templates/ported/organization/organization_team_permissions.cedar +3 -2
  20. package/_schemas/agent_ops/templates/ported/threat-detection/security_advanced_block_critical_severity.cedar +3 -3
  21. package/_schemas/agent_ops/templates/ported/threat-detection/security_chat_assistant.cedar +5 -4
  22. package/_schemas/agent_ops/templates/ported/threat-detection/security_data_pipeline_block_injection.cedar +3 -2
  23. package/_schemas/agent_ops/templates/ported/threat-detection/security_injection.cedar +9 -8
  24. package/_schemas/agent_ops/templates/ported/threat-detection/security_patterns.cedar +5 -3
  25. package/_schemas/agent_ops/templates/ported/threat-detection/semantic_defaults.cedar +9 -7
  26. package/_schemas/ai_gateway/context.json +308 -0
  27. package/_schemas/ai_gateway/schema.cedarschema +44 -7
  28. package/_schemas/ai_gateway/templates/defaults/agent_security.cedar +5 -16
  29. package/_schemas/ai_gateway/templates/defaults/pii.cedar +14 -7
  30. package/_schemas/ai_gateway/templates/defaults/pii_advanced.cedar +42 -21
  31. package/_schemas/ai_gateway/templates/defaults/pii_model.cedar +10 -5
  32. package/_schemas/ai_gateway/templates/defaults/semantic.cedar +9 -74
  33. package/_schemas/ai_gateway/templates/defaults/tools.cedar +7 -38
  34. package/_schemas/ai_gateway/templates/secrets.cedar +3 -3
  35. package/_schemas/guardrails/context.json +701 -0
  36. package/_schemas/guardrails/schema.cedarschema +30 -2
  37. package/_schemas/guardrails/templates/defaults/agent_identity.cedar +3 -3
  38. package/_schemas/guardrails/templates/defaults/injection.cedar +9 -8
  39. package/_schemas/guardrails/templates/defaults/pii.cedar +14 -7
  40. package/_schemas/guardrails/templates/defaults/pii_model.cedar +10 -5
  41. package/_schemas/guardrails/templates/defaults/secrets.cedar +4 -2
  42. package/_schemas/guardrails/templates/defaults/security_patterns.cedar +5 -3
  43. package/_schemas/guardrails/templates/defaults/semantic.cedar +6 -6
  44. package/_schemas/guardrails/templates/defaults/toxicity.cedar +6 -6
  45. package/_schemas/guardrails/templates/profiles/a2a_security/cross_origin.cedar +6 -4
  46. package/_schemas/guardrails/templates/profiles/a2a_security/escalation_detection.cedar +5 -4
  47. package/_schemas/guardrails/templates/profiles/a2a_security/inter_agent_injection.cedar +9 -8
  48. package/_schemas/guardrails/templates/profiles/advanced_detection/pii.cedar +42 -21
  49. package/_schemas/guardrails/templates/profiles/advanced_detection/secrets.cedar +4 -2
  50. package/_schemas/guardrails/templates/profiles/chat_assistant/security.cedar +5 -4
  51. package/_schemas/guardrails/templates/profiles/chat_assistant/trust_safety.cedar +5 -4
  52. package/_schemas/guardrails/templates/profiles/data_pipeline/data_protection.cedar +2 -1
  53. package/_schemas/guardrails/templates/profiles/data_pipeline/security.cedar +3 -2
  54. package/_schemas/guardrails/templates/profiles/multi_agent/agent_trust.cedar +6 -6
  55. package/_schemas/guardrails/templates/templates.json +0 -9
  56. package/_schemas/overwatch/context.json +30 -0
  57. package/_schemas/overwatch/schema.cedarschema +7 -0
  58. package/dist/agent_ops-context.gen.d.ts +6 -1
  59. package/dist/agent_ops-context.gen.js +10 -0
  60. package/dist/agent_ops-defaults.gen.js +177 -120
  61. package/dist/agent_ops-entities.gen.js +5 -1
  62. package/dist/ai_gateway-context.gen.d.ts +6 -1
  63. package/dist/ai_gateway-context.gen.js +10 -0
  64. package/dist/ai_gateway-defaults.gen.js +90 -164
  65. package/dist/ai_gateway-detectors.gen.js +6 -1
  66. package/dist/ai_gateway-entities.gen.js +6 -2
  67. package/dist/guardrails-context.gen.d.ts +6 -1
  68. package/dist/guardrails-context.gen.js +10 -0
  69. package/dist/guardrails-defaults.gen.js +144 -145
  70. package/dist/guardrails-detectors.gen.js +6 -1
  71. package/dist/guardrails-entities.gen.js +5 -1
  72. package/dist/overwatch-context.gen.d.ts +6 -1
  73. package/dist/overwatch-context.gen.js +10 -0
  74. package/dist/overwatch-detectors.gen.js +6 -1
  75. package/dist/service-schemas.gen.d.ts +4 -4
  76. package/dist/service-schemas.gen.js +528 -118
  77. package/package.json +1 -1
  78. package/_schemas/guardrails/templates/profiles/advanced_detection/threat_severity.cedar +0 -30
@@ -22,7 +22,8 @@
22
22
  @reject_message("Content blocked: national identifier (SSN, passport, or driver's license) patterns detected.")
23
23
  forbid (
24
24
  principal,
25
- action in [Guardrails::Action::"process_prompt", Guardrails::Action::"call_tool", Guardrails::Action::"read_file", Guardrails::Action::"write_file"],
25
+ action in [Guardrails::Action::"process_prompt",
26
+ Guardrails::Action::"process_response", Guardrails::Action::"call_tool", Guardrails::Action::"read_file", Guardrails::Action::"write_file"],
26
27
  resource
27
28
  )
28
29
  when {
@@ -44,7 +45,8 @@ when {
44
45
  @reject_message("Content blocked: credit card number patterns detected.")
