| jwSrSSKJr SSKJr SSKJr SSKJr SSK J r SSK J r J r SSKJr SS KJr "S S 5r"S S 5r"SS5rg)a Two-tier Memory Architecture for Agent Context Management. Working context: the conversation message list, owned by the agent loop and compressed by acli.compression (not tracked in this module). Mid-term (Session Memory): Per-session persistent state (plan, reflection, tool_chains) Long-term (Persistent Memory): Cross-session persistent memory (experience, user profile) Design inspired by context engineering principles (Weng 2026, TMLR Survey 2026). ) annotations)Path)Any)ExperienceTracker) PlanTracker)ReflectionTrackerStagnationTracker)ToolChainLibrary) TraceLoggerc,\rSrSrSrSSjrSSjrSrg) SessionMemoryzMid-term memory: per-session persistent state. Tracks plan, reflection, and tool chains for the current session. Persisted to .acli/session/ directory. cUS- UlURRSSS9 [5Ul[ 5Ul[ 5Ul[UR5Ul g)NsessionT)parentsexist_ok) session_dirmkdirrplanr reflectionr stagnationr tool_chainsself workspace_dirs ګC:\Users\johnn\AppData\Local\Packages\PythonSoftwareFoundation.Python.3.13_qbz5n2kfra8p0\LocalCache\local-packages\Python313\site-packages\dashscope/acli/memory/manager.py__init__SessionMemory.__init__&sX(94 td;M +-+-+D,<,<=cURR5 URR5 URR5 gz/Reset session memory (called at session start).N)r clear_planrresetrrs rr#SessionMemory.reset/s2   r)rrrrrNrrreturnNone)__name__ __module__ __qualname____firstlineno____doc__rr#__static_attributes__rrr r s > rr ch\rSrSrSrS SjrS S SjjrS S SjjrSSjrSr g)PersistentMemory6zLong-term memory: cross-session persistent memory. Stores experience and user profile that persist across sessions. Persisted to .acli/ directory. c0Xl[U5UlgN)rr experiencers rrPersistentMemory.__init__=s*+M:rc@URRUUUU5 g)zRecord a task experience.N)r6record_experiencer task_summary tools_usedoutcomelessons rr9"PersistentMemory.record_experienceA" ))      rc8URRX5$)z Search for relevant experiences.)r6search_experiences)rquerylimits rrB#PersistentMemory.search_experiencesPs 11%??rc8URRU5$)z(Format experiences for prompt injection.)r6format_experiences_for_prompt)r experiencess rformat_for_prompt"PersistentMemory.format_for_promptXs<<[IIr)r6rNr& r;strr<z list[str]r=rNr>rNr(r)))rCrNrDintr(list[dict[str, Any]])rHrQr(rN) r*r+r,r-r.rr9rBrIr/r0rrr2r26sv ;            $@@@  @Jrr2cv\rSrSrSrS Sjr\S Sj5rS SjrSSSjjr SSjr SSjr S r g ) MemoryManager]uUTwo-tier memory manager: per-session state + cross-session persistent memory. Coordinates mid-term (session: plan, reflection, tool chains) and long-term (persistent: experience, user profile) memory, plus trace logging. The conversation's working context lives in Agent.messages and is compressed by acli.compression — not tracked here. cd[U5Ul[U5Ul[ U5Ulgr5)r rr2 persistentr tracers rrMemoryManager.__init__gs&$]3 *=9 / rcSSKnURU5nURRS- UR 5R SS- n[ U5UlURUlURUlU$)a Build an isolated manager for a sub-agent. The child gets a fresh SessionMemory (own plan/reflection/tool_chains, under a ``child-*`` subdirectory) so its tool failures and plans never touch the parent's counters, while sharing the parent's persistent experience store and trace logger. rNchildren) uuid__new__rruuid4hexr rVrW)clsparentr\childchild_workspaces r derive_childMemoryManager.derive_childlsp  C  NN & & 3djjl6F6Fr6J J &o6 !,,ll  rcNURRRX5 g)z2Record a tool execution for reflection monitoring.N)rrrecord_tool_execution)r tool_namesuccesss rrg#MemoryManager.record_tool_executions 55iIrc@URRUUUU5 g)z.Record a task experience to persistent memory.N)rVr9r:s rr9MemoryManager.record_experiencer@rc:URRX5 g)zLog a trace event.N)rWlog)r event_typedatas r log_traceMemoryManager.log_traces z(rc8URR5 gr!)rr#r$s r reset_sessionMemoryManager.reset_sessions r)rVrrWNr&)ra'MemoryManager'r(rv)rhrNriboolr(r)rKrM)rorNrpzdict[str, Any]r(r)r') r*r+r,r-r.r classmethodrdrgr9rqrtr/r0rrrSrS]sj0 &J            )rrSN)r. __future__rpathlibrtypingr dashscope.acli.memory.experiencerdashscope.acli.memory.planr dashscope.acli.memory.reflectionrr !dashscope.acli.memory.tool_chainsr dashscope.acli.memory.tracer r r2rSr0rrrsI #>2?3  .$J$JN<<r