Observability review

Design or review metrics, events, alerts, and diagnostics so they support real decisions and expose real failure.

Canonical skill profile. This page is generated from the repository contract. The linked SKILL.md remains the source of authority.

Jump to

• Overview
• When to use
• When not to use
• Core moves
• Expected output
• Verification
• Handoff signals
• Pairs well with
• Anti-patterns

View the canonical contract on GitHub.

Overview

Use this skill when the question is what to measure, how to interpret it, and how to avoid vanity or blind instrumentation.

When to Use

• instrumentation planning
• health metrics
• alerts
• operational dashboards
• diagnostics

When NOT to Use

• when there is no actual decision or operational risk behind the metric

Core Moves

1. Name the decision the signal should support.
2. Choose the signal type and unit of analysis.
3. Define the minimum reading needed.
4. State owner, threshold, and action.

Optional Modules

• Coverage gaps — Check whether silence could hide failure.
• Incentive review — Test whether the metric would drive harmful behavior.
• Segmentation pass — Break the signal by source, queue, user type, or time window when averages would hide degradation.
• Born-measurable check (feature governance) — When reviewing a feature being shipped, require that it is born measurable: a single primary metric (with direction), at least one guardrail metric that must not get worse, an owner, and 30/90-day review dates. These map to the primary_metric, guardrail_metrics, owner, and review_dates fields of the Feature Value Governance Contract. No primary metric → no broad rollout.

Activation Triggers

• Use coverage gaps when the absence of data could be misread as stability.
• Use incentive review when the metric may influence team behavior.
• Use segmentation when aggregation can hide partial failure.
• Use the born-measurable check whenever the subject is a feature heading to rollout; block broad rollout if primary metric, guardrail, owner, or review dates are missing.

Expected Output

• decision-linked metric or alert
• clear owner and cadence
• misread risk note

Verification

• The signal has an action attached to it.
• The unit of analysis is explicit.
• The reading distinguishes health from value when needed.

Handoff Signals

• Instrumentation depends on engineering changes by another owner.
• A policy or product decision is required before the metric can be finalized.

Pairs Well With

• feature-planning
• qa-review
• business-design

Anti-patterns

• Measuring what is easy instead of what is useful.
• Treating activity as value.
• Hiding partial degradation behind averages.