axiomhq/gilfoyle

Overview

This skill is an SRE incident-response agent modeled as Bertram Gilfoyle. It queries your observability stack, performs disciplined root cause analysis, and produces verifiable findings without handwaving. Use it when you need precise, data-backed diagnosis and remediation steps during outages or degradations.

How this skill works

On activation it runs scripts/init to discover available environments, datasets, and datasources. It then follows a strict investigation loop: discover, gather code context, form falsifiable hypotheses, execute targeted queries (logs, metrics, profiling), verify results, and record verified facts. All claims are traceable to query outputs and stored via the memory commands.

When to use it

• P1 incidents where traffic, errors, or latency spike and fast triage is required
• Root cause analysis after a deploy, config change, or feature-flag toggle
• Log investigations to map error strings back to code paths
• Comparative/differential analysis between good and bad time windows
• Profiling or performance regressions that need concrete telemetry evidence

Best practices

• Run scripts/init immediately to discover actual datasets and UIDs; never guess names
• Design queries to disprove hypotheses first; avoid confirmation bias
• Filter by time first and be selective with fields to keep queries fast
• Record verified facts and useful queries to memory right away
• Always include shareable links to the queries that support each finding

Example use cases

• Investigating sudden 500s: compare last 30m vs prior 30m with spotlight to find over-represented dimensions
• Confirming whether a deploy caused an outage: check changelog, then run error-rate and latency queries across services
• Finding a noisy backend dependency: grep logs for exact error strings, map to code, verify with metrics
• Mitigating a DDoS: identify traffic spike, block or rate-limit offending sources, announce rollback if needed
• Post-incident documentation: save the queries, findings, and incident summary to memory for postmortem