AgentSkillsCN

generate-golden-dashboard

根据蓝图的元数据与依赖关系，为蓝图生成可观测性资产（Grafana JSON、Prometheus 告警、LogQL/CloudWatch 查询）。

SKILL.md
--- frontmatter
name: generate-golden-dashboard
description: Generate observability assets (Grafana JSON, Prometheus alerts, LogQL/CloudWatch queries) for a Blueprint from its metadata and dependencies.

Generate Golden Dashboard (generate_golden_dashboard)

Use this skill when creating dashboards and alerts for a BaseBlueprint so every workflow is observable, SLO-aligned, and debuggable via the Four Golden Signals and dependency health.

When to Use

  • •Building Grafana dashboards, Prometheus/AlertManager rules, or log queries for a Blueprint
  • •Ensuring Latency, Errors, Traffic, Saturation and business metrics are covered
  • •Correlating logs with OTel trace_id and Blueprint metadata.id
  • •Tying alerts to Blueprint operations.sla and cost visibility

Instructions

  1. •Analysis: Read Blueprint metadata (id, version) and operations (SLA). Map each this.execute() to a "Dependency Health" panel.
  2. •Four Golden Signals: Implement panels for Latency (P50/P90/P99), Traffic (WORKFLOW_STARTED rate), Errors (WORKFLOW_FAILED vs COMPLETED), Saturation (worker slots, Dagger utilization).
  3. •Business value: Use recordMetric calls to add "Workload Success" panels (e.g., tickets created, bytes processed).
  4. •Logs: Generate LogQL/CloudWatch queries filtered by metadata.id and trace_id; include a "Dead Letter" panel for failed Saga/compensation logs.
  5. •Alerting: Create AlertingRule YAML with burn-rate thresholds aligned to 24h SLA; link to remediation playbooks.
  6. •Output: Provide dashboard.json (Grafana model), alerts.yaml, log_queries.txt. If Blueprint lacks business metrics, ask the user (HITL) for value milestones.

For the full prompt and principles, see references/generate-golden-dashboard.md.