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-research`, or explicitly asks Ralph Specum in Codex to run the research phase.
.claude/skills/tzachbon-ralph-specum-research/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-04 | ✗→✓ | ▲ Improved | 214% | 0% |
| case-09 | ✗→✓ | ▲ Improved | 48% | 0% |
| case-10 | ✗→✓ | ▲ Improved | 42% | 0% |
| case-14 | ✗→✓ | ▲ Improved | 112% | 0% |
| case-16 | ✗→✓ | ▲ Improved | -14% | 0% |
You are a coordinator, not a researcher -- delegate ALL work to a research-analyst sub-agent.
.current-spec.claude/ralph-specum.local.md when present./specs.progress.md, current state, indexed codebase context, related specs, and epic context when present.research-analyst sub-agent. Pass the goal, existing context, and interview results. The sub-agent writes research.md in the spec directory. Do NOT write research.md yourself.phase: "research" and awaitingApproval: true (or false when --quick is active)..progress.md with the research summary, blockers, learnings, next step, and verification tooling notes when relevant.--quick: STOP HERE. Display the walkthrough summary and approval prompt. Do NOT continue to requirements. Wait for the user to explicitly approve and request the next phase.--quick: Continue directly into requirements.The result should identify existing code patterns, external references, constraints, related specs, risks, verification tooling, and a clear recommendation for the next phase.
research.md, name research.md and summarize the research briefly.approve current artifactrequest changescontinue to requirementscontinue to requirements as approval of research.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 | 14,969 | 5,562 | -63% | 1 | 1 | 0% | 2,272 | 855 | -62% | 0 | 0 | — |
case-02 | fail→fail | 4,106 | 10,761 | +162% | 1 | 1 | 0% | 226 | 825 | +265% | 0 | 0 | — |
case-03 | fail→fail | 3,788 | 5,192 | +37% | 1 | 1 | 0% | 115 | 765 | +565% | 0 | 0 | — |
case-04 | fail→pass | 3,652 | 3,795 | +4% | 1 | 1 | 0% | 350 | 1,099 | +214% | 0 | 0 | — |
case-05 | fail→fail | 14,624 | 11,021 | -25% | 1 | 1 | 0% | 2,147 | 1,021 | -52% | 0 | 0 | — |
case-06 | fail→fail | 12,257 | 6,455 | -47% | 1 | 1 | 0% | 1,873 | 869 | -54% | 0 | 0 | — |
case-07 | fail→fail | 7,906 | 16,045 | +103% | 1 | 1 | 0% | 907 | 2,416 | +166% | 0 | 0 | — |
case-08 | fail→fail | 38,156 | 7,594 | -80% | 1 | 1 | 0% | 6,175 | 1,141 | -82% | 0 | 0 | — |
case-09 | fail→pass | 6,431 | 9,848 | +53% | 1 | 1 | 0% | 907 | 1,339 | +48% | 0 | 0 | — |
case-10 | fail→pass | 6,041 | 4,744 | -21% | 1 | 1 | 0% | 830 | 1,175 | +42% | 0 | 0 | — |
case-11 | fail→fail | 4,747 | 6,173 | +30% | 1 | 1 | 0% | 213 | 776 | +264% | 0 | 0 | — |
case-12 | pass→fail | 3,425 | 7,051 | +106% | 1 | 1 | 0% | 569 | 1,000 | +76% | 0 | 0 | — |
case-13 | fail→fail | 19,359 | 4,596 | -76% | 1 | 1 | 0% | 2,469 | 662 | -73% | 0 | 0 | — |
case-14 | fail→pass | 7,766 | 11,474 | +48% | 1 | 1 | 0% | 1,089 | 2,313 | +112% | 0 | 0 | — |
case-15 | fail→fail | 7,875 | 2,184 | -72% | 1 | 1 | 0% | 1,070 | 816 | -24% | 0 | 0 | — |
case-16 | fail→pass | 9,168 | 4,713 | -49% | 1 | 1 | 0% | 1,369 | 1,182 | -14% | 0 | 0 | — |
case-17 | fail→fail | 11,707 | 3,799 | -68% | 1 | 1 | 0% | 1,724 | 1,058 | -39% | 0 | 0 | — |
case-18 | pass→fail | 3,357 | 6,870 | +105% | 1 | 1 | 0% | 479 | 770 | +61% | 0 | 0 | — |
case-19 | pass→pass | 6,277 | 9,227 | +47% | 1 | 1 | 0% | 1,035 | 1,925 | +86% | 0 | 0 | — |
case-20 | pass→fail | 25,007 | 13,999 | -44% | 1 | 1 | 0% | 4,643 | 2,768 | -40% | 0 | 0 | — |
case-21 | pass→fail | 22,182 | 20,443 | -8% | 1 | 1 | 0% | 4,409 | 2,863 | -35% | 0 | 0 | — |
case-22 | pass→fail | 13,526 | 10,073 | -26% | 1 | 1 | 0% | 2,309 | 2,104 | -9% | 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 12 counted toward the lift figure. The other 10 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 0 percentage points is the difference between those two pass rates over the 12 comparable cases. 5 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.