45
46
  forbid (
46
47
  principal,
47
- action in [Guardrails::Action::"process_prompt", Guardrails::Action::"call_tool", Guardrails::Action::"read_file", Guardrails::Action::"write_file"],
48
+ action in [Guardrails::Action::"process_prompt",
49
+ Guardrails::Action::"process_response", Guardrails::Action::"call_tool", Guardrails::Action::"read_file", Guardrails::Action::"write_file"],
48
50
  resource
49
51
  )
50
52
  when {
@@ -63,7 +65,8 @@ when {
63
65
  @reject_message("Content blocked: bank account (IBAN) number patterns detected.")
64
66
  forbid (
65
67
  principal,
66
- action in [Guardrails::Action::"process_prompt", Guardrails::Action::"call_tool", Guardrails::Action::"read_file", Guardrails::Action::"write_file"],
68
+ action in [Guardrails::Action::"process_prompt",
69
+ Guardrails::Action::"process_response", Guardrails::Action::"call_tool", Guardrails::Action::"read_file", Guardrails::Action::"write_file"],
67
70
  resource
68
71
  )
69
72
  when {
@@ -85,7 +88,8 @@ when {
85
88
  @reject_message("Content blocked: medical record identifier patterns detected.")
86
89
  forbid (
87
90
  principal,
88
- action in [Guardrails::Action::"process_prompt", Guardrails::Action::"call_tool", Guardrails::Action::"read_file", Guardrails::Action::"write_file"],
91
+ action in [Guardrails::Action::"process_prompt",
92
+ Guardrails::Action::"process_response", Guardrails::Action::"call_tool", Guardrails::Action::"read_file", Guardrails::Action::"write_file"],
89
93
  resource
90
94
  )
91
95
  when {
@@ -109,7 +113,8 @@ when {
109
113
  @reject_message("Content blocked: tax identifier (ITIN or EIN) patterns detected.")
110
114
  forbid (
111
115
  principal,
112
- action in [Guardrails::Action::"process_prompt", Guardrails::Action::"call_tool", Guardrails::Action::"read_file", Guardrails::Action::"write_file"],
116
+ action in [Guardrails::Action::"process_prompt",
117
+ Guardrails::Action::"process_response", Guardrails::Action::"call_tool", Guardrails::Action::"read_file", Guardrails::Action::"write_file"],
113
118
  resource
114
119
  )
115
120
  when {
@@ -129,7 +134,8 @@ when {
129
134
  @reject_message("Content blocked: credential or API key patterns detected.")
130
135
  forbid (
131
136
  principal,
132
- action in [Guardrails::Action::"process_prompt", Guardrails::Action::"call_tool", Guardrails::Action::"read_file", Guardrails::Action::"write_file"],
137
+ action in [Guardrails::Action::"process_prompt",
138
+ Guardrails::Action::"process_response", Guardrails::Action::"call_tool", Guardrails::Action::"read_file", Guardrails::Action::"write_file"],
133
139
  resource
134
140
  )
135
141
  when {
@@ -160,7 +166,8 @@ when {
160
166
  @reject_message("Content blocked: cryptocurrency wallet address patterns detected.")
161
167
  forbid (
162
168
  principal,
163
- action in [Guardrails::Action::"process_prompt", Guardrails::Action::"call_tool", Guardrails::Action::"read_file", Guardrails::Action::"write_file"],
169
+ action in [Guardrails::Action::"process_prompt",
170
+ Guardrails::Action::"process_response", Guardrails::Action::"call_tool", Guardrails::Action::"read_file", Guardrails::Action::"write_file"],
164
171
  resource
165
172
  )
166
173
  when {
@@ -179,7 +186,8 @@ when {
179
186
  @reject_message("Content blocked: personal name patterns detected.")
180
187
  forbid (
181
188
  principal,
182
- action in [Guardrails::Action::"process_prompt", Guardrails::Action::"call_tool", Guardrails::Action::"read_file", Guardrails::Action::"write_file"],
189
+ action in [Guardrails::Action::"process_prompt",
190
+ Guardrails::Action::"process_response", Guardrails::Action::"call_tool", Guardrails::Action::"read_file", Guardrails::Action::"write_file"],
183
191
  resource
184
192
  )
185
193
  when {
@@ -198,7 +206,8 @@ when {
198
206
  @reject_message("Content blocked: email address patterns detected.")
199
207
  forbid (
200
208
  principal,
201
- action in [Guardrails::Action::"process_prompt", Guardrails::Action::"call_tool", Guardrails::Action::"read_file", Guardrails::Action::"write_file"],
209
+ action in [Guardrails::Action::"process_prompt",
210
+ Guardrails::Action::"process_response", Guardrails::Action::"call_tool", Guardrails::Action::"read_file", Guardrails::Action::"write_file"],
202
211
  resource
203
212
  )
204
213
  when {
@@ -217,7 +226,8 @@ when {
217
226
  @reject_message("Content blocked: phone number patterns detected.")
218
227
  forbid (
219
228
  principal,
220
- action in [Guardrails::Action::"process_prompt", Guardrails::Action::"call_tool", Guardrails::Action::"read_file", Guardrails::Action::"write_file"],
229
+ action in [Guardrails::Action::"process_prompt",
230
+ Guardrails::Action::"process_response", Guardrails::Action::"call_tool", Guardrails::Action::"read_file", Guardrails::Action::"write_file"],
221
231
  resource
222
232
  )
223
233
  when {
@@ -236,7 +246,8 @@ when {
236
246
  @reject_message("Content blocked: date or time patterns detected.")
