AgentSkillsCN

adk-observability

精通ADK智能体的可观测性、链路追踪与结构化日志记录。可用于调试复杂的推理轨迹、监控工具调用，以及分析隐含模型的输出。

SKILL.md
--- frontmatter
name: adk-observability
description: Expert in ADK agent observability, tracing, and structured logging in Python. Use for debugging complex reasoning traces, monitoring tool calls, and analyzing latent model outputs.

ADK Observability Specialist (Python Edition)

Philosophy & Architecture

Observability enables measurement of an agent's internal state via external telemetry and structured logs. Basic I/O monitoring is insufficient for reasoning-heavy agents.

Core Capabilities

  • •Reasoning Traces: Inspect the sequence of thought and tool selection.
  • •Structured Logs: Capture JSON-based execution history for Cloud Logging.
  • •Trace View (Local): Use adk web to visualize timeline, spans, and payloads.

Implementations

  • •LoggingPlugin: Console-based structured output.
  • •DebugLoggingPlugin: Exhaustive YAML capture for offline debugging.
  • •OpenTelemetry: Distributed tracing via Monocle integration.
  • •Read references/observability.md for full configuration details.

Success Criteria

  • •Valid logging module configuration (programmatic or CLI).
  • •Successful extraction of debugging info from the Trace View.
  • •Effective use of BigQuery for performance trend analysis.