debug-master_skill

Overview

This skill is a senior-level SRE and debug architect focused on observability, distributed tracing, and autonomous incident remediation. It packages a proven incident resolution protocol, AI-assisted observability patterns, and agentic remediation loops designed to minimize MTTR and improve system resilience.

How this skill works

The skill inspects telemetry — metrics, logs, and OpenTelemetry traces — to construct an observability graph and surface the true blast radius of failures. It uses adaptive sampling, context propagation, and AI-assisted querying to correlate evidence, then proposes surgical fixes or safe rollback actions. Autonomous agents can execute non-destructive remediations with human-in-the-loop approval for destructive steps.

When to use it

• During active incidents to rapidly collect correlated evidence and determine blast radius.
• To implement or audit distributed tracing and OpenTelemetry instrumentation across services.
• When building agentic remediation workflows that require safe HITL gates for destructive actions.
• For running predictive observability and anomaly-detection campaigns to prevent outages.
• To standardize post-incident reporting and long-term knowledge storage in vector memory.

Best practices

• Always collect unified telemetry (metrics, logs, traces) before proposing fixes.
• Instrument standard spans and mandatory context propagation headers for cross-service calls.
• Use adaptive sampling: full capture for errors, low-rate sampling for healthy traffic.
• Prefer surgical fixes or feature-branch rollbacks over hotfixing production.
• Gate destructive agent actions behind human approval and audit logging.

Example use cases

• Triage a multi-region outage: correlate traces to find the failing service and isolate the faulty deployment.
• Automate cache invalidation and safe scaling in response to detected memory pressure via an agentic loop.
• Run weekly chaos experiments with agents to validate resilience and adjust dynamic SLOs.
• Detect slow memory leaks using predictive observability and create remediation runbooks automatically.
• Generate post-mortem summaries from correlated telemetry and store them in long-term vector memory for reuse.