Skip to content

Latest commit

 

History

History
35 lines (27 loc) · 1.12 KB

File metadata and controls

35 lines (27 loc) · 1.12 KB

Safe Workflows for claude-code-java Skills

This guide provides step-by-step workflows to maintain safety and control while using AI skills.

Workflow: Human-in-the-loop (Recommended)

  1. Analyze
    • Use read-only skills to inspect code, tests, or issues.
  2. Explain
    • AI summarizes findings, highlights potential changes.
  3. Propose
    • AI generates a diff or suggested edits in a sandbox.
  4. Review
    • Developer reviews all proposed changes.
    • Approves/rejects file by file.
  5. Apply
    • Changes are manually applied to project or git staging.
  6. Commit
    • Developer commits changes to git.
    • Optionally tag for auditing.

Levels of Skill Access

Level Description Recommended?
0 Read-only analysis ✅ Highly recommended
1 Proposal skills (sandbox write) ✅ Use with review
2 Auto-apply writes ⚠️ Only in very controlled experiments
3 Auto git commits ❌ Not recommended

Notes

  • Always check diffs before approving.
  • Never rely solely on AI-generated decisions for public APIs or critical functionality.