Categories
Indexed Skills (24)
technology-job
One or two sentences: what this diagnoses or produces, and the concrete symptoms or requests that should trigger it. Name the error strings an engineer would type. 40-1024 characters.
aws-production-debugging
Diagnose AWS production problems - IAM AccessDenied, security group and NACL drops, load balancer target health failures, service quota and throttling errors, EBS and instance credit exhaustion, cross-account role failures, and VPC endpoint or NAT reachability. Use when an AWS API call is denied or throttled, traffic does not reach an instance or container, or an AWS-hosted service is degraded.
github-actions-debugging
Debug GitHub Actions workflows - jobs that do not trigger, permission and OIDC failures, cache misses, flaky or slow runners, secrets unavailable in forks, matrix and concurrency problems, deployment jobs that hang on approval, and pipelines that pass in CI but fail in production. Use when a workflow fails, does not run, is slow, or behaves differently from a local build.
postgresql-troubleshooting
Diagnose PostgreSQL production problems - connection exhaustion, slow queries and missing indexes, lock contention and blocked transactions, idle-in-transaction sessions, table and index bloat, replication lag, autovacuum falling behind, and transaction ID wraparound warnings. Use when a database is slow, refusing connections, blocking, or when queries that were fast are suddenly not.
container-image-review
Review a Dockerfile or container image for correctness, size, build speed, security and reproducibility - base image choice, layer caching, multi-stage builds, non-root users, pinned versions, secrets leaked into layers, signal handling and healthchecks. Use when reviewing a Dockerfile, when an image is unexpectedly large or slow to build, or before an image is used in production.
docker-debugging
Debug containers that fail to start, exit immediately, cannot reach the network, or behave differently from the host - exit codes, entrypoint and shell-form problems, missing files after a build, volume and permission errors, port publishing, DNS inside containers, and architecture mismatches. Use when a container will not run, runs differently than expected, or works locally and fails elsewhere.
postmortem-analysis
Write a blameless postmortem that produces real change - a factual timeline, contributing factors instead of a single root cause, detection and response analysis, and action items that are specific, owned and small enough to finish. Use after an incident is resolved, when reviewing a draft postmortem, or when the same incident keeps recurring.
kubernetes-networking
Diagnose traffic that does not reach a healthy Kubernetes pod - Service with no endpoints, intermittent 502/503/504 from an Ingress, connection refused or timeout between services, NetworkPolicy drops, CoreDNS resolution failures, and load balancers that never become ready. Use when pods are Running and Ready but clients still fail, or when service-to-service calls fail while the process itself is healthy.
kubernetes-production-readiness
Review a Kubernetes workload before it carries production traffic - probes, resource requests and limits, replica count and spread, PodDisruptionBudgets, graceful shutdown, rollout strategy, security context, image pinning, and observability. Use when reviewing a manifest or Helm chart, onboarding a service to production, or answering whether a deployment is safe to ship.
kubernetes-resource-debugging
Diagnose Kubernetes CPU and memory problems with evidence instead of guesswork - OOMKilled containers, CPU throttling, pods Pending on Insufficient cpu/memory, node pressure and evictions, QoS class effects, and requests or limits that are wrong in either direction. Use when a workload is slow, is being killed, will not schedule, or when someone is about to change a resource request or limit.
kubernetes-troubleshooting
Diagnose unhealthy Kubernetes workloads from symptom to root cause - CrashLoopBackOff, Pending, ImagePullBackOff, OOMKilled, Init:Error, readiness probe failures, pods stuck Terminating, and rollouts that never complete. Use whenever a pod, deployment, job or statefulset is not running as expected and the cause is not yet known, before changing any manifest.
linux-network-debugging
Debug network failures from a Linux host - connection refused versus timeout, conntrack table exhaustion, listen backlog overflow, ephemeral port exhaustion, TIME_WAIT accumulation, MTU and path MTU black holes, packet loss, and routing or firewall drops. Use when connections fail, hang, or reset from a host, and when tcpdump or ss output needs interpreting.
