After a compromise, this engineer moved all AI coding agents off his desktop
rseroter · x · 2026-11-17
Engineer Jared Wray shares the secure development workflow he rebuilt after the ChainDrop incident compromised his desktop.
- Two conclusions: agents can usually write and test code without external credentials, with CI/CD validating changes; the desktop holds sensitive files and credentials, so untrusted code there puts too much at risk
- Minimal desktop: Safari for web/email, Brave for Claude/Codex/Cursor/GitHub, VS Code via GitHub Codespaces so code runs in the cloud, and a password manager fully inaccessible to coding agents
- AI setup: the Keyv repo publishes his .cursor, .devcontainer, and .claude security configs, with .codex planned
More from coding & agent
- Compound launches IC Deck Bench, first benchmark for IC-ready PowerPoint agents — peterjliu · 2026-11-17
- Confluent's Kafka Agent Skills Praised for Quality, Ship With Built-in Evals — AdiPolak · 2026-11-17
- itr8 puts Claude, GPT, Gemini and Grok in one group chat to collaborate on your work — cantrell · 2026-11-17
- 1000-agent swarm pybot experiment writeup coming within days — lvwerra · 2026-11-17
- Developer: My Elaborate AI Workflows Kept Failing — Just Focus on the Outputs — brandon_galang · 2026-11-17
- Gemini CLI Fix Stops Custom Model Aliases From Silently Losing Generation Settings — sayaklearner-prog · 2026-11-17