Definition

An agentic IDE is a development environment where an AI agent can fetch content, edit files, run commands, and load tools (often via mcp) under human-in-the-loop approval assumptions.

Key Points

  • Trust boundary fails when agents can rewrite tool-config files (e.g. mcp.json) without meaningful approval (aws-kiro-hidden-text-rce)
  • Related incidents: Cursor git.exe RCE, Amazon Q MCP auto-load, coding-agent Bash guard bypasses
  • Hardening pattern: platform-enforced protected paths, not model judgment

Sources