Radar MCP + Claude Code tops SREGym, fixing 95.2% of live Kubernetes faults
tianyin_xu · x · 2026-11-05
Skyhook announced that Claude Code, stripped of kubectl and given only Radar's MCP tools, now ranks #1 on SREGym's full 90-fault suite: 95.2% end-to-end diagnosis-and-fix rate on live Kubernetes clusters, 23 points ahead of #2 GitHub Copilot at 72.2%. Diagnosis takes 40.6s and a full fix 111.2s — roughly 6x faster — while using only 619K tokens per run, about a quarter of the runner-up's 2.45M. The benchmark covers OS-level faults like crashed kubelets, silent data corruption, retry storms, and recreations of real postmortems such as Cloudflare's 2019 WAF regex outage. SREGym (NeurIPS 2026) verifies fixes against the live system; Devin is also on the board and Datadog used it to validate Scout 1.0. Results were verified by the SREGym team; demo at KubeCon booth 365.
More from coding & agent
- OpenAI upgrades Codex: marked-up images now pinpoint edits via API — OpenAIDevs · 2026-11-05
- Oumi lets you plug in Claude or GPT to train a replacement model you fully own — testingcatalog · 2026-11-05
- AI agents find Linux kernel bugs and 10 novel exploitation techniques — fouadmatin · 2026-11-05
- Baseten Router cuts coding-task costs 72% with only 0.8pt pass-rate drop vs Opus 5.5 — baseten · 2026-11-05
- Mirage's Tesseract lets Claude or ChatGPT build 3D scenes and render video — LinusEkenstam · 2026-11-05
- BrowserForge: open-source MCP server gives agents your real browser with human approval gates — mostafazog · 2026-11-05