schematize
Skills · Claude Code · Agent Skills

A system you cannot see is a system you cannot operate — no negotiation.

OpenTelemetry with a Collector, correlated traces, metrics and logs, SLOs with burn-rate and alerts that point to a runbook. Observability is born with the service, not afterwards.

v0.2.1Disciplina · Observabilidadegithub.com/schematizeme/skill-observability

Onderhouden door Lucas Andrade

schematize-observability · v0.2.1bash
$ schematize install observability
GitHub →download skill (.zip)/observability-help · github.com/schematizeme/schematize-cli

Non-negotiable floors

Never switched off "to go faster". Full list in the skill.

OpenTelemetry and a Collector

The SDK speaks OTLP to a Collector, never to the backend; a vendor SDK in service code is forbidden.

Born with the service

A service is only done when it emits structured logs, metrics and traces, exposes probes and ships with a versioned dashboard, alerts and SLO.

Correlation in every log

trace_id and span_id in every JSON log, a static message with data in attributes, and deployment.environment.name on the Resource.

No PII or secrets in telemetry

Mask at the source and use redaction in the Collector; user_id, e-mail and IP never become a metric label.

Sampling never drops errors

100% of error or slow traces through tail sampling; the RED metric is not derived from sampled traces.

Every service has an SLO

Multi-window burn-rate on the error budget, and an exhausted budget freezes features.

Alerts are symptoms, with a runbook

User-symptom alerts with severity, for and runbook_url; an alert nobody acts on gets deleted.

Separate environments

Dev and hml telemetry never reaches the prd backend, and hml wakes nobody up.

Commands

Installed in .claude/commands/. Run /obs-help anytime.

/obs-alertsAudits or writes alerts (user symptom, severity, for, runbook) and runs check-alertas.mjs.
/obs-ccContext compact: writes context + checklist to the archive and runs /compact.
/obs-claudeCreates or merges the always-on observability CLAUDE.md at the repo root (won't overwrite other skills' blocks).
/obs-dashboardDashboard as code (SLO, RED, USE and links).
/obs-handoffWrites the context handoff without compacting — session end or task switch.
/obs-helpLists every command.
/obs-instrumentInstruments a service to the standard (SDK to Collector, semconv, JSON logs, healthchecks), in the house language set.
/obs-loadForce-loads the whole normative body and starts applying it.
/obs-reviewReviews a service's observability (signals, correlation, PII, cardinality, probes, SLO) and runs the gates.
/obs-sloDefines SLI, SLO and error budget in /docs/slo.md and generates the burn-rate rules.

For AI agents

Install with no human in the loop. Metadata in llms.txt.

  • Run: git clone https://github.com/schematizeme/skill-observability.git && ./skill-observability/install.sh
  • Or download: https://github.com/schematizeme/skill-observability/releases/latest/download/skill-observability.zip and unzip into .claude/skills/
  • Copy assets/CLAUDE.md to the repo root (always-on rule).
  • Confirm with /obs-help.

Wil je niet installeren?

Lees de volledige standaarden en kopieer ze rechtstreeks naar je project — of installeer de skill zodat de agent ze voor je afdwingt.

GitHub →Packs