tPj_lUdZddlZddlZddlmZmZmZmZmZddl m Z ddl m Z ddl mZmZmZddlmZddlmZdd lmZmZdd lmZdd lmZerdd lmZd Zee d<Gdde!Z"Gdde!Z#GddeZ$y)z'Gray Swan Cygnal guardrail integration.N) TYPE_CHECKINGAnyFinalLiteralOptional) HTTPException)verbose_proxy_logger)CustomGuardrailModifyResponseExceptionlog_guardrail_information) safe_dumps)safe_json_loads)get_async_httpx_clienthttpxSpecialProvider)GuardrailEventHooks)GenericGuardrailAPIInputs)LoggingzBlocked by Gray Swan GuardrailGRAYSWAN_BLOCK_ERROR_MSGceZdZdZy)GraySwanGuardrailMissingSecretsz-Raised when the Gray Swan API key is missing.N)__name__ __module__ __qualname____doc__/home/oem/CyberBrain26-unified/.venv/lib/python3.12/site-packages/litellm/proxy/guardrails/guardrail_hooks/grayswan/grayswan.pyrrs7rrc8eZdZdZddededzddffd ZxZS)GraySwanGuardrailAPIErrorz/Raised when the Gray Swan API returns an error.Nmessage status_codereturnc2t||||_yN)super__init__r!)selfr r! __class__s rr&z"GraySwanGuardrailAPIError.__init__%s !&rr$)rrrrstrintr& __classcell__r(s@rrr"s)9''#*'''rrceZdZdZhdZdZdZdZhdZ d;de dzd e dzd e dzd e dzd e dzd e dzde e e fdzde dzde de de dzde dzdeddffd ZedeefdZde ffd Ze dd0ed)e de fd1Z#d2ede fd3Z$d4e dzde fd5Z%d4e dzde dzfd6Z& dblockmonitor passthroughr0zhttps://api.grayswan.aiz/cygnal/monitor>offhybridthinkingNguardrail_nameapi_keyapi_baseon_flagged_actionviolation_thresholdreasoning_mode categories policy_idstreaming_end_of_stream_onlystreaming_sampling_rate fail_openguardrail_timeoutkwargsr"c  ttj|_|xst j d}|s t d||_|xs#t j dxs |j}|jd|_ |j|j|_ |}|r2|j|jvr|j|_n4|r!t!j"d||j$|j$|_|j'||_|j+||_||_||_| dn t3| |_| dn t7| |_| |_| |_t!j>d | | tA|d |tE|jGd | y) N) llm_providerGRAYSWAN_API_KEYzDGray Swan API key missing. Set `GRAYSWAN_API_KEY` or pass `api_key`.GRAYSWAN_API_BASE/zLGray Swan Guardrail: Unsupported on_flagged_action '%s', defaulting to '%s'.T>@zNGraySwan __init__: streaming_end_of_stream_only=%s, streaming_sampling_rate=%s)r5supported_event_hooksr)$rrGuardrailCallback async_handlerosgetenvrr6 BASE_API_URLrstripr7 MONITOR_PATH monitor_urllowerSUPPORTED_ON_FLAGGED_ACTIONSr8r warningDEFAULT_ON_FLAGGED_ACTION_resolve_thresholdr9_resolve_reasoning_moder:r;r<boolr?floatr@r=r>debugr%r&listget_supported_event_hooks)r'r5r6r7r8r9r:r;r<r=r>r?r@rA api_key_valuebaseactionr(s rr&zGraySwanGuardrail.__init__@s 4AUAgAgh&G"))4F*G 1V * U")),?"@UDDUDU C( "mm_T->->,?@) flln(I(II%+\\^D "$,,b22 &*%C%CD "#'#:#:;N#O "::>J$"!