237
247
  forbid (
238
248
  principal,
239
- action in [Guardrails::Action::"process_prompt", Guardrails::Action::"call_tool", Guardrails::Action::"read_file", Guardrails::Action::"write_file"],
249
+ action in [Guardrails::Action::"process_prompt",
250
+ Guardrails::Action::"process_response", Guardrails::Action::"call_tool", Guardrails::Action::"read_file", Guardrails::Action::"write_file"],
240
251
  resource
241
252
  )
242
253
  when {
@@ -258,7 +269,8 @@ when {
258
269
  @reject_message("Content blocked: URL patterns detected.")
259
270
  forbid (
260
271
  principal,
261
- action in [Guardrails::Action::"process_prompt", Guardrails::Action::"call_tool", Guardrails::Action::"read_file", Guardrails::Action::"write_file"],
272
+ action in [Guardrails::Action::"process_prompt",
273
+ Guardrails::Action::"process_response", Guardrails::Action::"call_tool", Guardrails::Action::"read_file", Guardrails::Action::"write_file"],
262
274
  resource
263
275
  )
264
276
  when {
@@ -277,7 +289,8 @@ when {
277
289
  @reject_message("Content blocked: street address or location patterns detected.")
278
290
  forbid (
279
291
  principal,
280
- action in [Guardrails::Action::"process_prompt", Guardrails::Action::"call_tool", Guardrails::Action::"read_file", Guardrails::Action::"write_file"],
292
+ action in [Guardrails::Action::"process_prompt",
293
+ Guardrails::Action::"process_response", Guardrails::Action::"call_tool", Guardrails::Action::"read_file", Guardrails::Action::"write_file"],
281
294
  resource
282
295
  )
283
296
  when {
@@ -298,7 +311,8 @@ when {
298
311
  @reject_message("Content blocked: organization name patterns detected.")
299
312
  forbid (
300
313
  principal,
301
- action in [Guardrails::Action::"process_prompt", Guardrails::Action::"call_tool", Guardrails::Action::"read_file", Guardrails::Action::"write_file"],
314
+ action in [Guardrails::Action::"process_prompt",
315
+ Guardrails::Action::"process_response", Guardrails::Action::"call_tool", Guardrails::Action::"read_file", Guardrails::Action::"write_file"],
302
316
  resource
303
317
  )
304
318
  when {
@@ -317,7 +331,8 @@ when {
317
331
  @reject_message("Content blocked: occupation patterns detected.")
318
332
  forbid (
319
333
  principal,
320
- action in [Guardrails::Action::"process_prompt", Guardrails::Action::"call_tool", Guardrails::Action::"read_file", Guardrails::Action::"write_file"],
334
+ action in [Guardrails::Action::"process_prompt",
335
+ Guardrails::Action::"process_response", Guardrails::Action::"call_tool", Guardrails::Action::"read_file", Guardrails::Action::"write_file"],
321
336
  resource
322
337
  )
323
338
  when {
@@ -336,7 +351,8 @@ when {
336
351
  @reject_message("Content blocked: username patterns detected.")
337
352
  forbid (
338
353
  principal,
339
- action in [Guardrails::Action::"process_prompt", Guardrails::Action::"call_tool", Guardrails::Action::"read_file", Guardrails::Action::"write_file"],
354
+ action in [Guardrails::Action::"process_prompt",
355
+ Guardrails::Action::"process_response", Guardrails::Action::"call_tool", Guardrails::Action::"read_file", Guardrails::Action::"write_file"],
340
356
  resource
341
357
  )
342
358
  when {
@@ -355,7 +371,8 @@ when {
355
371
  @reject_message("Content blocked: employee or customer identifier patterns detected.")
356
372
  forbid (
357
373
  principal,
358
- action in [Guardrails::Action::"process_prompt", Guardrails::Action::"call_tool", Guardrails::Action::"read_file", Guardrails::Action::"write_file"],
374
+ action in [Guardrails::Action::"process_prompt",
375
+ Guardrails::Action::"process_response", Guardrails::Action::"call_tool", Guardrails::Action::"read_file", Guardrails::Action::"write_file"],
359
376
  resource
360
377
  )
361
378
  when {
@@ -376,7 +393,8 @@ when {
376
393
  @reject_message("Content blocked: device or network identifier patterns detected.")
377
394
  forbid (
378
395
  principal,
379
- action in [Guardrails::Action::"process_prompt", Guardrails::Action::"call_tool", Guardrails::Action::"read_file", Guardrails::Action::"write_file"],
396
+ action in [Guardrails::Action::"process_prompt",
397
+ Guardrails::Action::"process_response", Guardrails::Action::"call_tool", Guardrails::Action::"read_file", Guardrails::Action::"write_file"],
380
398
  resource
381
399
  )
382
400
  when {
@@ -403,7 +421,8 @@ when {
403
421
  @reject_message("Content blocked: vehicle identifier (VIN or plate) patterns detected.")
404
422
  forbid (
405
423
  principal,
406
- action in [Guardrails::Action::"process_prompt", Guardrails::Action::"call_tool", Guardrails::Action::"read_file", Guardrails::Action::"write_file"],
424
+ action in [Guardrails::Action::"process_prompt",
425
+ Guardrails::Action::"process_response", Guardrails::Action::"call_tool", Guardrails::Action::"read_file", Guardrails::Action::"write_file"],
407
426
  resource
408
427
  )
409
428
  when {
@@ -423,7 +442,8 @@ when {
423
442
  @reject_message("Content blocked: salary or financial amount patterns detected.")
