infra-agent-skills

← All skills

Metric Query

metric-query

Read historical or current metric samples (time series) for a resource from a monitoring platform, for troubleshooting or before/after comparison. Use when you need observed numbers for a known metric over a time window. Covers Prometheus, Zabbix, and Amazon CloudWatch. Read-only. Assumes an authenticated API client for the chosen platform is already available. Not for creating monitors or acknowledging alerts (see alert-monitor-management).

ObservabilityByItential
Works with
View full skill on GitHubInstall docs
To install (framework-agnostic: plain markdown, nothing to compile or pip install. Copy the skill folder to wherever your agent framework looks for skills, e.g. for Claude Code):
git clone https://github.com/itential/infra-agent-skills.git
cp -r infra-agent-skills/skills/<skill-name> ~/.claude/skills/

Metric Query

Reads time-series samples for a named metric or monitored item over a bounded window from a monitoring platform, and reports them together with the scope and completeness of what came back.

When to use this skill

  • Fetching observed values (utilisation, error counters, latency) for a known metric and interval to support troubleshooting or to compare against an earlier window.
  • Checking whether a platform returned partial, truncated, or empty data for a query before drawing conclusions.
  • Not for creating monitors, thresholds or alert rules, or acknowledging alerts: see alert-monitor-management. Not for dashboards or log searches.

Universal concepts

This capability has no wire-level standard behind it; each platform defines its own query model. Where documented, the following hold on every platform listed:

  • A query names a series (metric or item, plus any label/dimension selection) and a time window.
  • Results carry timestamps and values, or aggregates over a period, and their meaning depends on the platform’s own aggregation, resolution, and units.
  • A result can be bounded, paginated, or aggregated by the platform; a bounded or aggregated result is not the full raw record.
  • An empty result or an error does not show that a resource is healthy, idle, or unmonitored.

Procedure

  1. Identify the platform from the request and find it in references/INDEX.md. If it is not listed, stop: this skill does not cover it.
  2. Open that platform’s procedure.md and follow it exactly.
  3. Report the outcome using that procedure’s own status vocabulary, with the query scope, window, and any warnings or incompleteness.

Known limitations

  • Read-only: no monitor, alert, retention, or collection changes, and no thresholds or diagnoses are produced.
  • The caller supplies the metric identity; discovery is covered only where a platform procedure includes it.

Contributing to this skill

This router changes only to correct shared concepts or to add a pointer to a newly supported platform. Platform detail belongs under references/<platform>/.

Tools

An authenticated API client for the selected platform. Each platform’s procedure.md names its operations and endpoints.md binds them.

Verification checklist

  • The platform, metric, and window in the result match the request.
  • Warnings, truncation, pagination state, and platform status values are carried into the report.
  • No health conclusion is drawn from an empty or failed read.