Debug Errors with Smart Diagnostics

Error Diagnostics Smart Debug is a skill for AI-assisted debugging, running a 10-step workflow from an error report through to a validated fix and prevention. It parses the issue for error messages/stack traces, reproduction steps, affected components, performance characteristics, environment, and failure pattern (intermittent vs consistent).

Initial triage uses a debugger subagent for pattern recognition, stack-trace analysis, dependency analysis, and 3-5 ranked hypotheses with a recommended strategy. For production/staging issues it gathers observability data - error tracking (Sentry, Rollbar, Bugsnag), APM metrics (DataDog, New Relic, Dynatrace), distributed traces (Jaeger, Zipkin, Honeycomb), log aggregation (ELK, Splunk, Loki), and session replays (LogRocket, FullStory) - querying for frequency/trends, affected cohorts, and deployment-timeline correlation. Each hypothesis gets a probability score, supporting evidence, falsification criteria, and a testing approach, categorized as logic errors, state management issues, integration failures, resource exhaustion, configuration drift, or data corruption. Strategy is matched to the issue: interactive step-through for locally reproducible bugs, observability-driven trace analysis for production issues, time-travel record/replay for complex state, chaos engineering for load-dependent intermittent failures, or statistical delta debugging for a small failure percentage. AI suggests optimal instrumentation points (entry points, decision nodes, state mutations, integration boundaries, error paths), and production-safe techniques include OpenTelemetry spans, feature-flagged debug logging, sampling-based continuous profiling, protected read-only debug endpoints, and canary traffic-shifted debug deployments.

Root cause analysis reconstructs the full execution path, tracks variable state at decision points, and identifies code smells and similar bug patterns. The fix step generates code changes with an impact assessment, risk level, test-coverage needs, and rollback strategy. Validation runs the test suite, compares performance to baseline, canary-deploys while monitoring error rate, and gets an AI code review - succeeding only if tests pass, there's no performance regression, and the error rate holds or improves. Prevention generates regression tests, updates the knowledge base with the root cause, adds monitoring for similar issues, and documents troubleshooting steps in a runbook.

Use it for a specific bug or production incident that needs structured, evidence-based root cause analysis rather than guesswork - especially when the failure is intermittent, production-only, or has an unclear cause. For a straightforward, locally reproducible bug with an obvious cause, its own strategy-selection guidance still points to simple interactive step-through debugging rather than the full observability-driven workflow.

Input is the issue description (error, reproduction steps, environment, frequency). Output is a structured report: issue summary, root cause with evidence, a fix proposal (code changes, risk, impact), a validation plan, and prevention steps (tests, monitoring, documentation).

It queries error-tracking (Sentry, Rollbar, Bugsnag), APM (DataDog, New Relic, Dynatrace), tracing (Jaeger, Zipkin, Honeycomb), log aggregation (ELK, Splunk, Loki), session replay (LogRocket, FullStory), and profiling (Pyroscope) tools, and instruments code with OpenTelemetry spans.

Developers and SREs debugging intermittent or production issues who want a structured, evidence-based process from hypothesis through validated fix and prevention rather than ad hoc guesswork.