424
443
  forbid (
425
444
  principal,
426
- action in [Guardrails::Action::"process_prompt", Guardrails::Action::"call_tool", Guardrails::Action::"read_file", Guardrails::Action::"write_file"],
445
+ action in [Guardrails::Action::"process_prompt",
446
+ Guardrails::Action::"process_response", Guardrails::Action::"call_tool", Guardrails::Action::"read_file", Guardrails::Action::"write_file"],
427
447
  resource
428
448
  )
429
449
  when {
@@ -444,7 +464,8 @@ when {
444
464
  @reject_message("Content blocked: sensitive attribute (ethnicity, religion, etc.) patterns detected.")
445
465
  forbid (
446
466
  principal,
447
- action in [Guardrails::Action::"process_prompt", Guardrails::Action::"call_tool", Guardrails::Action::"read_file", Guardrails::Action::"write_file"],
467
+ action in [Guardrails::Action::"process_prompt",
468
+ Guardrails::Action::"process_response", Guardrails::Action::"call_tool", Guardrails::Action::"read_file", Guardrails::Action::"write_file"],
448
469
  resource
449
470
  )
450
471
  when {
@@ -18,7 +18,7 @@
18
18
 
19
19
  @id("data-protection.advanced-block-high-risk-secrets")
20
20
  @name("Block high-risk credential types")
21
- @description("Blocks process_prompt, call_tool, read_file, and write_file when secret_types contains a cloud, GitHub, SSH, or database credential.")
21
+ @description("Blocks process_prompt, process_response, call_tool, read_file, and write_file when secret_types contains a cloud, GitHub, SSH, or database credential.")
22
22
  @severity("critical")
23
23
  @tags("category:data-protection,threat:secrets,detection:rule,owasp:llm06")
24
24
  @reject_message("Content blocked: high-risk credentials detected (cloud keys, GitHub tokens, SSH keys, or database URLs).")
@@ -26,6 +26,7 @@ forbid (
26
26
  principal,
27
27
  action in [
28
28
  Guardrails::Action::"process_prompt",
29
+ Guardrails::Action::"process_response",
29
30
  Guardrails::Action::"call_tool",
30
31
  Guardrails::Action::"read_file",
31
32
  Guardrails::Action::"write_file"
@@ -48,7 +49,7 @@ when {
48
49
 
49
50
  @id("data-protection.advanced-block-api-tokens")
50
51
  @name("Block API keys and bearer tokens")
51
- @description("Blocks process_prompt, call_tool, read_file, and write_file when secret_types contains api_key, bearer_token, jwt_token, or OAuth credentials.")
52
+ @description("Blocks process_prompt, process_response, call_tool, read_file, and write_file when secret_types contains api_key, bearer_token, jwt_token, or OAuth credentials.")
52
53
  @severity("high")
53
54
  @tags("category:data-protection,threat:secrets,detection:rule,owasp:llm06")
54
55
  @reject_message("Content blocked: API keys, bearer tokens, or OAuth credentials detected.")
@@ -56,6 +57,7 @@ forbid (
56
57
  principal,
57
58
  action in [
58
59
  Guardrails::Action::"process_prompt",
60
+ Guardrails::Action::"process_response",
59
61
  Guardrails::Action::"call_tool",
60
62
  Guardrails::Action::"read_file",
61
63
  Guardrails::Action::"write_file"
@@ -18,13 +18,14 @@
18
18
 
19
19
  @id("security.chat-block-injection")
20
20
  @name("Block injection (chat profile)")
21
- @description("Blocks process_prompt and call_tool when injection_score >= 71 (lower threshold for public chat).")
21
+ @description("Blocks process_prompt, process_response, and call_tool when injection_score >= 71 (lower threshold for public chat).")
22
22
  @severity("high")
23
23
  @tags("category:security,threat:injection,detection:ml,owasp:llm01")
24
24
  @reject_message("Prompt blocked: prompt injection detected with high confidence (chat profile).")
25
25
  forbid (
26
26
  principal,
27
- action in [Guardrails::Action::"process_prompt", Guardrails::Action::"call_tool"],
27
+ action in [Guardrails::Action::"process_prompt",
28
+ Guardrails::Action::"process_response", Guardrails::Action::"call_tool"],
28
29
  resource
29
30
  )
30
31
  when {
@@ -33,13 +34,13 @@ when {
33
34
 
34
35
  @id("security.chat-block-jailbreak")
35
36
  @name("Block jailbreak (chat profile)")
36
- @description("Blocks process_prompt when jailbreak_score >= 66 (lower threshold for public chat).")
37
+ @description("Blocks process_prompt and process_response when jailbreak_score >= 66 (lower threshold for public chat).")
37
38
  @severity("high")
38
39
  @tags("category:security,threat:jailbreak,detection:ml,owasp:llm02")
39
40
  @reject_message("Prompt blocked: jailbreak attempt detected with high confidence (chat profile).")
40
41
  forbid (
41
42
  principal,
42
- action == Guardrails::Action::"process_prompt",
43
+ action in [Guardrails::Action::"process_prompt", Guardrails::Action::"process_response"],
43
44
  resource
44
45
  )
45
46
  when {
@@ -19,13 +19,13 @@
19
19
 
20
20
  @id("trust-safety.chat-block-toxicity")
21
21
  @name("Block toxicity (chat profile)")
22
- @description("Blocks process_prompt when any toxicity score is >= 71 (lower threshold for public chat).")
22
+ @description("Blocks process_prompt and process_response when any toxicity score is >= 71 (lower threshold for public chat).")
23
23
  @severity("critical")
24
24
  @tags("category:trust-safety,threat:harmful,threat:hate-speech,detection:ml,compliance:eu-ai-act")
25
25
  @reject_message("Prompt blocked: toxic content detected at the chat-profile threshold.")
