Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Focus within the user's chosen field(s). Identify specific sub-directions through deep paper and web research, then present ranked candidates. Use after landscape-reconnaissance has identified fields of interest.
.claude/skills/yogsoth-ai-direction-narrowing/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-06 | ✗→✓ | ▲ Improved | -11% | 0% |
| case-02 | ✗→✓ | ▲ Improved | 95% | 0% |
| case-03 | ✗→✓ | ▲ Improved | 41% | 0% |
| case-04 | ✗→✓ | ▲ Improved | 122% | 0% |
| case-05 | ✗→✓ | ▲ Improved | 4% | 0% |
Focus within chosen field(s). Identify specific sub-directions and present ranked candidates.
| SOP | Purpose | Execution | |-----|---------|-----------| | broad-paper-search | Scan papers in the chosen field(s) | import: literature-overview | | deep-web-search | Deep reading of web resources in the field | subagent | | present-candidates | Present ranked sub-directions to user | dialogue |
present-candidates depth scales by start mode:broad-paper-search: at least 80 papers scanneddeep-web-search: at least 30 web pages read in fullRankedCandidates[] + user's selection
<!-- BEGIN available-tables (generated) -->
Optional, no fixed order; the final leaf is always a sop.
| SOP | When to use | | --- | --- | | deep-web-search | Full-page web reading for non-academic perspectives — blogs, tech reports, product pages, industry analysis. Spawns a subagent to read pages in isolated context. Hard constraint: at least 30 web pages read in full. | | north-star-crystallization-broad-paper-search | Paper landscape scan within selected field(s). Strict import of literature-engine/literature-overview skill. Hard constraint: at least 80 papers scanned. | | present-candidates | Analyze sub-directions within the user's chosen field and present ranked candidates. Combines sub-direction identification, skill-gap matching, and presentation into a single SOP. Depth scales by start mode: cold-start shows broad sub-directions, warm-start shows specific sub-problems, hot-start shows granular technical details. |
<!-- 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-07 | fail→fail | 19,587 | 25,899 | +32% | 1 | 1 | 0% | 2,760 | 4,373 | +58% | 0 | 0 | — |
case-06 | fail→pass | 21,958 | 16,550 | -25% | 1 | 1 | 0% | 3,341 | 2,960 | -11% | 0 | 0 | — |
case-01 | fail→fail | 21,810 | 10,009 | -54% | 1 | 1 | 0% | 3,368 | 1,083 | -68% | 0 | 0 | — |
case-02 | fail→pass | 24,818 | 45,647 | +84% | 1 | 1 | 0% | 3,660 | 7,135 | +95% | 0 | 0 | — |
case-03 | fail→pass | 25,725 | 34,922 | +36% | 1 | 1 | 0% | 3,771 | 5,300 | +41% | 0 | 0 | — |
case-04 | fail→pass | 20,062 | 39,268 | +96% | 1 | 1 | 0% | 2,985 | 6,640 | +122% | 0 | 0 | — |
case-05 | fail→pass | 20,590 | 25,502 | +24% | 1 | 1 | 0% | 3,441 | 3,581 | +4% | 0 | 0 | — |
case-08 | pass→fail | 18,692 | 29,388 | +57% | 1 | 1 | 0% | 2,841 | 5,003 | +76% | 0 | 0 | — |
case-09 | pass→pass | 19,704 | 21,089 | +7% | 1 | 1 | 0% | 2,812 | 3,740 | +33% | 0 | 0 | — |
case-10 | fail→fail | 21,743 | 7,976 | -63% | 1 | 1 | 0% | 3,376 | 862 | -74% | 0 | 0 | — |
case-11 | fail→fail | 19,043 | 18,766 | -1% | 1 | 1 | 0% | 2,755 | 3,379 | +23% | 0 | 0 | — |
case-12 | fail→pass | 23,020 | 24,694 | +7% | 1 | 1 | 0% | 3,625 | 4,281 | +18% | 0 | 0 | — |
case-13 | fail→pass | 24,810 | 39,840 | +61% | 1 | 1 | 0% | 4,196 | 6,676 | +59% | 0 | 0 | — |
case-14 | fail→fail | 25,346 | 8,315 | -67% | 1 | 1 | 0% | 3,765 | 896 | -76% | 0 | 0 | — |
case-15 | fail→pass | 26,641 | 19,947 | -25% | 1 | 1 | 0% | 3,840 | 3,410 | -11% | 0 | 0 | — |
case-16 | fail→fail | 25,313 | 3,759 | -85% | 1 | 1 | 0% | 3,971 | 1,012 | -75% | 0 | 0 | — |
case-17 | fail→pass | 25,177 | 32,824 | +30% | 1 | 1 | 0% | 3,784 | 4,535 | +20% | 0 | 0 | — |
case-18 | pass→pass | 20,732 | 26,294 | +27% | 1 | 1 | 0% | 3,048 | 4,590 | +51% | 0 | 0 | — |
case-19 | fail→fail | 35,900 | 7,459 | -79% | 1 | 1 | 0% | 5,642 | 1,035 | -82% | 0 | 0 | — |
case-20 | fail→fail | 28,201 | 27,953 | -1% | 1 | 1 | 0% | 4,675 | 4,580 | -2% | 0 | 0 | — |
case-21 | pass→fail | 34,835 | 28,575 | -18% | 1 | 1 | 0% | 6,190 | 5,599 | -10% | 0 | 0 | — |
case-22 | pass→fail | 40,075 | 38,169 | -5% | 1 | 1 | 0% | 6,187 | 6,673 | +8% | 0 | 0 | — |
case-23 | pass→pass | 34,032 | 38,294 | +13% | 1 | 1 | 0% | 5,640 | 6,663 | +18% | 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 19 counted toward the lift figure. The other 4 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 +26 percentage points is the difference between those two pass rates over the 19 comparable cases. 4 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.