Skip to the content
AI and Cybersecurity Learning and Practice Center Test automation Open the Learning Center

Best practiceCoding assistants and CI

Review agent configuration files like CI configuration

OWASP ASI04OWASP ASI06

.cursor/rules, .claude/settings.json, AGENTS.md, .vscode/tasks.json and .mcp.json can make an agent run commands, trust a workspace or load hidden instructions. Rules File Backdoor hid instructions in invisible Unicode; Mini Shai-Hulud used Claude Code hooks for persistence.

What to do

  1. Add these paths to CODEOWNERS and require review for any change; diff them with invisible characters made visible.
  2. Do not auto-trust a cloned repository; open unknown projects in a sandbox until the config has been read.
  3. Keep hooks, allowed commands and MCP servers in version control and alert on drift on developer machines.

Sources

Check your own AI plan against the practices: Plan check