26
26
  forbid (
27
27
  principal,
28
- action == Guardrails::Action::"process_prompt",
28
+ action in [Guardrails::Action::"process_prompt", Guardrails::Action::"process_response"],
29
29
  resource
30
30
  )
31
31
  when {
@@ -37,13 +37,14 @@ when {
37
37
 
38
38
  @id("trust-safety.chat-block-restricted-topics")
39
39
  @name("Block restricted topics (chat profile)")
40
- @description("Blocks process_prompt and call_tool when topic_confidence >= 71 and content_topics contains a restricted category.")
40
+ @description("Blocks process_prompt, process_response, and call_tool when topic_confidence >= 71 and content_topics contains a restricted category.")
41
41
  @severity("high")
42
42
  @tags("category:trust-safety,threat:harmful,detection:ml,compliance:eu-ai-act")
43
43
  @reject_message("Prompt blocked: content discusses a topic restricted for chat assistants.")
44
44
  forbid (
45
45
  principal,
46
- action in [Guardrails::Action::"process_prompt", Guardrails::Action::"call_tool"],
46
+ action in [Guardrails::Action::"process_prompt",
47
+ Guardrails::Action::"process_response", Guardrails::Action::"call_tool"],
47
48
  resource
48
49
  )
49
50
  when {
@@ -17,7 +17,7 @@
17
17
 
18
18
  @id("data-protection.data-pipeline-block-secrets")
19
19
  @name("Block secrets in data pipeline")
20
- @description("Blocks process_prompt, call_tool, read_file, and write_file when secrets_detected is true.")
20
+ @description("Blocks process_prompt, process_response, call_tool, read_file, and write_file when secrets_detected is true.")
21
21
  @severity("critical")
22
22
  @tags("category:data-protection,threat:secrets,detection:rule,owasp:llm06")
23
23
  @reject_message("Request blocked: secrets detected in a data pipeline — any credential exposure is unacceptable here.")
@@ -25,6 +25,7 @@ forbid (
25
25
  principal,
26
26
  action in [
27
27
  Guardrails::Action::"process_prompt",
28
+ Guardrails::Action::"process_response",
28
29
  Guardrails::Action::"call_tool",
29
30
  Guardrails::Action::"read_file",
30
31
  Guardrails::Action::"write_file"
@@ -17,13 +17,14 @@
17
17
 
18
18
  @id("security.data-pipeline-block-injection")
19
19
  @name("Block pipeline injection (data-pipeline profile)")
20
- @description("Blocks process_prompt and call_tool when injection_score >= 66 (lower threshold for data pipelines).")
20
+ @description("Blocks process_prompt, process_response, and call_tool when injection_score >= 66 (lower threshold for data pipelines).")
21
21
  @severity("high")
22
22
  @tags("category:security,threat:injection,detection:ml,owasp:llm01")
23
23
  @reject_message("Request blocked: prompt injection detected at the data-pipeline threshold — RAG inputs are high-risk for indirect injection.")
24
24
  forbid (
25
25
  principal,
26
- action in [Guardrails::Action::"process_prompt", Guardrails::Action::"call_tool"],
26
+ action in [Guardrails::Action::"process_prompt",
27
+ Guardrails::Action::"process_response", Guardrails::Action::"call_tool"],
27
28
  resource
28
29
  )
29
30
  when {
@@ -93,13 +93,13 @@ when {
93
93
 
94
94
  @id("agent-identity.multi-agent-autonomous-injection")
95
95
  @name("Block injection on autonomous agents (stricter)")
96
- @description("Blocks process_prompt when agent_type is \"autonomous\" and injection_score >= 51.")
96
+ @description("Blocks process_prompt and process_response when agent_type is \"autonomous\" and injection_score >= 51.")
97
97
  @severity("high")
98
- @tags("category:agent-identity,threat:injection,detection:ml,surface:process-prompt,owasp:llm01")
98
+ @tags("category:agent-identity,threat:injection,detection:ml,surface:process-prompt,surface:process-response,owasp:llm01")
99
99
  @reject_message("Prompt blocked: lower injection threshold applies to autonomous agents and was exceeded.")
100
100
  forbid (
101
101
  principal is Guardrails::Agent,
102
- action == Guardrails::Action::"process_prompt",
102
+ action in [Guardrails::Action::"process_prompt", Guardrails::Action::"process_response"],
103
103
  resource
104
104
  )
105
105
  when {
@@ -109,13 +109,13 @@ when {
109
109
 
110
110
  @id("agent-identity.multi-agent-autonomous-jailbreak")
111
111
  @name("Block jailbreak on autonomous agents (stricter)")
112
- @description("Blocks process_prompt when agent_type is \"autonomous\" and jailbreak_score >= 51.")
112
+ @description("Blocks process_prompt and process_response when agent_type is \"autonomous\" and jailbreak_score >= 51.")
113
113
  @severity("high")
114
- @tags("category:agent-identity,threat:jailbreak,detection:ml,surface:process-prompt,owasp:llm02")
114
+ @tags("category:agent-identity,threat:jailbreak,detection:ml,surface:process-prompt,surface:process-response,owasp:llm02")
115
115
  @reject_message("Prompt blocked: lower jailbreak threshold applies to autonomous agents and was exceeded.")
