Anthropic Open Sources AI On-Call Kit: Claude Fixes Bugs in Minutes
新智元 · wechat · 2026-08-22
Anthropic revealed its internal AI-driven CI/CD and On-call system, centered around ClaudeTag. Integrated with tools like Grafana and Datadog via MCP, it automates the incident response lifecycle.
Key Features:
- Detection & Triage: Claude monitors alerts 24/7, filtering noise based on on-call playbooks to reduce fatigue.
- Orchestration: Upon incident escalation, an orchestrator agent spawns sub-agents to gather logs and metrics, generating comprehensive situational reports.
- Auto-Remediation: Claude submits PRs, manages feature flags for gradual rollouts, and verifies fixes automatically.
- Knowledge Accumulation: The system updates lessons.md after every incident. Claude reviews these past lessons before starting new investigations, learning from history—and even correcting human engineers' biases.
Context: To keep pace with the 8x increase in code output (mostly AI-generated), Anthropic open-sourced the "oncall-kit". This toolkit helps teams transition from manual incident review to AI-augmented workflows, using simulation data for training.
More from coding & agent
- Building an Astronomy App with Agent Skills — adnan_hashmi · 2026-08-22
- Vercel releases Is Agentic to score site readiness for AI agents — EdenEmarco177 · 2026-08-22
- After Hours of Tinkering: Jan.ai MCP Integration with ComfyUI Works Locally — tostane · 2026-08-22
- Content: Self-hosted MCP server for converting files to transcripts and audio — EgalitarianMonkey · 2026-08-22
- Grokbot 101: Always-on Cloud Agent Without SSH Hassle — eptwts · 2026-08-22
- Where should an AI agent's permissions actually be enforced? — Arc_bong · 2026-08-22