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 `$ralph-specum-triage`, or explicitly asks Ralph Specum in Codex to triage a large effort into multiple specs.
.claude/skills/tzachbon-ralph-specum-triage/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-07 | ✗→✓ | ▲ Improved | -35% | 0% |
| case-13 | ✗→✓ | ▲ Improved | 1% | 0% |
| case-14 | ✓→✗ | ▼ Worse | 59% | 0% |
| case-16 | ✓→✗ | ▼ Worse | -73% | 0% |
| case-19 | ✓→✗ | ▼ Worse | -5% | 0% |
You are a coordinator, not a triage analyst -- delegate decomposition work to a triage-analyst sub-agent.
specs/_epics/<epic-name>/specs/.current-epicspecs/.current-epic. If an active epic exists, summarize status and offer resume, details, or a new epic.research.md, epic.md, .progress.md, and .epic-state.json as needed.triage-analyst sub-agent. The sub-agent runs the four-stage triage flow:Do NOT decompose or generate epic content yourself.
epic.md by aggregating and formatting the sub-agent's output (without altering substantive content) into:.epic-state.json with each spec, its status, and dependencies.specs/.current-epic to the active epic name.$ralph-specum-start for per-spec execution.The result should make it clear:
--quick: STOP HERE. Display the epic summary and approval prompt. Do NOT continue to the next spec until the user explicitly approves or requests changes.--quick: Continue directly to the first unblocked spec.epic.md, name epic.md and summarize the epic plan briefly.approve current artifactrequest changescontinue to the next speccontinue to the next spec as approval of epic.md.| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-06 | fail→fail | 18,460 | 4,740 | -74% | 1 | 1 | 0% | 2,830 | 793 | -72% | 0 | 0 | — |
case-01 | fail→fail | 20,168 | 4,536 | -78% | 1 | 1 | 0% | 3,406 | 1,052 | -69% | 0 | 0 | — |
case-02 | fail→fail | 16,072 | 5,765 | -64% | 1 | 1 | 0% | 2,602 | 929 | -64% | 0 | 0 | — |
case-03 | fail→fail | 18,954 | 7,107 | -63% | 1 | 1 | 0% | 3,454 | 915 | -74% | 0 | 0 | — |
case-04 | fail→fail | 2,523 | 4,855 | +92% | 1 | 1 | 0% | 181 | 724 | +300% | 0 | 0 | — |
case-05 | fail→fail | 14,951 | 7,733 | -48% | 1 | 1 | 0% | 2,433 | 1,130 | -54% | 0 | 0 | — |
case-07 | fail→pass | 21,244 | 8,740 | -59% | 1 | 1 | 0% | 3,148 | 2,044 | -35% | 0 | 0 | — |
case-08 | fail→fail | 7,179 | 5,374 | -25% | 1 | 1 | 0% | 1,011 | 770 | -24% | 0 | 0 | — |
case-09 | fail→fail | 4,908 | 5,889 | +20% | 1 | 1 | 0% | 707 | 827 | +17% | 0 | 0 | — |
case-10 | fail→fail | 10,661 | 3,557 | -67% | 1 | 1 | 0% | 1,619 | 916 | -43% | 0 | 0 | — |
case-11 | fail→fail | 7,247 | 5,583 | -23% | 1 | 1 | 0% | 1,023 | 782 | -24% | 0 | 0 | — |
case-12 | fail→fail | 7,951 | 7,140 | -10% | 1 | 1 | 0% | 1,126 | 915 | -19% | 0 | 0 | — |
case-13 | fail→pass | 13,335 | 8,824 | -34% | 1 | 1 | 0% | 2,003 | 2,015 | +1% | 0 | 0 | — |
case-14 | pass→fail | 4,155 | 5,046 | +21% | 1 | 1 | 0% | 511 | 815 | +59% | 0 | 0 | — |
case-15 | fail→fail | 28,783 | 7,863 | -73% | 1 | 1 | 0% | 5,552 | 1,007 | -82% | 0 | 0 | — |
case-16 | pass→fail | 19,935 | 6,535 | -67% | 1 | 1 | 0% | 3,420 | 918 | -73% | 0 | 0 | — |
case-17 | fail→fail | 15,603 | 6,986 | -55% | 1 | 1 | 0% | 2,262 | 896 | -60% | 0 | 0 | — |
case-18 | fail→fail | 14,366 | 6,629 | -54% | 1 | 1 | 0% | 2,066 | 826 | -60% | 0 | 0 | — |
case-19 | pass→fail | 14,348 | 11,719 | -18% | 1 | 1 | 0% | 2,585 | 2,446 | -5% | 0 | 0 | — |
case-20 | pass→fail | 6,327 | 5,524 | -13% | 1 | 1 | 0% | 995 | 757 | -24% | 0 | 0 | — |
case-21 | pass→fail | 18,794 | 5,613 | -70% | 1 | 1 | 0% | 3,077 | 801 | -74% | 0 | 0 | — |
case-22 | pass→pass | 22,337 | 16,853 | -25% | 1 | 1 | 0% | 3,669 | 3,643 | -1% | 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 -14 percentage points is the difference between those two pass rates over the 6 comparable cases. 7 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.