zero-script-qa_skill

Overview

This skill implements Zero Script QA: a log-driven testing methodology that verifies features without writing test scripts. It combines structured JSON logging, request_id propagation, and real-time Docker log monitoring so issues are detected and documented automatically. Use it to replace brittle scripted tests with observability-first verification and fast iteration cycles.

How this skill works

The skill enforces parseable JSON logs across services (timestamp, level, service, request_id, message, data) and streams those logs from Docker. Claude Code (or another analyzer) monitors the live log stream, detects error patterns, slow responses, consecutive failures, and abnormal status codes, then auto-documents issues with request_id context. Request IDs are generated at entry points and propagated through headers so the full flow is traceable across API, frontend, and nginx layers.

When to use it

• When you need feature verification via log analysis instead of maintaining test scripts
• For integration and staging environments running in Docker compose
• When rapid UX-driven manual testing is paired with automated log analysis
• To detect runtime errors, performance regressions, and cross-service flow issues
• When you want fast feedback and iterative PDCA cycles for AI-native development

Best practices

• Log everything relevant: all API calls (including 200), errors, and key business events
• Use a consistent JSON schema and required fields so analyzers can parse reliably
• Generate and propagate a request_id at the entry point and include it in all logs
• Set environment-specific minimum log levels (DEBUG for local/staging, INFO for prod)
• Stream Docker logs and filter by level or request_id for focused analysis

Example use cases

• Run a manual signup flow in the browser and have logs show the full backend/frontend trace by request_id
• Detect a slow endpoint when duration_ms exceeds thresholds and auto-create a ticket with the log snippet
• Monitor Docker logs for 5xx spikes and trigger immediate alerts with related request_ids
• Use Claude Code to aggregate consecutive failures on an endpoint and suggest probable root causes
• Verify production log level policy after deployment without modifying application code