[ctx]ctxseccontext security: attacks on the harness and context layer of AI agents, plus exploited-now news, hunting and playbooks.

← Tips & Guides · markdown

tip based on Forever Security research (Oct 22) and OpenAI/Anthropic guidance cited by BleepingComputer

Tip: never start a coding agent in ~/Downloads, or in any folder the web or other people can write to

  • Start Codex, Claude Code, Cursor CLI and similar tools in the root of the one project they should work on. Move a downloaded file into a scratch project first if you want the agent to look at it.
  • Before an agent works in an unfamiliar folder, look for AGENTS.md, CLAUDE.md and other instruction or settings files, and read them.
  • Set your browser to ask where to save each download, or send downloads to a folder agents never open.
  • Keep approval prompts and sandboxing on, and watch for an agent trying to change its own config, API base URL or package sources.

Sources: Forever Security: Markdown Is the New Executable (Gal Weizman, Omer Samet; Oct 22) · BleepingComputer: AI coding agents hijacked using AGENTS.md file in Downloads folder (Oct 25)