AskaEth
de1fa83b79
修的是「文档说 ✅、生产不可达」：此前 Scheduler.Spawn 只有测试调用方、无 bus 入口、无派发工具；Span.Record 无生产调用方 → 泳道画不出、节点日志恒空。
- 入口：subagent_spawn 工具（internal/runtime/orchestration/spawntool.go，同包便于用真 Scheduler/loop/Tracer/Flow 做集成测试；注册在 cmd/cyrene/subagent_tool.go 照 terminal_tool 范式）。默认 owner-only，不设 AllowInGroup/AllowInStranger；安全仍由 restrictSchemas + applyToolSecurityGate 双层保证，未绕代审
- 父会话归属：AgentSpec.ParentSession（请求语义）→ Scheduler 转运行期 ctx 标记 loop.WithFlowBranch → 分支 span 以父 session 发布；**allow 谓词一行未改**；单图语义靠「分支 Run 不调 StartRound/EndRound」+ FlowPublisher 不新建图；未归属 spawn 一张图都不发（有测试锁）
- 前台最多等 2s（对齐 task.autoBackgroundThreshold）→ 合并结果作工具结果；转后台则广播 bus subagent.result（design/05 §9 回填 hook）；新增 Scheduler.Wait(ids) 按批等待（Join 会串并发批次，故不重用）
- 子代理跑在 WithoutCancel(ctx)：保留 span 归属、不被工具超时/回合结束顺手杀掉；生命周期由 PerAgentTimeout + CancelAll 约束；内部 WithMutedBroadcast 静音（不进前台 stream）
- 节点日志：loop/spanlog.go 新增 llm.done / tool.done / subagent.done；工具参数**只记键名不记值**；仅步骤真正结束才追加（跳过/被拒路径不发，不伪造）
- 脱敏（红线）：新增 trace/redact.go（形状规则 sk-/Bearer/JWT/key:value + 键名规则 api_key/token/authorization/password…，递归嵌套、不改调用方事件），graph.go 发布前必经；**顺带修掉一个既有泄漏**：result_summary prop 原先未脱敏（工具结果里的密钥会进图），由图级变异测试抓出
- 测试：新增 22 个（工具级 / 图级 / 日志级），含 4 次变异验证（去掉 WithFlowBranch、去掉 record*、还原未脱敏、去掉静音 各自 FAIL）；-race 过 orchestration+trace+loop
- 已知限制：真实模型端到端未验（无凭据/不起服务）；工具白名单是 design gap 未做；subagent.result 暂无消费者（design/15 未实现）；后台子代理跨轮次边界迟到 span 会丢泳道根；PerAgentTimeout=10min 待确认
68 lines
2.5 KiB
Go
68 lines
2.5 KiB
Go
package loop
|
|
|
|
import (
|
|
"context"
|
|
"fmt"
|
|
|
|
"git.yeij.top/AskaEth/Cyrene_v2/internal/runtime/agent"
|
|
"git.yeij.top/AskaEth/Cyrene_v2/internal/runtime/model"
|
|
)
|
|
|
|
// streamStep drives a streaming model call. It is the streaming counterpart of
|
|
// agent.RunStep: it builds the request from the agent's public surface, runs the
|
|
// loop's input guardrail, streams deltas through OnStream, runs the output
|
|
// guardrail and decodes any tool calls. Because RunStep does not expose a stream
|
|
// path, this method is owned by the loop; it mirrors the agent's semantics so the
|
|
// two behave identically apart from the delta callback.
|
|
func (l *Loop) streamStep(ctx context.Context, messages []model.Message) (*stepOutcome, error) {
|
|
adapter := l.agent.Adapter()
|
|
if adapter == nil {
|
|
return nil, fmt.Errorf("loop: stream step: %w", ErrNoAdapter)
|
|
}
|
|
req := l.buildRequest(ctx, messages)
|
|
|
|
if l.guards != nil {
|
|
verdict, err := l.guards.CheckInput(ctx, &req)
|
|
if err != nil {
|
|
return nil, fmt.Errorf("loop: check input: %w", err)
|
|
}
|
|
if verdict != nil && !verdict.Allow {
|
|
return &stepOutcome{final: true, text: verdict.Reason, stopReason: agent.StopReasonGuardrailInput}, nil
|
|
}
|
|
}
|
|
|
|
resp, err := adapter.ChatStream(ctx, req, func(ev model.StreamEvent) {
|
|
// Stamp the session/turn identity at the loop boundary: the adapter runs
|
|
// below the loop and cannot know which conversation the stream belongs to,
|
|
// so the loop fills it from the context it was driven with (design/16 §4.2)
|
|
// and the frontend can route the delta to the active session.
|
|
ev.SessionID = loopSession(ctx)
|
|
ev.TurnID = loopTurn(ctx)
|
|
// Suppress UI broadcast for internal autonomous passes (backthink),
|
|
// whose intermediate reasoning must not leak into the chat stream.
|
|
if l.onStream != nil && !isMutedBroadcast(ctx) {
|
|
l.onStream(ev)
|
|
}
|
|
})
|
|
if err != nil {
|
|
return nil, fmt.Errorf("loop: chat stream: %w", err)
|
|
}
|
|
|
|
if l.guards != nil {
|
|
verdict, err := l.guards.CheckOutput(ctx, &resp)
|
|
if err != nil {
|
|
return nil, fmt.Errorf("loop: check output: %w", err)
|
|
}
|
|
if verdict != nil && !verdict.Allow {
|
|
return &stepOutcome{final: true, text: verdict.Reason, stopReason: agent.StopReasonGuardrailOutput}, nil
|
|
}
|
|
}
|
|
|
|
if calls, ok := decodeToolCalls(resp); ok {
|
|
return &stepOutcome{final: false, toolCalls: calls, stopReason: agent.StopReasonToolCalls,
|
|
modelStopReason: resp.StopReason, model: resp.Model}, nil
|
|
}
|
|
return &stepOutcome{final: true, text: resp.Content, stopReason: agent.StopReasonStop,
|
|
modelStopReason: resp.StopReason, model: resp.Model}, nil
|
|
}
|