Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →This skill should be used only when the user explicitly asks to use `$ralph-specum-design`, or explicitly asks Ralph Specum in Codex to run the design phase.
.claude/skills/tzachbon-ralph-specum-design/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-15 | ✗→✓ | ▲ Improved | 9% | 0% |
| case-16 | ✗→✓ | ▲ Improved | -47% | 0% |
| case-21 | ✗→✓ | ▲ Improved | -12% | 0% |
| case-08 | ✓→✗ | ▼ Worse | -75% | 0% |
| case-11 | ✓→✗ | ▼ Worse | -77% | 0% |
You are a coordinator, not an architect -- delegate ALL work to an architect-reviewer sub-agent.
.current-specrequirements.mdrequirements.md. Read research.md when present, .progress.md, and current state.awaitingApproval: false before generation.architect-reviewer sub-agent. Pass requirements, research, and interview context. The sub-agent writes design.md. Do NOT write design.md yourself.phase: "design" and awaitingApproval: true (or false when --quick is active)..progress.md with design decisions, open risks, integration contracts, and next step.--quick: STOP HERE. Display the walkthrough summary and approval prompt. Do NOT continue to tasks. Wait for the user to explicitly approve and request the next phase.--quick: Continue directly into tasks.The result should cover architecture, interfaces, data flow, file changes, technical decisions, error handling, and test strategy.
design.md, name design.md and summarize the design briefly.approve current artifactrequest changescontinue to taskscontinue to tasks as approval of design.md.| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→fail | 25,239 | 4,570 | -82% | 1 | 1 | 0% | 4,335 | 774 | -82% | 0 | 0 | — |
case-02 | fail→fail | 16,947 | 6,410 | -62% | 1 | 1 | 0% | 2,845 | 843 | -70% | 0 | 0 | — |
case-03 | fail→fail | 20,784 | 5,984 | -71% | 1 | 1 | 0% | 3,741 | 796 | -79% | 0 | 0 | — |
case-04 | fail→fail | 17,706 | 4,871 | -72% | 1 | 1 | 0% | 2,894 | 708 | -76% | 0 | 0 | — |
case-05 | fail→fail | 14,327 | 6,843 | -52% | 1 | 1 | 0% | 2,227 | 780 | -65% | 0 | 0 | — |
case-06 | fail→fail | 33,718 | 5,268 | -84% | 1 | 1 | 0% | 5,130 | 792 | -85% | 0 | 0 | — |
case-07 | fail→fail | 19,928 | 5,249 | -74% | 1 | 1 | 0% | 3,223 | 757 | -77% | 0 | 0 | — |
case-08 | pass→fail | 18,748 | 9,526 | -49% | 1 | 1 | 0% | 3,067 | 754 | -75% | 0 | 0 | — |
case-09 | fail→fail | 24,343 | 5,834 | -76% | 1 | 1 | 0% | 4,312 | 786 | -82% | 0 | 0 | — |
case-10 | fail→fail | 11,877 | 6,564 | -45% | 1 | 1 | 0% | 1,894 | 750 | -60% | 0 | 0 | — |
case-11 | pass→fail | 24,997 | 6,825 | -73% | 1 | 1 | 0% | 3,816 | 862 | -77% | 0 | 0 | — |
case-12 | pass→pass | 5,170 | 6,013 | +16% | 1 | 1 | 0% | 761 | 851 | +12% | 0 | 0 | — |
case-13 | fail→fail | 9,514 | 6,861 | -28% | 1 | 1 | 0% | 1,399 | 780 | -44% | 0 | 0 | — |
case-14 | fail→fail | 21,766 | 6,253 | -71% | 1 | 1 | 0% | 3,859 | 839 | -78% | 0 | 0 | — |
case-15 | fail→pass | 6,995 | 4,692 | -33% | 1 | 1 | 0% | 1,022 | 1,116 | +9% | 0 | 0 | — |
case-16 | fail→pass | 9,506 | 2,050 | -78% | 1 | 1 | 0% | 1,406 | 745 | -47% | 0 | 0 | — |
case-17 | fail→fail | 21,686 | 7,069 | -67% | 1 | 1 | 0% | 3,226 | 823 | -74% | 0 | 0 | — |
case-18 | pass→pass | 12,566 | 3,355 | -73% | 1 | 1 | 0% | 1,910 | 948 | -50% | 0 | 0 | — |
case-19 | fail→fail | 12,877 | 2,453 | -81% | 1 | 1 | 0% | 1,850 | 795 | -57% | 0 | 0 | — |
case-20 | pass→pass | 8,697 | 4,498 | -48% | 1 | 1 | 0% | 1,353 | 1,142 | -16% | 0 | 0 | — |
case-21 | fail→pass | 11,626 | 6,594 | -43% | 1 | 1 | 0% | 1,678 | 1,473 | -12% | 0 | 0 | — |
case-22 | pass→pass | 7,091 | 6,264 | -12% | 1 | 1 | 0% | 1,107 | 1,339 | +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 8 counted toward the lift figure. The other 14 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 +5 percentage points is the difference between those two pass rates over the 8 comparable cases. 3 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.