metrics
能把度量落地到代码与仪表盘，支撑可信决策与告警。
定义、实现与验证可操作的产品/工程/可靠性指标与监测方案。
帮你定义、实现并验证能真正驱动决策的指标和监测方案，从产品漏斗、KPI 到 SLI/SLO、DORA 指标都能覆盖。适合不知道该量什么、要埋点、需要仪表盘或报警设计，或怀疑现有指标误导决策时用。方法强调做最小可信的度量：明确指标来源与质量、埋点实现到代码和测试并做本地验证，确保指标能产生可执行的响应。
▸ 展开 SKILL.md 英文原文
Identifies, defines, implements, and verifies actionable software, product, delivery, reliability, and AI-system metrics. Use when the user asks what to measure, wants KPIs or a measurement plan, needs telemetry or instrumentation added to a repository, wants a dashboard or alert metric designed, or mentions funnels, adoption, retention, SLIs/SLOs, DORA metrics, experiment guardrails, evaluation metrics, metric quality, or cardinality. Also use when existing metrics are misleading, unactionable, expensive, privacy-sensitive, or inconsistent. Do not use for a one-off arithmetic calculation or for logs/traces when no metric or measurement decision is involved.
帮我安装这个 skill：https://raw.githubusercontent.com/lpbayliss/claude-skills/main/skills/metrics/SKILL.mdcurl -fsSL "https://raw.githubusercontent.com/lpbayliss/claude-skills/main/skills/metrics/SKILL.md"# Metrics Workflow Create the smallest trustworthy measurement system that supports a real decision. A metric is useful only when its meaning, source, quality, and response are clear. When the user asks to **add, create, instrument, or implement** metrics in a repository, carry the work through code, tests, and local verification. When they ask **what to measure**, stop at a decision-ready measurement plan unless implementation is also requested. ## Non-negotiables - Start from the decision or behaviour the metric must inform; do not begin with available fields or a generic KPI catalogue. - Inspect the current code, schemas, telemetry stack, naming conventions, dashboards, tests, privacy