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-requirements`, or explicitly asks Ralph Specum in Codex to run the requirements phase.
.claude/skills/tzachbon-ralph-specum-requirements/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-13 | ✗→✓ | ▲ Improved | -33% | 0% |
| case-19 | ✗→✓ | ▲ Improved | -36% | 0% |
| case-07 | ✓→✗ | ▼ Worse | -22% | 0% |
| case-18 | ✓→✗ | ▼ Worse | -47% | 0% |
| case-21 | ✓→✗ | ▼ Worse | -79% | 0% |
You are a coordinator, not a product manager -- delegate ALL work to a product-manager sub-agent.
.current-specresearch.md when present, .progress.md, and the current state.awaitingApproval: false before generation.product-manager sub-agent. Pass research context, goal, and interview results. The sub-agent writes requirements.md. Do NOT write requirements.md yourself.phase: "requirements" and awaitingApproval: true (or false when --quick is active)..progress.md with approved research context, user decisions, blockers, next step, and any epic constraints that must carry forward.--quick: STOP HERE. Display the walkthrough summary and approval prompt. Do NOT continue to design. Wait for the user to explicitly approve and request the next phase.--quick: Continue directly into design.The result should include user stories, acceptance criteria, functional requirements, non-functional requirements, dependencies, exclusions, and success criteria.
requirements.md, name requirements.md and summarize the requirements briefly.approve current artifactrequest changescontinue to designcontinue to design as approval of requirements.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 | 3,830 | 5,863 | +53% | 1 | 1 | 0% | 209 | 780 | +273% | 0 | 0 | — |
case-02 | fail→fail | 17,496 | 5,500 | -69% | 1 | 1 | 0% | 2,708 | 654 | -76% | 0 | 0 | — |
case-03 | fail→fail | 11,244 | 5,136 | -54% | 1 | 1 | 0% | 1,788 | 731 | -59% | 0 | 0 | — |
case-04 | fail→fail | 6,275 | 5,040 | -20% | 1 | 1 | 0% | 466 | 728 | +56% | 0 | 0 | — |
case-05 | fail→fail | 7,556 | 5,949 | -21% | 1 | 1 | 0% | 561 | 702 | +25% | 0 | 0 | — |
case-06 | fail→fail | 3,426 | 7,957 | +132% | 1 | 1 | 0% | 347 | 842 | +143% | 0 | 0 | — |
case-22 | pass→pass | 15,610 | 24,085 | +54% | 1 | 1 | 0% | 3,236 | 4,645 | +44% | 0 | 0 | — |
case-07 | pass→fail | 5,489 | 5,241 | -5% | 1 | 1 | 0% | 790 | 615 | -22% | 0 | 0 | — |
case-08 | fail→fail | 15,422 | 6,401 | -58% | 1 | 1 | 0% | 2,305 | 869 | -62% | 0 | 0 | — |
case-09 | fail→fail | 9,050 | 7,423 | -18% | 1 | 1 | 0% | 1,320 | 959 | -27% | 0 | 0 | — |
case-10 | fail→fail | 5,349 | 7,263 | +36% | 1 | 1 | 0% | 224 | 867 | +287% | 0 | 0 | — |
case-11 | fail→fail | 5,371 | 6,808 | +27% | 1 | 1 | 0% | 221 | 867 | +292% | 0 | 0 | — |
case-12 | fail→fail | 10,375 | 7,395 | -29% | 1 | 1 | 0% | 1,501 | 993 | -34% | 0 | 0 | — |
case-13 | fail→pass | 22,310 | 11,985 | -46% | 1 | 1 | 0% | 3,441 | 2,322 | -33% | 0 | 0 | — |
case-14 | fail→fail | 16,859 | 5,707 | -66% | 1 | 1 | 0% | 2,725 | 849 | -69% | 0 | 0 | — |
case-15 | fail→fail | 6,813 | 18,509 | +172% | 1 | 1 | 0% | 998 | 2,747 | +175% | 0 | 0 | — |
case-16 | fail→fail | 8,987 | 9,029 | +0% | 1 | 1 | 0% | 1,332 | 972 | -27% | 0 | 0 | — |
case-17 | pass→pass | 13,016 | 5,380 | -59% | 1 | 1 | 0% | 1,878 | 1,250 | -33% | 0 | 0 | — |
case-18 | pass→fail | 6,912 | 4,399 | -36% | 1 | 1 | 0% | 1,145 | 606 | -47% | 0 | 0 | — |
case-19 | fail→pass | 7,455 | 1,959 | -74% | 1 | 1 | 0% | 1,122 | 717 | -36% | 0 | 0 | — |
case-20 | pass→pass | 30,980 | 17,423 | -44% | 1 | 1 | 0% | 5,409 | 3,315 | -39% | 0 | 0 | — |
case-21 | pass→fail | 20,826 | 6,291 | -70% | 1 | 1 | 0% | 3,397 | 703 | -79% | 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 6 counted toward the lift figure. The other 16 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 6 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.