linux-performance
Find the bottleneck on a slow Linux host using the USE method - CPU saturation versus steal versus iowait, memory pressure and swap, disk latency and queue depth, run queue and context switching, and separating a slow host from a slow application. Use when a server or node is healthy but slow, latency is high with no errors, or when someone is about to resize an instance without knowing what is saturated.
linux-troubleshooting
Diagnose a Linux host that is unhealthy - disk full, out of memory, OOM killer activity, a service that will not start under systemd, runaway processes, inode exhaustion, clock skew, and failures that appear application-level but are actually host-level. Use when a server or node is degraded, a systemd unit fails, or when a containerised symptom needs to be traced to the host underneath.
dns-troubleshooting
Diagnose DNS failures - NXDOMAIN, SERVFAIL, intermittent resolution errors, stale records after a change, CoreDNS problems in Kubernetes, ndots and search domain surprises, negative caching, and TTL-driven propagation delays. Use whenever a name does not resolve, resolves to the wrong address, or resolution is slow or intermittent.
http-debugging
Debug HTTP failures through proxies and load balancers - what 502, 503, 504 and 499 actually mean, keepalive and idle-timeout races, retry storms, redirect loops, compression and chunked encoding problems, CORS failures, and header size or body limits. Use when an HTTP request fails or is slow and you need to know which hop produced the response.
tls-debugging
Diagnose TLS and certificate failures - expired certificates, incomplete chains that work in browsers but fail in code, hostname and SAN mismatches, SNI problems, protocol and cipher mismatches, mutual TLS failures, and clock skew. Use when a connection fails with a certificate or handshake error, or when HTTPS works from one client but not another.
observability-debugging
Use metrics, logs and traces together to find where a production problem actually is - starting from a symptom, narrowing by service, and correlating the three signals instead of staring at one. Use when a service is slow or erroring and the cause is not obvious, when dashboards disagree, or when you have telemetry but no answer.
prometheus-troubleshooting
Fix Prometheus problems - targets down or missing, metrics that disappear, high cardinality and OOMing Prometheus servers, queries that time out, rate() returning nothing, staleness, scrape failures, and alerts that do not fire or fire constantly. Use when Prometheus data is missing, wrong, or expensive, or when a PromQL query returns something unexpected.
slo-design
Design SLIs, SLOs, error budgets and burn-rate alerts that page a human only when users are actually affected. Use when defining reliability targets for a service, replacing threshold alerts that cause fatigue, choosing what to measure, deciding availability targets, or building alerting that survives contact with production.
secrets-management-review
Review how an application or infrastructure handles secrets - credentials in code, in Git history, in container layers, in Terraform state and in CI logs; Kubernetes Secret handling; rotation; least privilege; and what to do when a secret is exposed. Use when reviewing secret handling, onboarding a service, or responding to a leaked credential.
terraform-review
Review a Terraform or OpenTofu plan and its code before apply - find resource replacements, data loss, drift, hardcoded secrets, unpinned providers, missing lifecycle guards, and blast radius that the diff does not make obvious. Use whenever reviewing a terraform plan output, an infrastructure pull request, or a module before it is used in production.
terraform-safe-changes
Execute Terraform or OpenTofu changes against live infrastructure without causing an outage - staged applies, plan files, lifecycle guards, moved blocks instead of replacements, imports, state surgery, and rollback. Use when applying infrastructure changes to production, refactoring modules, renaming resources, importing existing infrastructure, or when a plan wants to destroy something it should not.
terraform-troubleshooting
Fix Terraform and OpenTofu errors - state lock timeouts, "resource already exists", provider authentication failures, cycle errors, "Invalid for_each argument", inconsistent final plan, failed applies that left state half-converged, and plans that will not converge. Use when a terraform command fails, hangs, or produces a diff that makes no sense.
Bio shown is the top-scored skill's repo description as a fallback — real GitHub bios land in a future update.