Assess
Check Grafana alerts, Jenkins build status, and the CubeAPM service inventory.
grafana alert rule list -o jsonAn agent workspace wiring read-only observability, CI/CD, and infrastructure CLIs into one isolated credential environment so a coding agent can investigate incidents, run RCAs, and understand production behavior.
read-only, always
# Assess alerts and failed delivery jobs grafana alert rule list -o json jenkins job list --recursive --status FAILURE -o json # Follow the affected service into traces cubeapm traces search --service checkout --status error --last 1h -o json # Check whether the failure lines up with cluster events kubectl get events -n checkout --sort-by=.lastTimestamp | tail -30 # Correlate the timestamps and save the RCA under incidents/
Speaks the language of your stack
GrafanaJenkinsCubeAPMAWSGitHub ActionsKafkaKubernetesRedisMongoDBPostgreSQLMySQLElasticsearch
How an investigation flows
start broad, then narrow
Investigate errors
Search error traces and related logs, then open the trace waterfall for the affected service.
cubeapm traces search --status error --last 1hCheck metrics
Compare error rate, latency, and service health across the incident window.
cubeapm metrics query 'error_rate' --last 1hCheck deployments
Review deploy annotations, recent Jenkins builds, build logs, merged pull requests, and releases.
jenkins build list checkout/deploy --limit 5Map dependencies
Trace error propagation through the dependency graph and check the supporting infrastructure.
cubeapm traces get <trace-id> -o jsonCorrelate
Match deployment timestamps with error spikes, Grafana annotations, and cluster events.
kubectl get events -n checkout --sort-by=.lastTimestampEverything, in one workspace
A read-only toolbelt for a coding agent that investigates like your best on-call.
01
Incident investigations
Move from an alert to traces, metrics, logs, deployment history, and infrastructure state with a disciplined RCA workflow.
02
Cross-system correlation
Match error and latency windows with builds, releases, cluster events, queue lag, cache health, and database evidence.
03
Read-only safety
Production-only conventions, read-only roles, session safeguards, command allowlists, and approval prompts keep investigations read-shaped.
04
One operations toolbelt
Use twelve wired CLIs plus optional Elasticsearch access without switching credential contexts or stitching together separate workspaces.
05
Runtime-agnostic agents
Run the same workspace under Claude Code, Codex CLI, OpenCode, or another agent runtime that follows AGENTS.md.
06
Credential isolation
direnv pins CLI profiles and environment credentials inside the clone, keeping production access separate from global configuration.
Correlate the timeline
Wire together twelve read-only CLIs in one isolated credential environment so a coding agent can correlate signals across systems in one session.
correlate the timestamps
incident / checkout correlation view
01Grafana alert
grafana alert rule list -o jsonEstablish the active alert window.
02Jenkins build
jenkins job list --recursive --status FAILURE -o jsonCompare build start and completion times.
03CubeAPM error rate
cubeapm metrics query 'rate(http_requests_total{status=~"5.."}[5m])' -o jsonLocate the error spike.
04Kubernetes event
kubectl get events -n checkout --sort-by=.lastTimestamp | tail -30Check whether the failure lines up with cluster events.