Skip to content

Latest commit

 

History

History
36 lines (24 loc) · 1.13 KB

File metadata and controls

36 lines (24 loc) · 1.13 KB

Design Principles for claude-code-java

This document explains the core philosophy behind the AI skills and workflows in this workspace.

1. Human-in-the-loop

  • Developer decides: The AI proposes changes; the developer reviews and approves.
  • No automatic commits or writes to git without explicit approval.
  • Skills generate outputs in a controlled environment, such as diffs or read-only views.

2. Reproducibility

  • Skills are versioned and reusable.
  • Workflows should produce consistent results when rerun with the same context.
  • Metrics and outputs are measurable: token usage, time saved, coverage improvements.

3. Safety

  • AI should never modify files outside its defined scope.
  • Any skill that writes to disk should:
    • indicate what files will change
    • allow for review before application

4. Transparency

  • Every proposed change should be explainable:
    • why the change is needed
    • what files are affected
  • Logs or explanations should be available alongside diffs.

5. Incremental adoption

  • Start with read-only / analysis skills.
  • Move to proposal skills.
  • Only then consider mutation skills with review.