Anthropic shipped its internal CI/CD on-call agent as a public setup kit, with skills in git and a lessons.md learning loop
The August 18 writeup describes Claude Tag running as Anthropic's first responder for CI failures, with a dedicated service account, MCP connectors to Datadog, Grafana, PagerDuty, GitHub and Kubernetes, and an orchestrator that spawns executor subagents to check dependencies in parallel. Standing instructions live as markdown skills in a GitHub repo, including an oncall.md with deterministic escalation criteria and an auto-updating lessons.md incident journal; median time to first analysis is 14 minutes. The generalized kit at github.com/anthropics/oncall-kit (21 stars, reference code, not maintained) is a five-phase gated setup that mines your own incident history into playbooks, keeps Claude strictly read-only, and validates drafted playbooks against held-out incidents before install.
↳ Follow the thread