Reported by sourceThe source says this happened. Top 200 Repos has not reproduced it.
Build an on-call investigation agent from your own alert history
Goutham Veeramachaneni connected Claude Code to Grafana's gcx CLI, his code and runbooks, then trained its skill on two weeks of past pages for his services.
For Faro, a frontend observability backend with a few stateless services and MySQL, he gave Claude Code access to Grafana Cloud logs, traces, profiles, alerts and metrics through the gcx CLI, plus the codebase and team runbooks as files. He guided it through the previous two weeks of pages; after each investigation it updated its skill and knowledge base and proposed runbook changes as pull requests.
Results and limitations
He says it was on target after four or five guided investigations, that his team adopted it for troubleshooting and alert tuning, and he estimates false positives below 10%, though without formal evaluation.
Before you try it
No formal evals; scope is limited to his own services, it is triggered manually and cannot see deployment manifests. Early runs went off the rails and needed steering, and his view that this suits most services is an extrapolation.
Needs read-only observability credentials and a history of resolved alerts to learn from.
Try it yourself: Teach an agent to investigate past alerts
Fill the fields to adapt the prompt. Empty fields stay as visible [placeholders].
Preview
You are helping me build an on-call investigation skill for [Service]. Using read-only access through [Observability CLI], review the alerts from [Alert history window] one at a time. For each, gather the relevant logs, metrics and traces, propose a likely cause with evidence, and ask me to confirm or correct it before moving on. After each confirmed investigation, update [Skill file] with what you learned and draft, but do not merge, a runbook change in [Runbook folder]. Never silence alerts, edit dashboards, scale services or run write queries. Track how often your first guess turned out to be right.
Keep credentials read-only and measure accuracy on past incidents before relying on it for live pages.
Simon Willison sends research questions to cloud coding agents such as Claude Code for web and Codex Cloud, which write and run experiments in a dedicated repo and open a PR.
Install Anthropic's xlsx skill in Claude Code so the agent writes live formulas, recalculates the workbook with LibreOffice and checks it for formula errors.
Simon Willison gave Claude Code full permissions inside a Docker container on an NVIDIA Spark and had it get DeepSeek-OCR working; it took four prompts and under 40 minutes, with one nudge from him.
Red Hat's Roy Belio containerized Karpathy's autoresearch on OpenShift AI with H100 GPUs and let Claude Code run 198 training experiments over 24 hours.
B12 CTO Adam Marcus used Claude Code to analyze gigabyte core dumps filling web server disks and traced them to an RSS endpoint reading an endless radio stream.
The open-source Google Workspace MCP server exposes Calendar, Gmail and other Workspace tools to Claude Desktop, Claude Code and other MCP clients, using your own OAuth client and optional read-only mode.
GhidraMCP exposes Ghidra's decompiler and renaming tools over MCP, so an agent such as Claude Code can list functions, read decompiled code and propose clearer names.
