Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Identify key variables in the causal system
.claude/skills/yogsoth-ai-knowledge-structuring-variable-identification/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-01 | ✗→✓ | ▲ Improved | 23% | 0% |
| case-18 | ✗→✓ | ▲ Improved | 21% | 0% |
| case-02 | ✗→✓ | ▲ Improved | 84% | 0% |
| case-03 | ✗→✓ | ▲ Improved | 95% | 0% |
| case-05 | ✗→✓ | ▲ Improved | 155% | 0% |
Systematically surface all variables that participate in the causal system under study. This strategy drives CC to read source material, extract candidate variables, create variable pages, and flag potential confounders before any mechanism edges are drawn.
CC must resist the urge to jump straight to causal edges. The quality of the entire causal model depends on having a complete, well-scoped variable inventory first. Every variable should be grounded in at least one source page, typed (exposure, outcome, mediator, moderator, confounder, or instrument), and given a plain-language description. Confounders deserve explicit flagging because they are the most common source of spurious causal claims.
| Metric | S | M | L | |--------|---|---|---| | Variables identified | 6 | 15 | 30 | | Source pages analyzed | 5 | 15 | 30 | | Confounders flagged | 2 | 5 | 10 |
| Metric | Target | Current | Status |
|---------------------|--------|---------|--------|
| Variables identified | S:6 / M:15 / L:30 | 0 | ⬜ |
| Source pages analyzed | S:5 / M:15 / L:30 | 0 | ⬜ |
| Confounders flagged | S:2 / M:5 / L:10 | 0 | ⬜ |<HARD-GATE> Cannot exit until 80% of budget met. Print state ledger before each iteration decision. </HARD-GATE>
<!-- BEGIN available-tables (generated) -->
Optional, no fixed order; the final leaf is always a sop.
| Tactic | When to use | | --- | --- | | counterfactual-reasoning | Tactic for reasoning about what would happen if variables were different — supports causal identification and intervention analysis. | | evidence-weighing | Tactic for assessing the strength and relevance of evidence for causal claims — distinguishes correlation from causation. |
Optional, no fixed order; the final leaf is always a sop.
| SOP | When to use | | --- | --- | | variable-page-creation | SOP for creating a variable page in the causal model — documents a measurable quantity with its properties. |
<!-- 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-01 | fail→pass | 50,293 | 72,442 | +44% | 1 | 1 | 0% | 4,993 | 6,161 | +23% | 0 | 0 | — |
case-18 | fail→pass | 12,020 | 12,081 | +1% | 1 | 1 | 0% | 1,895 | 2,284 | +21% | 0 | 0 | — |
case-02 | fail→pass | 46,054 | 60,047 | +30% | 1 | 1 | 0% | 4,811 | 8,854 | +84% | 0 | 0 | — |
case-03 | fail→pass | 26,745 | 51,046 | +91% | 1 | 1 | 0% | 4,542 | 8,849 | +95% | 0 | 0 | — |
case-04 | pass→fail | 13,078 | 34,760 | +166% | 1 | 1 | 0% | 2,086 | 6,572 | +215% | 0 | 0 | — |
case-05 | fail→pass | 19,557 | 38,788 | +98% | 1 | 1 | 0% | 2,779 | 7,075 | +155% | 0 | 0 | — |
case-06 | pass→fail | 26,900 | 49,783 | +85% | 1 | 1 | 0% | 4,739 | 8,791 | +86% | 0 | 0 | — |
case-19 | pass→pass | 15,213 | 39,095 | +157% | 1 | 1 | 0% | 2,398 | 6,496 | +171% | 0 | 0 | — |
case-07 | fail→pass | 16,473 | 23,194 | +41% | 1 | 1 | 0% | 2,849 | 1,623 | -43% | 0 | 0 | — |
case-08 | pass→pass | 20,975 | 37,706 | +80% | 1 | 1 | 0% | 2,634 | 3,936 | +49% | 0 | 0 | — |
case-09 | pass→pass | 10,518 | 30,609 | +191% | 1 | 1 | 0% | 1,428 | 7,117 | +398% | 0 | 0 | — |
case-10 | fail→pass | 5,882 | 6,391 | +9% | 1 | 1 | 0% | 1,119 | 1,759 | +57% | 0 | 0 | — |
case-11 | pass→pass | 17,791 | 18,345 | +3% | 1 | 1 | 0% | 2,293 | 3,833 | +67% | 0 | 0 | — |
case-12 | pass→pass | 15,943 | 19,304 | +21% | 1 | 1 | 0% | 1,745 | 2,878 | +65% | 0 | 0 | — |
case-13 | fail→pass | 18,667 | 10,645 | -43% | 1 | 1 | 0% | 3,379 | 1,449 | -57% | 0 | 0 | — |
case-14 | fail→pass | 21,896 | 30,896 | +41% | 1 | 1 | 0% | 2,257 | 6,950 | +208% | 0 | 0 | — |
case-15 | pass→pass | 42,186 | 20,536 | -51% | 1 | 1 | 0% | 2,012 | 1,802 | -10% | 0 | 0 | — |
case-16 | pass→pass | 14,280 | 27,528 | +93% | 1 | 1 | 0% | 2,191 | 5,672 | +159% | 0 | 0 | — |
case-17 | fail→pass | 18,618 | 3,747 | -80% | 1 | 1 | 0% | 1,021 | 1,344 | +32% | 0 | 0 | — |
case-20 | pass→pass | 5,962 | 16,099 | +170% | 1 | 1 | 0% | 940 | 3,693 | +293% | 0 | 0 | — |
case-21 | pass→pass | 8,382 | 16,485 | +97% | 1 | 1 | 0% | 1,419 | 2,957 | +108% | 0 | 0 | — |
case-22 | pass→pass | 5,068 | 13,756 | +171% | 1 | 1 | 0% | 806 | 2,765 | +243% | 0 | 0 | — |
case-23 | fail→pass | 13,371 | 3,984 | -70% | 1 | 1 | 0% | 2,377 | 1,117 | -53% | 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. 23 cases were attempted, and 22 counted toward the lift figure. The other 1 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 +39 percentage points is the difference between those two pass rates over the 22 comparable cases. 2 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.