Aska
648292d035
- 每轮 assemble(transformContext 等价)：run.go 每 step 前 assembleContext——依次 Inject(每轮 重新 Retriever.Inject 不缓存首轮)、WorkingMemory 前置 Summarize、可插拔 TransformContext 投影/ 过滤；生成的 view 交 step，msgs 保持纯对话供工具结果追加，记忆不重复注入、不写回 checkpoint。 - Compactor：结构分支感知压缩——estimateContextTokens 启发式超 MaxContextTokens 且消息≥3 才动； findKeepStart 按 user 消息切整轮从后往前累加(至少保留最后一整轮→尾随工具结果绝不切片/不跨轮)； 旧轮 user/assistant 对压成一条 RoleSystem 摘要置顶，既有摘要前缀识别合并防叠；capSummary 上限 4000 rune；ctx 即 abortSignal(取消返回 ctx.Err，压缩失败仅记日志不中断 run)。 - 选项：Options 增 MaxContextTokens/Compactor/TransformContext，均 nil-tolerant(默认=现有行为， 零值不压缩)；DefaultMaxContextTokens=32768。Compactor 接口预留注入 LLM 摘要。 go build/vet/test -count=1 全绿 + windows/darwin 交叉编译。12 断言覆盖尾随工具结果保留/不跨轮切割/ 摘要上限/摘要合并/ctx 取消/每轮重检索/TransformContext 逐轮/超限压缩/默认关闭。-race 受宿主 TSan VMA 限制无法运行(与代码无关)。 **至此 Top5(1~5) 全部完成并推库。**
90 lines
4.6 KiB
Go
90 lines
4.6 KiB
Go
package loop
|
|
|
|
import (
|
|
"context"
|
|
"log/slog"
|
|
|
|
"git.yeij.top/AskaEth/Cyrene_v2/internal/runtime/agent"
|
|
"git.yeij.top/AskaEth/Cyrene_v2/internal/runtime/guardrail"
|
|
"git.yeij.top/AskaEth/Cyrene_v2/internal/runtime/memory"
|
|
"git.yeij.top/AskaEth/Cyrene_v2/internal/runtime/model"
|
|
"git.yeij.top/AskaEth/Cyrene_v2/internal/runtime/persist"
|
|
"git.yeij.top/AskaEth/Cyrene_v2/internal/runtime/tool"
|
|
"git.yeij.top/AskaEth/Cyrene_v2/internal/runtime/trace"
|
|
)
|
|
|
|
// Options configures a Loop. All fields are optional and nil is tolerated until
|
|
// the step that actually needs the field: the agent is required for Run (Run
|
|
// returns ErrNilAgent otherwise); a missing tool dispatcher causes a requested
|
|
// call to be fed back as an error; a missing retriever/persona skips memory
|
|
// injection; a missing extractor/store skips the memory hook; a missing
|
|
// checkpoint skips persistence; a missing tracer skips tracing.
|
|
type Options struct {
|
|
// Agent is the loop's execution unit: an instruction string plus a backend
|
|
// adapter, an optional tool dispatcher and optional guardrails.
|
|
Agent *agent.Agent
|
|
// Tools is the tool dispatcher used to execute model-requested calls. When
|
|
// nil the loop falls back to the agent's own dispatcher (if any).
|
|
Tools *tool.Dispatcher
|
|
// Memory is the long-term memory store used to persist extracted items.
|
|
Memory *memory.Store
|
|
// Retriever injects relevant memories into the context, re-evaluated every
|
|
// round (see assembleContext) so retrieval tracks the latest conversation.
|
|
Retriever *memory.Retriever
|
|
// WorkingMemory is the per-session sliding window, provided for the
|
|
// orchestrator to inspect; the loop does not mutate it.
|
|
WorkingMemory *memory.WorkingMemory
|
|
// Extractor derives memory items to persist after every step.
|
|
Extractor memory.Extractor
|
|
// Checkpoint persists a durable snapshot of the loop state at every step.
|
|
Checkpoint *persist.Store
|
|
// Tracer emits a structured span for every step, LLM call and tool call.
|
|
Tracer *trace.Tracer
|
|
// Guards validates request/response payloads on the streaming path, where
|
|
// RunStep is bypassed. On the non-streaming path the agent runs its own
|
|
// guardrails inside RunStep.
|
|
Guards *guardrail.Pipeline
|
|
// PersonaID scopes memory retrieval and extraction. Blank skips both.
|
|
PersonaID string
|
|
// MaxRounds is the tool-loop upper bound. Non-positive selects
|
|
// DefaultMaxRounds (8).
|
|
MaxRounds int
|
|
// OnStream, when non-nil, switches the loop to the streaming path: the model
|
|
// adapter's ChatStream is driven directly and each structured stream event is
|
|
// forwarded to this callback (model.StreamEvent), so the frontend can render
|
|
// body text, tool calls and (eventually) reasoning segments independently.
|
|
OnStream func(model.StreamEvent)
|
|
// ToolExecution selects how a batch of tool calls is run: "parallel" (the
|
|
// default) or "sequential". Empty means parallel.
|
|
ToolExecution string
|
|
// ToolParallelism caps how many tool calls run concurrently when ToolExecution
|
|
// is "parallel". Non-positive selects DefaultToolParallelism (4).
|
|
ToolParallelism int
|
|
// Hooks installs optional per-batch/per-call tool-loop hooks. Every field is
|
|
// nil-tolerant; these are reserved for "敏感操作代审" (design/11) and the
|
|
// future asynchronous-tool path. A zero Hooks value is a no-op.
|
|
Hooks ToolLoopHooks
|
|
// MaxContextTokens is the context budget that triggers overflow compaction.
|
|
// Non-positive (the zero value) disables compaction, preserving the loop's
|
|
// previous behavior. Setting it to a positive value (e.g.
|
|
// DefaultMaxContextTokens) enables compaction: when the loop's estimated
|
|
// context exceeds the budget it summarizes older turns into a leading summary
|
|
// message while keeping the recent tail (including trailing tool results)
|
|
// intact (design §1.3). Nil-tolerant: leaving it unset changes nothing.
|
|
MaxContextTokens int
|
|
// Compactor compacts an over-budget conversation. Nil selects the built-in
|
|
// structural compactor (no model call); a non-nil value lets a caller inject
|
|
// an LLM-backed or custom summarizer. It is only invoked when
|
|
// MaxContextTokens is positive.
|
|
Compactor Compactor
|
|
// TransformContext, when non-nil, is called immediately before every model
|
|
// step with the assembled message list (after memory injection and any
|
|
// compaction) and the current round number (1-based). It may project, filter
|
|
// or re-order the messages the model sees. It is the loop's equivalent of
|
|
// OpenClaw's transformContext (design §1.3). Nil = no projection.
|
|
TransformContext func(ctx context.Context, messages []model.Message, round int) []model.Message
|
|
// Log is the base logger. Nil is replaced with a discard logger tagged with
|
|
// source "runtime/loop".
|
|
Log *slog.Logger
|
|
}
|