Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Classify each pre-segmented text unit independently against a fixed label set (Argumentative Zoning, CoreSC, PubMed-RCT, Swales move/step, CODA-19, TDMS, or CSFCube's facet labels), single-layer with no cross-unit dependency. Use this after unit-segmentation has split the text, whenever a sentence- or clause-level rhetorical/functional classification is needed; do not use this for methods requiring document-level coreference reasoning (see multi-stage-cascade-extraction instead).
.claude/skills/yogsoth-ai-unit-classification/SKILL.md| Model | Eval pass | Runs |
|---|---|---|
| gemini-3.6-flash | 100% | 12 |
| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-06 | ✗→✓ | ▲ Improved | -51% | 0% |
| case-12 | ✗→✓ | ▲ Improved | -16% | 0% |
| case-22 | ✗→✓ | ▲ Improved | -60% | 0% |
| case-04 | ✓→✓ | ＝ Same ✓ | -61% | 0% |
| case-05 | ✓→✓ | ＝ Same ✓ | -6% | 0% |
Single-layer per-unit classification against a fixed, parameterized label set — no cross-unit or document-level dependency. Covers 7 methods (AZ/CoreSC/PubMed-RCT/NICTA-PIBOSO/CSAbstruct/CODA-19/Swales) plus TDMS's tuple-output variant, plus CSFCube's 3 facet labels as one more label_set option.
Subagent — spawned via spawn-agent skill.
An earlier graph draft tried to fold SciERC/SciREX into this node via a boolean toggle; the coverage audit (S6) found this doesn't work — those methods need document-level coreference clustering and (for SciREX) saliency judgment over ALL mentions in the paper, not per-unit independent classification. A boolean can't absorb that difference; they live in multi-stage-cascade-extraction instead.
csfcube-facet is documented as out-of-scope as its own SOP (its real task — multi-document pairwise relevance ranking — has no single-paper analog), but its 3 facet-label definitions (Background/Objective, Method, Result) are reused here as one more valid label_set option, per spec §3.
<!-- BEGIN available-tables (generated) -->
| SOP | When to use | | --- | --- | | spawn-agent | Spawn a customized CC subagent with full MCP tool access. |
<!-- END available-tables (generated) -->
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-10 | fail→fail | 20,498 | 12,159 | -41% | 1 | 1 | 0% | 3,094 | 1,609 | -48% | 0 | 0 | — |
case-01 | fail→fail | 8,752 | 11,793 | +35% | 1 | 1 | 0% | 614 | 1,590 | +159% | 0 | 0 | — |
case-02 | fail→fail | 10,198 | 14,263 | +40% | 1 | 1 | 0% | 854 | 1,913 | +124% | 0 | 0 | — |
case-03 | fail→fail | 8,713 | 14,819 | +70% | 1 | 1 | 0% | 676 | 2,150 | +218% | 0 | 0 | — |
case-04 | pass→pass | 22,552 | 9,095 | -60% | 1 | 1 | 0% | 2,761 | 1,065 | -61% | 0 | 0 | — |
case-05 | pass→pass | 12,516 | 9,065 | -28% | 1 | 1 | 0% | 1,260 | 1,183 | -6% | 0 | 0 | — |
case-06 | fail→pass | 34,348 | 9,509 | -72% | 1 | 1 | 0% | 2,311 | 1,137 | -51% | 0 | 0 | — |
case-07 | fail→fail | 19,902 | 10,229 | -49% | 1 | 1 | 0% | 2,494 | 1,253 | -50% | 0 | 0 | — |
case-08 | fail→fail | 22,296 | 13,693 | -39% | 1 | 1 | 0% | 2,787 | 1,801 | -35% | 0 | 0 | — |
case-09 | fail→fail | 19,524 | 7,827 | -60% | 1 | 1 | 0% | 2,342 | 843 | -64% | 0 | 0 | — |
case-11 | fail→fail | 16,455 | 7,435 | -55% | 1 | 1 | 0% | 1,749 | 711 | -59% | 0 | 0 | — |
case-12 | fail→pass | 21,903 | 15,785 | -28% | 1 | 1 | 0% | 2,698 | 2,256 | -16% | 0 | 0 | — |
case-13 | fail→fail | 20,504 | 8,561 | -58% | 1 | 1 | 0% | 2,714 | 947 | -65% | 0 | 0 | — |
case-14 | fail→fail | 22,756 | 19,346 | -15% | 1 | 1 | 0% | 2,613 | 2,557 | -2% | 0 | 0 | — |
case-15 | pass→pass | 17,157 | 9,258 | -46% | 1 | 1 | 0% | 1,937 | 1,104 | -43% | 0 | 0 | — |
case-16 | pass→pass | 19,533 | 11,727 | -40% | 1 | 1 | 0% | 2,114 | 1,571 | -26% | 0 | 0 | — |
case-17 | fail→fail | 18,684 | 10,391 | -44% | 1 | 1 | 0% | 1,979 | 1,363 | -31% | 0 | 0 | — |
case-18 | pass→pass | 20,026 | 11,915 | -41% | 1 | 1 | 0% | 2,536 | 1,524 | -40% | 0 | 0 | — |
case-19 | pass→pass | 23,858 | 15,991 | -33% | 1 | 1 | 0% | 2,687 | 2,079 | -23% | 0 | 0 | — |
case-20 | pass→pass | 19,097 | 10,931 | -43% | 1 | 1 | 0% | 1,999 | 1,280 | -36% | 0 | 0 | — |
case-21 | pass→pass | 16,511 | 10,622 | -36% | 1 | 1 | 0% | 1,823 | 1,375 | -25% | 0 | 0 | — |
case-22 | fail→pass | 16,000 | 7,103 | -56% | 1 | 1 | 0% | 1,702 | 687 | -60% | 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. The headline lift of +14 percentage points is the difference between those two pass rates over the 22 comparable cases.
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.