Telemetry

Telemetry & Observability

Three pillars: Errors (what broke), Metrics (how it's performing), Logs (what happened).

Error Tracking

Strategy

• Use a dedicated error tracking service (Sentry, Bugsnag, Rollbar, etc.)
• Capture unhandled exceptions automatically
• Add breadcrumbs at key application events for context
• Group errors by root cause, not by stack trace variation

Setup Pattern

1. Install SDK
2. Initialize early in application startup
3. Configure environment (dev/staging/production)
4. Set sample rate for performance tracing (10-20% is typical)
5. Upload source maps / debug symbols for readable stack traces

Breadcrumbs

Add context at key moments so error reports tell a story:

• User actions (navigation, form submissions, key interactions)
• State transitions (login, mode changes, configuration updates)
• External calls (API requests, DB queries)

Metrics

Key Metrics to Track

| Category | Metrics | |----------|---------| | Availability | Uptime, error rate, success rate | | Performance | Response time (p50, p95, p99), throughput | | Resources | CPU, memory, disk, connections | | Business | Active users, key actions, conversion |

Alerting Rules

• Alert on symptoms, not causes (high error rate, not "disk at 80%")
• Set meaningful thresholds based on actual baselines
• Include runbook links in alerts
• Avoid alert fatigue - every alert should require action

Logging

Levels

| Level | When | |-------|------| | ERROR | Something broke and needs attention | | WARN | Something unexpected but handled | | INFO | Significant application events | | DEBUG | Detailed diagnostic info (dev/staging only) |

Rules

• Structured logging (JSON) in production for parseability
• Include correlation IDs for request tracing
• Never log secrets, PII, or full request bodies with sensitive data
• Keep log volume manageable - noisy logs get ignored

Analytics Events

Define events that answer business questions:

EVENT: [name]
WHEN: [trigger condition]
PAYLOAD: [relevant data fields]
QUESTION IT ANSWERS: [what you learn from this]

Offline-First Pattern

Queue events locally when offline. Flush on connectivity. Batch inserts for efficiency. Dedup on the server if needed.

Phase Rollout

1. First: Error tracking (know when things break)
2. Next: Key metrics (know how things perform)
3. Then: Structured logging (know what happened)
4. Finally: Analytics (know how users behave)

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: pvnarp
• Source: pvnarp/agent-skills
• License: MIT

Install and usage instructions live in the source repository linked above.