Skip to main content
US Army Corps of EngineersInstitute for Water Resources, Risk Management Center

AI-Assisted Development

Using AI Assistants

Developers may use Claude Code or Codex for AI-assisted development. To ensure consistency:

Required files in every repository:

  • CLAUDE.md or AGENT.md - Shared AI assistant instructions (see separate document)

AI assistant best practices:

  1. Always check AI suggestions - AI is a tool, not a replacement for judgment
  2. Review generated code carefully - Ensure it meets team standards
  3. Test AI-generated code thoroughly - Don't assume it's correct
  4. Update the assistant guidance file (CLAUDE.md or AGENT.md) - If the assistant repeatedly makes the same mistake

Prompting standards (every session):

  • Explicitly instruct the agent to read and follow CLAUDE.md or AGENT.md before doing any work
  • Require a brief implementation plan before any code revisions
  • Reinforce instruction adherence by asking the agent to confirm it read the file(s)
  • Repeat the instruction if the agent deviates and ask it to re-check the guidance

What AI should help with:

  • Boilerplate code generation
  • Code refactoring
  • Writing tests
  • Documentation
  • Debugging assistance
  • Explaining complex code

What requires human decision:

  • Architecture decisions
  • Security-sensitive code
  • Performance-critical code
  • Database schema changes
  • API contract changes

Version Control with AI Assistants

When using Claude Code or Codex:

  • Review all changes before committing
  • Don't commit AI output as one massive commit
  • Break AI-generated changes into logical commits
  • Authors write their own commit messages (don't just use AI suggestions)