Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Use when a complex or risky code change should be split into Kent Beck's two-step method by first making the change easy through behavior-preserving preparatory refactoring, then making the intended behavior change once the design supports it. This should trigger for requests such as Apply two-step change; Make this risky change safer; Refactor before changing behavior; Separate preparation from behavior change. Part of Plinth Toolkit
.claude/skills/jabrena-051-design-two-steps-methods/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-01 | ✗→✓ | ▲ Improved | -22% | 0% |
| case-02 | ✗→✓ | ▲ Improved | -46% | 0% |
| case-03 | ✗→✓ | ▲ Improved | 3% | 0% |
| case-07 | ✗→✓ | ▲ Improved | 0% | 0% |
| case-09 | ✗→✓ | ▲ Improved | 19% | 0% |
Guide Java developers through complex or risky changes by keeping design preparation separate from behavior modification. This is an interactive SKILL.
What is covered in this Skill?
Separate behavior-preserving preparation from behavior-changing work, and validate after each step.
Read references/051-design-two-steps-methods.md, inspect the relevant code and tests, and state the exact behavior or capability that must change. Identify why the current design makes the change complex, risky, or hard to verify.
Choose the smallest behavior-preserving refactoring that reduces the obstacle: extract method or class, clarify names, isolate dependencies, add seams for testing, move responsibilities, improve types, or add characterization tests before touching behavior.
Apply Step 1 as focused preparatory refactoring only. Keep commits, notes, or task boundaries clear enough that reviewers can see no intended behavior change is included.
Run the relevant existing tests, build checks, characterization tests, or manual verification. If behavior changes unexpectedly, fix or revert the preparation before proceeding.
Apply the smallest intended behavior change now that the design supports it. Use focused Java, framework, persistence, messaging, API, or testing skills when detailed implementation guidance is needed.
Verify the intended behavior with targeted tests and relevant project validation. Report what was preparation, what changed behavior, what was verified after each step, and any remaining risks.
For detailed guidance, examples, and constraints, see references/051-design-two-steps-methods.md.
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→pass | 17,509 | 9,802 | -44% | 1 | 1 | 0% | 2,947 | 2,295 | -22% | 0 | 0 | — |
case-02 | fail→pass | 16,119 | 6,496 | -60% | 1 | 1 | 0% | 3,050 | 1,637 | -46% | 0 | 0 | — |
case-03 | fail→pass | 19,702 | 14,083 | -29% | 1 | 1 | 0% | 3,001 | 3,079 | +3% | 0 | 0 | — |
case-04 | pass→fail | 23,532 | 14,275 | -39% | 1 | 1 | 0% | 4,369 | 3,219 | -26% | 0 | 0 | — |
case-05 | pass→fail | 17,497 | 13,857 | -21% | 1 | 1 | 0% | 3,004 | 2,825 | -6% | 0 | 0 | — |
case-06 | pass→pass | 15,267 | 13,409 | -12% | 1 | 1 | 0% | 2,821 | 2,742 | -3% | 0 | 0 | — |
case-07 | fail→pass | 13,306 | 8,548 | -36% | 1 | 1 | 0% | 2,075 | 2,073 | -0% | 0 | 0 | — |
case-08 | fail→fail | 12,030 | 8,463 | -30% | 1 | 1 | 0% | 1,848 | 1,883 | +2% | 0 | 0 | — |
case-09 | fail→pass | 16,529 | 14,440 | -13% | 1 | 1 | 0% | 2,381 | 2,825 | +19% | 0 | 0 | — |
case-10 | fail→pass | 17,027 | 9,308 | -45% | 1 | 1 | 0% | 2,563 | 2,105 | -18% | 0 | 0 | — |
case-11 | fail→fail | 14,327 | 9,751 | -32% | 1 | 1 | 0% | 2,117 | 2,193 | +4% | 0 | 0 | — |
case-12 | fail→pass | 14,200 | 12,658 | -11% | 1 | 1 | 0% | 2,750 | 2,663 | -3% | 0 | 0 | — |
case-13 | fail→pass | 15,615 | 12,666 | -19% | 1 | 1 | 0% | 2,635 | 2,762 | +5% | 0 | 0 | — |
case-14 | fail→pass | 23,564 | 12,493 | -47% | 1 | 1 | 0% | 951 | 2,797 | +194% | 0 | 0 | — |
case-15 | fail→pass | 16,428 | 9,183 | -44% | 1 | 1 | 0% | 2,612 | 2,107 | -19% | 0 | 0 | — |
case-16 | fail→pass | 16,406 | 10,348 | -37% | 1 | 1 | 0% | 2,453 | 2,243 | -9% | 0 | 0 | — |
case-17 | fail→pass | 16,021 | 9,948 | -38% | 1 | 1 | 0% | 2,350 | 2,224 | -5% | 0 | 0 | — |
case-18 | fail→pass | 23,183 | 10,512 | -55% | 1 | 1 | 0% | 3,649 | 2,161 | -41% | 0 | 0 | — |
case-19 | fail→pass | 13,392 | 8,253 | -38% | 1 | 1 | 0% | 2,063 | 1,904 | -8% | 0 | 0 | — |
case-20 | fail→pass | 15,350 | 9,683 | -37% | 1 | 1 | 0% | 2,449 | 2,193 | -10% | 0 | 0 | — |
case-21 | fail→fail | 16,484 | 13,477 | -18% | 1 | 1 | 0% | 2,494 | 2,606 | +4% | 0 | 0 | — |
case-22 | fail→pass | 17,372 | 10,096 | -42% | 1 | 1 | 0% | 2,936 | 2,329 | -21% | 0 | 0 | — |
DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 22 cases were attempted, and 21 counted toward the lift figure. The other 1 produced results that are not comparable between the two arms, so they are excluded from the headline rather than averaged into it. The headline lift of +64 percentage points is the difference between those two pass rates over the 21 comparable cases. 2 cases got worse with the skill loaded, and they are included in that figure.
Without the skill loaded, the model failed this case. With it loaded, the same prompt on the same model passed. This is one improved case from the latest verified run; every case, including any that regressed, is in the table above.
Other measured skills in the registry, with their headline benchmark lift.