*!2Y):)BN_H`-I)'>$"" \ ( #  )"&t'E'E'G"H  rc`tjtjtjgSr$)rpre_call during_call post_call)clss rr[z+GraySwanGuardrail.get_supported_event_hookss) ( (  + +  ) )  rc t|||}dt|jv}t j d|||j |t|j|S)zOverride to add debug logging.apply_guardrailznGraySwan DEBUG: should_run_guardrail event_type=%s, result=%s, event_hook=%s, has_apply_guardrail=%s, class=%s)r%should_run_guardrailtype__dict__r rY event_hookr)r'data event_typeresulthas_apply_guardrailr(s rrfz&GraySwanGuardrail.should_run_guardrails^4T:F%6$t*:M:M%M"" |   OO  J     rinputs request_data input_type)requestresponse logging_objLiteLLMLoggingObjc Ktjd||jdr|jdgddnd|jdg}|stjd|Stjdt|||dk(rd nd }|Dcgc]}||d  }}|j |xsi} | rtjd t | |j || ||} | |Stj} |j| d{} |dk(} |j| ||| }|Scc}w7%#t$r}|j|rtj}t|ddxs t|dd}|j||xsi| |||jrtjd||cYd}~St!|t"r|t#t%|||d}~wwxYww)a Apply Gray Swan guardrail to extracted text content. This method is called by the unified guardrail system which handles extracting text from any request format (OpenAI, Anthropic, etc.). Args: inputs: Dictionary containing: - texts: List of texts to scan - images: Optional list of images (not currently used by GraySwan) - tool_calls: Optional list of tool calls (not currently used) request_data: The original request data input_type: "request" for pre-call, "response" for post-call logging_obj: Optional logging object Returns: GenericGuardrailAPIInputs - texts may be replaced with violation message in passthrough mode Raises: HTTPException: If content is blocked (block mode) Exception: If guardrail check fails zCGraySwan DEBUG: apply_guardrail called with input_type=%s, texts=%stextsNdNONEz%Gray Swan Guardrail: No texts to scanz/Gray Swan Guardrail: Scanning %d text(s) for %srr assistantuser)rolecontentz*Gray Swan Guardrail: dynamic extra_body=%s) response_jsonrorn is_outputr!exception_status_code)excro start_timeend_timer!zRGray Swan Guardrail: fail_open=True. Allowing request to proceed despite error: %sr!)r rYgetlen)get_guardrail_dynamic_request_body_paramsr _prepare_payloadtime_call_grayswan_api_process_response_internal Exception_is_grayswan_exceptiongetattr_log_guardrail_failurer?rS isinstancerr))r'rnrorprsrvr{textmessages dynamic_bodypayloadrr}r~rlrrr!s rrez!GraySwanGuardrail.apply_guardrails> "" Q -3ZZ-@FJJw #DS )f zz'2. & &'N OM"" = J  &0:%=k6GLMtDT:MM#LL\Z`^`  & &'SU_`lUm n..x|U`a ?M IIK  X)-)@)@)I#IM)Z7I ;;+)# <F M-N$J X**3/"iikH!(mT!B!qgcSjlpFqK  ' ')/R%!' ( ~~$,,h #89 +CH+NTW W) Xs\BH D>$A$H EEE=HE HA?G< HH)G<<HHrct|tryt|dd}t|tr|j dt k(Sy)NTdetailerrorF)rr rdictrr)r'rrs rrz(GraySwanGuardrail._is_grayswan_exceptionsB c2 3Xt4 fd #::g&*BB BrrcfK|j|d{}|j||S7w)z Run the GraySwan guardrail on a payload. This is a legacy method for testing purposes. Args: payload: The payload to scan Returns: Dict containing the GraySwan API response N)r_process_grayswan_response)r'rr}s rrun_grayswan_guardrailz(GraySwanGuardrail.run_grayswan_guardrail s5&*%<%>yIG;7;@TVZ+[//52::;KRPTTU_acdhhirs '4((:#%!--/ 9JC3x~~#99.1%j!#c(+ 9!