Agent 可观测性追踪器

分享海报

@user_638e256a/agent-trace

Agent observability tracer for the agent-cognitive-brain seven-layer architecture. Provides full-chain Span tracing (per-layer timing + status codes with parent-child relationships), decision path recording (reasoning → decision → execution with alternatives and confidence), metrics collection (P50/P95/P99 latency distributions, success/block/error rates, token consumption, tool call heatmaps), anomaly detection (baseline deviation alerts, dead-loop detection, silent failure detection), and exporters (JSON structured logs, OpenTelemetry-compatible format). This skill should be used when building agent observability, adding distributed tracing, implementing performance profiling, setting up anomaly detection, or creating decision audit trails. Trigger keywords include: Agent tracing, 可观测性, 链路追踪, Span, 性能分析, 指标采集, 异常检测, 决策审计, OTel, distributed tracing.