116
116
  forbid (
117
117
  principal is Guardrails::Agent,
118
- action == Guardrails::Action::"process_prompt",
118
+ action in [Guardrails::Action::"process_prompt", Guardrails::Action::"process_response"],
119
119
  resource
120
120
  )
121
121
  when {
@@ -482,15 +482,6 @@
482
482
  "compliance:hipaa"
483
483
  ]
484
484
  },
485
- {
486
- "id": "security.advanced-block-critical-severity",
487
- "name": "Advanced Detection — Threat Severity",
488
- "description": "Catch-all that blocks any content flagged as critical severity by any detector.",
489
- "category": "security",
490
- "file": "profiles/advanced_detection/threat_severity.cedar",
491
- "severity": "critical",
492
- "tags": ["category:security", "detection:aggregate", "posture:catch-all"]
493
- },
494
485
  {
495
486
  "id": "agent-security.param-validation",
496
487
  "name": "Action Parameter Validation",
@@ -680,6 +680,36 @@
680
680
  "required": false,
681
681
  "description": "All operation classes observed in the tool call, emitted by the AST command classifiers. Subset of: 'readonly', 'write_enabling', 'execute_enabling', 'network_access', 'unknown'. A single command can carry several classes; combine with contains() to match shapes — e.g. tool_operation_classes.contains(\"network_access\") && tool_operation_classes.contains(\"execute_enabling\") catches download-and-run (curl | sh). Used by the Bash Operation Classes template. 'unknown' appears when any unrecognized or variable-expanded command was seen and should be treated as deny-worthy."
682
682
  },
683
+ {
684
+ "key": "resolved_target_paths",
685
+ "type": "string",
686
+ "required": false,
687
+ "description": "Newline-delimited (leading+trailing \\n) union of the statically-resolved absolute target paths a shell command touches, any capability, emitted by the bash AST detector. Cedar like is full-string anchored and every value starts with the \\n delimiter, so patterns must begin with '*': anchored sub-tree \"*\\n/etc/*\" (matches only paths under /etc, unlike the loose \"*/etc/*\" which also hits a project etc/ dir at any depth), bare directory \"*\\n/etc\\n*\" (e.g. rm -rf /etc, no trailing slash), exact file \"*\\n/etc/passwd\\n*\". Resolution is purely lexical (cwd-join + path.Clean; quoting, escapes, wrappers, cd/VAR/bash -c normalized); remote endpoints and dynamic operands are never fabricated."
688
+ },
689
+ {
690
+ "key": "read_target_paths",
691
+ "type": "string",
692
+ "required": false,
693
+ "description": "Newline-delimited resolved absolute paths a shell command reads or enumerates, emitted by the bash AST detector. \\n-anchored: match with like patterns that begin with '*' (see resolved_target_paths)."
694
+ },
695
+ {
696
+ "key": "write_target_paths",
697
+ "type": "string",
698
+ "required": false,
699
+ "description": "Newline-delimited resolved absolute paths a shell command writes, creates, or destroys (redirects, cp/mv/dd, curl/wget -o, scp/rsync local dest, tar/unzip/cpio extraction, yq -i), emitted by the bash AST detector. \\n-anchored: e.g. write_target_paths like \"*\\n/etc/cron.d/*\"."
700
+ },
701
+ {
702
+ "key": "exec_target_paths",
703
+ "type": "string",
704
+ "required": false,
705
+ "description": "Newline-delimited resolved absolute paths a shell command executes as a script, emitted by the bash AST detector. \\n-anchored: match with like patterns that begin with '*' (see resolved_target_paths)."
706
+ },
707
+ {
708
+ "key": "unresolved_target",
709
+ "type": "boolean",
710
+ "required": false,
711
+ "description": "True when a shell command's target hid behind a variable, glob, command substitution, or inline shell code that static analysis could not resolve, or when the 64-path / 4096-byte path-list cap truncated the results. Benign dev traffic sets this frequently (variables, globs, and command substitution are everywhere, and the truncation cap trips it too), so it is NOT a deny signal on its own — combine it with a path or operation match, and reserve a blanket deny-on-unresolved rule for high-assurance or allowlisted environments."
712
+ },
683
713
  {
684
714
  "key": "package_install_detected",
685
715
  "type": "boolean",
@@ -246,6 +246,13 @@ action call_tool appliesTo {
246
246
  // --- Tool Operation Classifier ---
247
247
  tool_operation_classes?: Set<String>, // subset of {"readonly", "write_enabling", "execute_enabling", "network_access", "unknown"}
248
248
 
249
+ // --- Bash AST resolved target paths (#411 / CAP-DET-007) — newline-delimited (leading+trailing \n); Cedar `like` is anchored so patterns start with "*": sub-tree "*\n/etc/*", bare dir "*\n/etc\n*", exact "*\n/etc/passwd\n*" ---
250
+ resolved_target_paths?: String, // union of resolved absolute target paths, any capability
251
+ read_target_paths?: String, // paths the command reads or enumerates
252
+ write_target_paths?: String, // paths the command writes, creates, or destroys
253
+ exec_target_paths?: String, // paths executed as a script
254
+ unresolved_target?: Bool, // a target hid behind a variable/glob/$()/inline code
255
+
249
256
  // --- Session Detection History (cross-turn sticky flags) ---
250
257
  session_pii_detected?: Bool,
251
258
  session_pii_types?: Set<String>,
@@ -41,6 +41,7 @@ export declare const AgentOpsContextKey: {
41
41
  readonly EncodedTypes: "encoded_types";
42
42
  readonly Event: "event";
43
43
  readonly ExecAllowlist: "exec_allowlist";
44
+ readonly ExecTargetPaths: "exec_target_paths";
44
45
  readonly FactualityScore: "factuality_score";
45
46
  readonly FlowCompartments: "flow_compartments";
46
47
  readonly FlowConfidentiality: "flow_confidentiality";
@@ -108,7 +109,9 @@ export declare const AgentOpsContextKey: {
108
109
  readonly PrincipalCompartments: "principal_compartments";
109
110
  readonly PrivilegeScope: "privilege_scope";
110
111
  readonly ProfanityScore: "profanity_score";
112
+ readonly ReadTargetPaths: "read_target_paths";
111
113
  readonly RequestId: "request_id";
114
+ readonly ResolvedTargetPaths: "resolved_target_paths";
112
115
  readonly Role: "role";
113
116
  readonly RpmExceeded: "rpm_exceeded";
114
117
  readonly RpmRemainingPct: "rpm_remaining_pct";
@@ -158,15 +161,17 @@ export declare const AgentOpsContextKey: {
158
161
  readonly TopicConfidence: "topic_confidence";
159
162
  readonly TpmExceeded: "tpm_exceeded";
160
163
  readonly TpmRemainingPct: "tpm_remaining_pct";
164
+ readonly UnresolvedTarget: "unresolved_target";
161
165
  readonly UserEmail: "user_email";
162
166
  readonly ViolenceScore: "violence_score";
163
167
  readonly WeaponsScore: "weapons_score";
164
168
  readonly WorkspaceRoot: "workspace_root";
165
169
  readonly WritablePaths: "writable_paths";
170
+ readonly WriteTargetPaths: "write_target_paths";
166
171
  };
167
172
  export type AgentOpsContextKey = (typeof AgentOpsContextKey)[keyof typeof AgentOpsContextKey];
168
173
  /**
169
174
  * The full set of authorable context attribute keys for AgentOps.
