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
47 lines
2.0 KiB
Go
47 lines
2.0 KiB
Go
package loop
|
|
|
|
import "context"
|
|
|
|
// loopSessionIDKey is the context key for the session id the current Run is
|
|
// driving. It is seeded at the top of Run and read where the loop emits a
|
|
// session-scoped UI event (capacity_awaken), so the event carries the right
|
|
// session rather than being treated as global.
|
|
type loopSessionIDKey struct{}
|
|
|
|
// withLoopSession returns a context carrying the Run's session id.
|
|
func withLoopSession(ctx context.Context, sessionID string) context.Context {
|
|
return context.WithValue(ctx, loopSessionIDKey{}, sessionID)
|
|
}
|
|
|
|
// loopSession returns the session id on ctx, or "" when none was seeded (a
|
|
// defensive fallback for a caller that drives the loop outside Run).
|
|
func loopSession(ctx context.Context) string {
|
|
s, _ := ctx.Value(loopSessionIDKey{}).(string)
|
|
return s
|
|
}
|
|
|
|
// SessionFromContext returns the session id the enclosing Run is driving, or ""
|
|
// when none was attached. It is exported so a tool executing inside the loop can
|
|
// attribute the work it starts (a spawned sub-agent, a background task) to the
|
|
// conversation that launched it — never inferred from model output.
|
|
func SessionFromContext(ctx context.Context) string { return loopSession(ctx) }
|
|
|
|
// loopTurnIDKey is the context key for the assistant turn the current Run is
|
|
// producing. The gateway attaches it (WithTurnID) before starting a turn so the
|
|
// streaming events the loop emits can be correlated to that turn by the client.
|
|
type loopTurnIDKey struct{}
|
|
|
|
// WithTurnID attaches turnID to ctx so the streaming events the loop emits carry
|
|
// the turn they belong to. It is exported because the gateway calls it once per
|
|
// turn, before Run; a caller that attaches none leaves the events' TurnID empty,
|
|
// which keeps the correlation fields backward compatible.
|
|
func WithTurnID(ctx context.Context, turnID string) context.Context {
|
|
return context.WithValue(ctx, loopTurnIDKey{}, turnID)
|
|
}
|
|
|
|
// loopTurn returns the turn id on ctx, or "" when none was attached.
|
|
func loopTurn(ctx context.Context) string {
|
|
s, _ := ctx.Value(loopTurnIDKey{}).(string)
|
|
return s
|
|
}
|