backend-structured-logging

Backend Structured Logging

Purpose

Design structured JSON logging with consistent schema, context propagation, and sampling.

Agent Protocol

Trigger

Exact user phrases: "structured logging", "JSON logging", "log format", "logging best practice", "log levels", "distributed tracing", "log correlation", "structured log", "logging library", "log aggregation", "log output", "log schema", "correlation ID".

Input Context

Before activating, verify:

• Logging framework (Winston/Pino/Serilog/Log4j/logrus/zerolog)
• Log aggregation system (Elasticsearch/Loki/CloudWatch/Datadog)
• Compliance requirements (audit log retention, PII handling, access logs)

Output Artifact

Logging schema and configuration as formatted text.

Response Format

# Log schema (JSON fields)
# Log levels and sampling rules

// Logger configuration
// Context propagation middleware
// PII redaction config

No preamble. No postamble. No explanations. No filler/hedging/transitions. Compress output — why use many token when few do trick.

Completion Criteria

• Log schema defined with all required fields (timestamp, level, message, context)
• Log levels configured with production rules (ERROR/WARN/INFO, sampled DEBUG)
• Log output configured as JSON lines to stdout
• Context propagation implemented (correlation ID through all async boundaries)
• Sensitive data redaction with pattern-based masking
• Sampling strategy defined per log level

Max Response Length

200 lines of configuration and code.

Decision Tree

Which Log Level?

What happened?
  ├── Application cannot continue (OOM, DB connection lost for good)
  │   └── FATAL — page on-call immediately
  ├── Request failed, user got an error
  │   └── ERROR — alert if rate exceeds threshold
  ├── Request succeeded but degraded (fallback used, retry happened)
  │   └── WARN — investigate if persistent
  ├── State transition normal (order created, user registered)
  │   └── INFO — sampled in production
  ├── Need to trace request through the system
  │   └── DEBUG — header-activated per request
  └── Deep internal details (loop iterations, variable values)
      └── TRACE — never enabled in production