170
175
  * Iterate this to enumerate the authorable surface (cockpit, conformance).
171
176
  */
172
- export declare const AgentOpsContextKeys: readonly ["agent_framework", "agent_id", "agent_publisher", "agent_trust_level", "agent_type", "budget_exceeded", "budget_remaining_pct", "code_languages", "code_ratio", "command_injection_detected", "command_injection_score", "command_injection_type", "contains_code", "contains_non_ascii", "content_safety_blocked", "content_safety_score", "content_topics", "content_type", "conversation_turn", "crime_score", "cross_origin_detected", "cross_origin_score", "cross_origin_type", "cwd", "detected_language", "detected_script", "detected_threats", "detector_count", "direction", "egress_hosts", "encoded_content_detected", "encoded_count", "encoded_score", "encoded_types", "event", "exec_allowlist", "factuality_score", "flow_compartments", "flow_confidentiality", "flow_data_types", "flow_integrity", "flow_origins", "flow_resolution_status", "flow_sink", "flow_sink_effects", "flow_sink_is_external", "hallucination_score", "hate_speech_score", "highest_severity", "identity_type", "indirect_injection_score", "indirect_injection_type", "injection_deep_context_score", "injection_pulse_score", "injection_score", "injection_type", "invisible_chars_detected", "invisible_chars_score", "is_english", "is_latin_script", "isolation_tier", "jailbreak_deep_context_score", "jailbreak_pulse_score", "jailbreak_score", "keyword_categories", "keyword_count", "keyword_matched", "language_confidence", "loop_count", "loop_detected", "loop_tool", "malicious_package_detected", "max_threat_severity", "mcp_config_risk", "mcp_risk_score", "mcp_risk_type", "mcp_server", "mcp_server_verified", "mcp_tool", "mechanism_capabilities", "model_name", "model_provider", "multi_turn_detection", "network_egress", "package_check_status", "package_install_detected", "param_type_violation", "param_type_violations", "path", "path_traversal_detected", "path_traversal_severity", "path_traversal_type", "pattern_type", "phishing_detected", "pii_count", "pii_detected", "pii_score", "pii_types", "principal", "principal_clearances", "principal_compartments", "privilege_scope", "profanity_score", "request_id", "role", "rpm_exceeded", "rpm_remaining_pct", "rug_pull_detected", "rug_pull_score", "rug_pull_type", "script_confidence", "secret_count", "secret_types", "secrets_detected", "sentiment_score", "sequence_risk", "session_command_injection", "session_cumulative_risk_score", "session_injection_detected", "session_max_command_injection_score", "session_max_injection_score", "session_max_jailbreak_score", "session_max_pii_score", "session_max_secret_score", "session_max_sensitivity", "session_original_request", "session_pii_detected", "session_pii_types", "session_secret_types", "session_secrets_detected", "session_threat_turns", "sexual_score", "source", "sql_injection_detected", "sql_injection_score", "sql_injection_type", "surface", "suspicious_pattern", "threat_categories", "threat_count", "timestamp", "tool_category", "tool_is_builtin", "tool_is_sensitive", "tool_name", "tool_operation_classes", "tool_poisoning_detected", "tool_poisoning_score", "tool_poisoning_type", "tool_risk_score", "topic_confidence", "tpm_exceeded", "tpm_remaining_pct", "user_email", "violence_score", "weapons_score", "workspace_root", "writable_paths"];
177
+ export declare const AgentOpsContextKeys: readonly ["agent_framework", "agent_id", "agent_publisher", "agent_trust_level", "agent_type", "budget_exceeded", "budget_remaining_pct", "code_languages", "code_ratio", "command_injection_detected", "command_injection_score", "command_injection_type", "contains_code", "contains_non_ascii", "content_safety_blocked", "content_safety_score", "content_topics", "content_type", "conversation_turn", "crime_score", "cross_origin_detected", "cross_origin_score", "cross_origin_type", "cwd", "detected_language", "detected_script", "detected_threats", "detector_count", "direction", "egress_hosts", "encoded_content_detected", "encoded_count", "encoded_score", "encoded_types", "event", "exec_allowlist", "exec_target_paths", "factuality_score", "flow_compartments", "flow_confidentiality", "flow_data_types", "flow_integrity", "flow_origins", "flow_resolution_status", "flow_sink", "flow_sink_effects", "flow_sink_is_external", "hallucination_score", "hate_speech_score", "highest_severity", "identity_type", "indirect_injection_score", "indirect_injection_type", "injection_deep_context_score", "injection_pulse_score", "injection_score", "injection_type", "invisible_chars_detected", "invisible_chars_score", "is_english", "is_latin_script", "isolation_tier", "jailbreak_deep_context_score", "jailbreak_pulse_score