+ JiFRt^RIt^RIHtHtHtRRRlltRRRlltR#) aaAnthropic prompt caching (system_and_3 strategy). Reduces input token costs by ~75% on multi-turn conversations by caching the conversation prefix. Uses 4 cache_control breakpoints (Anthropic max): 1. System prompt (stable across all turns) 2-4. Last 3 non-system messages (rolling window) Pure functions -- no class state, no AIAgent dependency. N)AnyDictListc@V^8dQhR\R\R\RR/#)msg cache_markernative_anthropicreturnN)dictbool)formats"1/home/ubuntu/hermes-agent/agent/prompt_caching.py __annotate__rs)11T111Z^1cnVPRR4pVPR4pVR8XdV'dWR&R#VeVR8XdWR&R#\V\4'dRRRVRV/.VR&R#\V\4'd2V'd(VR ,p\V\4'dWR&R#R#R#R#) zFAdd cache_control to a single message, handling all format variations.rolecontenttool cache_controlNtypetext)get isinstancestrlistr )rrr rrlasts&&& r_apply_cache_markerrs 7762 Dggi G v~ #/ 'R-+O'3 VVWo| L I '4  Wr{ dD ! !$0 ! "&- rc V^8dQhR\\\\3,,R\R\R\\\\3,,/#)r api_messages cache_ttlr r )rrrrr )r s"rrr)sJtCH~& $sCx. rc\P!V4pV'gV#RR/pVR8XdRVR&^pV^,PR4R8Xd\V^,WBR7V^, p^V, p\ \ V44Uu.uF#qsV,PR4R8wgK!VNK% ppW)RFp \W9,WBR7K V#uupi) zApply system_and_3 caching strategy to messages for Anthropic models. Places up to 4 cache_control breakpoints: system prompt + last 3 non-system messages. Returns: Deep copy of messages with cache_control breakpoints injected. r ephemeral1httlrsystem)r N)copydeepcopyrrrangelen) r!r"r messagesmarkerbreakpoints_used remaininginon_sysidxs &&& rapply_anthropic_cache_controlr3)s}}\*H k "FDu {v(*HQKSA$$IH .V.Q1+//&2IX2Uqq.GVz{#HM6U$ O Ws C/C)F)5mF)__doc__r(typingrrrrr3rrr8s$ ""14r