agentsclimarketplace

Ops observability

Skill barcelosvinicius/basic-engineering/plugins/be/skills/ops-observability

Claude Code plugin + npm base for AI-assisted engineering: 25 skills, 12 agents, slash commands, session-continuity hook. Also works with Copilot, Cursor, and others.

Install
npx -y skills add barcelosvinicius/basic-engineering --skill ops-observability

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

One thing to look at

  • 0 stars0 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

What its author says it does

Copied from the file, not written here

Use when adding logging, metrics, or tracing; defining SLOs and alerts; writing runbooks; or preparing a service for production operation. Structured logs, the four golden signals, alert hygiene, and runbook discipline.

SKILL.md

3.5 KB, as published. Nobody here has run it

Skill: Observability and Runbooks

Defines the minimum observability a service needs before production and how to keep it operable afterwards. Reference: engineering-principles.md §7 (Resilience) and §9 (Observability).

Structured logging

  • Structured format (JSON or key=value) — logs are queried, not read.
  • Correlation ID per request — generated at the edge, propagated to every downstream call and log line.
  • Levels with meaning: ERROR = requires action; WARN = degraded but self-healing; INFO = business-relevant events; DEBUG = off in prod.
  • Never log secrets or PII — scrub Authorization headers, passwords, tokens, documents/IDs at the middleware level (see sec-secrets-management).
  • Log the outcome of failures, not just the exception: what request, which user (pseudonymized), what state.

Metrics — the four golden signals

Every service exposes at minimum:

SignalMetricTypical alert
Latencyp50/p95/p99 per endpointp99 above SLO for 5 min
TrafficRequests/sDrop to ~0 (outage indicator)
ErrorsError rate (5xx, business failures)Rate above baseline
SaturationCPU, memory, pool/queue usageSustained > 80%

Plus a health check endpoint (/health) that verifies real dependencies (database ping, queue connectivity), monitored externally.

SLOs and alert hygiene

  • Define 2–3 SLOs per service (e.g., "99.5% of requests under 500 ms", "99.9% availability monthly") — alert on SLO burn, not raw spikes.
  • Every alert must be actionable — if the response to an alert is "ignore it", delete or tune the alert. Alert fatigue is an outage risk.
  • Every alert links to its runbook.

Runbooks

One runbook per recurring operation or failure mode, created from the base template templates/docs/runbook.template.md into docs/processo/runbooks/.

Minimum content per runbook:

  1. Symptom — what the operator sees (alert name, error pattern).
  2. Impact — who/what is affected.
  3. Diagnosis — exact commands/queries to confirm the cause.
  4. Mitigation — step-by-step, copy-pasteable, no decisions left implicit.
  5. Escalation — who to call when the steps don't work.

Update the runbook in the same PR as any change that alters the procedure. After an incident, fold what was learned into the runbook and docs/lessons-learned.md.

Tracing (when there is more than one service)

  • Propagate trace context (W3C traceparent) across HTTP/queue boundaries.
  • Use OpenTelemetry-compatible instrumentation so the backend is swappable.
  • Trace external calls and database queries — that is where latency hides.

Common mistakes

MistakeCauseSolution
Logs unsearchable in incidentFree-text loggingStructured format + correlation ID
Alert storm during deploysAlerting on raw spikesAlert on SLO burn rates with windows
Health check always greenEndpoint returns 200 unconditionallyVerify real dependencies
Runbook outdated at 3 a.m.Procedure changed, doc didn'tRunbook update in the same PR
PII in logsLogging whole request objectsAllowlist fields; scrub at middleware

Keep looking

Skills are one crate of 328,083. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.