", "jailbreak_score", "keyword_categories", "keyword_count", "keyword_matched", "language_confidence", "loop_count", "loop_detected", "loop_tool", "malicious_package_detected", "max_threat_severity", "mcp_config_risk", "mcp_risk_score", "mcp_risk_type", "mcp_server", "mcp_server_verified", "mcp_tool", "mechanism_capabilities", "model_name", "model_provider", "multi_turn_detection", "network_egress", "package_check_status", "package_install_detected", "param_type_violation", "param_type_violations", "path", "path_traversal_detected", "path_traversal_severity", "path_traversal_type", "pattern_type", "phishing_detected", "pii_count", "pii_detected", "pii_score", "pii_types", "principal", "principal_clearances", "principal_compartments", "privilege_scope", "profanity_score", "read_target_paths", "request_id", "resolved_target_paths", "role", "rpm_exceeded", "rpm_remaining_pct", "rug_pull_detected", "rug_pull_score", "rug_pull_type", "script_confidence", "secret_count", "secret_types", "secrets_detected", "sentiment_score", "sequence_risk", "session_command_injection", "session_cumulative_risk_score", "session_injection_detected", "session_max_command_injection_score", "session_max_injection_score", "session_max_jailbreak_score", "session_max_pii_score", "session_max_secret_score", "session_max_sensitivity", "session_original_request", "session_pii_detected", "session_pii_types", "session_secret_types", "session_secrets_detected", "session_threat_turns", "sexual_score", "source", "sql_injection_detected", "sql_injection_score", "sql_injection_type", "surface", "suspicious_pattern", "threat_categories", "threat_count", "timestamp", "tool_category", "tool_is_builtin", "tool_is_sensitive", "tool_name", "tool_operation_classes", "tool_poisoning_detected", "tool_poisoning_score", "tool_poisoning_type", "tool_risk_score", "topic_confidence", "tpm_exceeded", "tpm_remaining_pct", "unresolved_target", "user_email", "violence_score", "weapons_score", "workspace_root", "writable_paths", "write_target_paths"];
@@ -43,6 +43,7 @@ export const AgentOpsContextKey = {
43
43
  EncodedTypes: 'encoded_types',
44
44
  Event: 'event',
45
45
  ExecAllowlist: 'exec_allowlist',
46
+ ExecTargetPaths: 'exec_target_paths',
46
47
  FactualityScore: 'factuality_score',
47
48
  FlowCompartments: 'flow_compartments',
48
49
  FlowConfidentiality: 'flow_confidentiality',
@@ -110,7 +111,9 @@ export const AgentOpsContextKey = {
110
111
  PrincipalCompartments: 'principal_compartments',
111
112
  PrivilegeScope: 'privilege_scope',
112
113
  ProfanityScore: 'profanity_score',
114
+ ReadTargetPaths: 'read_target_paths',
113
115
  RequestId: 'request_id',
116
+ ResolvedTargetPaths: 'resolved_target_paths',
114
117
  Role: 'role',
115
118
  RpmExceeded: 'rpm_exceeded',
116
119
  RpmRemainingPct: 'rpm_remaining_pct',
@@ -160,11 +163,13 @@ export const AgentOpsContextKey = {
160
163
  TopicConfidence: 'topic_confidence',
161
164
  TpmExceeded: 'tpm_exceeded',
162
165
  TpmRemainingPct: 'tpm_remaining_pct',
166
+ UnresolvedTarget: 'unresolved_target',
163
167
  UserEmail: 'user_email',
164
168
  ViolenceScore: 'violence_score',
165
169
  WeaponsScore: 'weapons_score',
166
170
  WorkspaceRoot: 'workspace_root',
167
171
  WritablePaths: 'writable_paths',
172
+ WriteTargetPaths: 'write_target_paths',
168
173
  };
169
174
  /**
170
175
  * The full set of authorable context attribute keys for AgentOps.
@@ -207,6 +212,7 @@ export const AgentOpsContextKeys = [
207
212
  AgentOpsContextKey.EncodedTypes,
208
213
  AgentOpsContextKey.Event,
209
214
  AgentOpsContextKey.ExecAllowlist,
215
+ AgentOpsContextKey.ExecTargetPaths,
210
216
  AgentOpsContextKey.FactualityScore,
211
217
  AgentOpsContextKey.FlowCompartments,
212
218
  AgentOpsContextKey.FlowConfidentiality,
@@ -274,7 +280,9 @@ export const AgentOpsContextKeys = [
274
280
  AgentOpsContextKey.PrincipalCompartments,
275
281
  AgentOpsContextKey.PrivilegeScope,
276
282
  AgentOpsContextKey.ProfanityScore,
283
+ AgentOpsContextKey.ReadTargetPaths,
277
284
  AgentOpsContextKey.RequestId,
285
+ AgentOpsContextKey.ResolvedTargetPaths,
278
286
  AgentOpsContextKey.Role,
279
287
  AgentOpsContextKey.RpmExceeded,
280
288
  AgentOpsContextKey.RpmRemainingPct,
@@ -324,9 +332,11 @@ export const AgentOpsContextKeys = [
324
332
  AgentOpsContextKey.TopicConfidence,
325
333
  AgentOpsContextKey.TpmExceeded,
326
334
  AgentOpsContextKey.TpmRemainingPct,
335
+ AgentOpsContextKey.UnresolvedTarget,
327
336
  AgentOpsContextKey.UserEmail,
328
337
  AgentOpsContextKey.ViolenceScore,
329
338
  AgentOpsContextKey.WeaponsScore,
330
339
  AgentOpsContextKey.WorkspaceRoot,
331
340
  AgentOpsContextKey.WritablePaths,
341
+ AgentOpsContextKey.WriteTargetPaths,
332
342
  ];