Claude waffled on a straightforward CI fix instead of just doing it

User reported a docker/login-action@v3 GHCR timeout in their GitHub Actions workflow. Instead of immediately applying a fix (retry logic), Claude:

1. Asked the user what they wanted to do about it
2. Suggested "just re-run the workflow" multiple times
3. Made the edit but didn't commit or push, then asked the user if it should commit
4. Required multiple prompts to actually complete the task end-to-end

The fix was simple — wrap the login in nick-fields/retry@v3. Claude should have applied it, committed, and pushed without hand-holding.

Expected behavior: When a user shares a CI error and asks for a fix, apply the fix and ship it. Don't ask 5 clarifying questions for a straightforward problem.