Sentry Operations Skill

Default output: return only the result, blockers, and required evidence. Omit preambles, process narration, repeated context, confidence scores, and follow-up offers. Use at most five bullets unless a required artifact or schema needs more.

Sentry Operations

Current Versions (Verify Before Use)

Check the Sentry SDK releases for your platform. SDKs follow semver independently.

Core Principles

Instrument before you need it. Sentry is most valuable when it's already running before an incident.
Alert on new issues, not total volume. A spike in errors you've already seen is less actionable than a new error type.
Breadcrumbs tell the story. Good instrumentation includes HTTP requests, state transitions, and user actions leading to the error.
Performance is part of observability. Enable distributed tracing to connect frontend errors to backend root causes.

SDK Integration Checklist

Error Tracking

[ ] DSN configured via environment variable (not hardcoded)
[ ] release set to git SHA or version tag
[ ] environment set (production, staging, development)
[ ] Global error handler captures unhandled exceptions
[ ] Rejected promises are captured (frontend + backend)
[ ] Source maps uploaded for JS/TS (enables readable stack traces)

Performance

[ ] Tracing enabled with appropriate sample rate
[ ] Custom spans for critical operations (DB queries, external API calls)
[ ] Distributed tracing headers propagated between services
[ ] Database queries auto-instrumented or manually spanned

Context

[ ] User ID attached (hashed, not PII)
[ ] Tags for team/service/component
[ ] Breadcrumbs for state changes

Alert Configuration

// Example: Alert on new issues in production
// Sentry UI → Alerts → Create Alert Rule
// When: A new issue is created
// If: Environment equals "production"
// Then: Send notification to Slack #alerts

Alert hierarchy:

New issue in production → Immediate notification
Issue count > 100/hour → Page on-call (regression)
Performance regression > 20% → Slack notification
Release crash rate > 1% → Block deployment

Performance Monitoring

Key Metrics

Apdex: User satisfaction score (0-1)
P50/P95/P99 Latency: Response time distribution
Throughput: Requests per minute
Failure Rate: Error % of total requests

Custom Spans

const span = Sentry.startSpan({ name: 'process-payment' }, () => {
  // critical operation
});

Common Anti-Patterns

| Anti-Pattern | Why It's Wrong | Fix | |---|---|---| | Hardcoded DSN | Secret exposure, can't rotate per env | Environment variable | | No source maps | Stack traces are minified gibberish | Upload maps in CI | | 100% trace sampling | Expensive, unnecessary | 1-10% for high traffic | | No release association | Can't correlate errors with deployments | Set release to git SHA | | Ignoring caught errors | Silent failures accumulate | Log or capture explicitly | | PII in context | Privacy violations | Hash IDs, strip emails/phones |

Troubleshooting Flow

New error spike: Check release timing → correlate with deployment
Minified stack trace: Source maps not uploaded or mismatch
Missing errors: Check filters, sampling rate, DSN correctness
Performance blind spot: Check trace propagation between services
Alert fatigue: Review alert rules, add thresholds, use issue states

Agent Skills: Sentry Operations

Install this agent skill to your local

Skill Files