(D(rrrcDd|i}|jdxs |j}|r||d<|jdxs |j}|r||d<|jdxs |j}|r||d<d|vr|d|d<|j ||} |jd} t | t r t | ni} | r,| jd} t | t ri| | n| | d<| r-tt| i} t | t r| r| |d<|S) Nrr;r<r:rlitellm_metadatar)default) rr;r<r:rrrrr )r'rrrorsrr;r<r:inbound_headersrcleaned_litellm_metadataexisting_headers sanitizeds rrz"GraySwanGuardrail._prepare_payloadsJ+5h)?(,,\:Mdoo $.GL !'++K8JDNN #,GK , 0 01A B YdFYFY (6G$ %  %".z":GJ !%!>!>|[!Y"."2"23E"FDNO_aeDf$/?*@ln &>&B&B9&M ;EFVX\;]7#77cr %Y / $.z:R/S]_`I)T*y.7*+rrct|trt|dkDr|d}t|tr|ni}|j dd}|j dg}|j dd}|j dd}|rdnd }d |d |d d g} |r(|j |} | r| j d| d |r| j d|r| j ddj| S)a Format detection info into a user-friendly violation message. Args: detection_info: Can be either: - A single dict with violation_score, violated_rules, mutation, ipi keys - A list of such dicts (legacy format) is_output: True if violation is in model output, False if in input Returns: Formatted violation message string rrrrrFrzthe model responsez input queryzOSorry I can't help with that. According to the Gray Swan Cygnal Guardrail, the z has a violation score of z.2f.zIt was violating the rule(s): zEMutation effort to make the harmful intention disguised was DETECTED.z'Indirect Prompt Injection was DETECTED. )rrZrrr_format_violated_rulesrjoin) r'rr~detection_dictrrrrr message_partsformatted_ruless rrz+GraySwanGuardrail._format_violation_message<s nd +N0Ca0G+A.N9C>SW8Xn^`!/!3!34Es!K . 2 23CR H(,,Z?#''u5>c|r,|j|jvr|jStjd}|r,|j|jvr|jSy)NGRAYSWAN_REASONING_MODE)rQSUPPORTED_REASONING_MODESrKrLrs rrVz)GraySwanGuardrail._resolve_reasoning_modesX U[[]d&D&DD;;= #<= w}}$*H*HH==? "rrrr!c  |jt||d||||z dtj d|r d|d|y d|y #t$rtjd|YGwxYw) z;Log guardrail failure and attach standard logging metadata.guardrail_failed_to_respondr)guardrail_json_responseroguardrail_statusrrdurationguardrail_providerzBGray Swan Guardrail: failed to log guardrail failure for error: %sz-Gray Swan Guardrail: API request failed%s: %sz (status_code=)rN):add_standard_logging_guardrail_information_to_request_datar)rr exceptionr)r'rrorrr!s rrz(GraySwanGuardrail._log_guardrail_failures   K K(+C)!>%!!J.#- L  "" ;/:n[M +  @B    * *T  s$A A.-A.) rNNNNNNNFTrGr$)NN)F))rrrrrRrTrMrOrr)rXrrWr*rr& classmethodrZrr[rfr rrrrerrrrrrrrrrrrUrVrr+r,s@rr.r.*s $G ),L$L =&0"#(,,0%),0 $-2'(!%*.> d > t> * > : > #T\ > d > cNT)> :> '+> "%> $;> !4<> >  > @ $/B*C    & 6: ]X)]X]X12 ]X 12 ]X # ]X]X~)DT#s(^(!04 `P`PTk`P'- `P  `PLXXc3hX&WCH~WW* W  W # Wz $sCx. 6:))12) c3h$  ):6: 'tCH~&'' ' 12 ' c3h$  'R*(*(*(Y\*(X$T$c$0   S4ZC$J#'        4Z    rr.)%rrKrtypingrrrrrfastapirlitellm._loggingr %litellm.integrations.custom_guardrailr r r *litellm.litellm_core_utils.safe_json_dumpsr *litellm.litellm_core_utils.safe_json_loadsr&litellm.llms.custom_httpx.http_handlerrrlitellm.types.guardrailsrlitellm.types.utilsr*litellm.litellm_core_utils.litellm_loggingrrtr__annotations__rrrr.rrrr sq- ??!1 BF99W"B%B8i8' 'E  E  r