Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Parallel external documentation lookup. Spawn multiple @scroll-docs agents simultaneously to fetch SDK refs, API docs, and framework guides for different topics in one pass.
.claude/skills/evolution-foundation-dev-external-context/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-08 | ✗→✓ | ▲ Improved | -10% | 0% |
| case-12 | ✗→✓ | ▲ Improved | 4% | 0% |
| case-16 | ✗→✓ | ▲ Improved | 27% | 0% |
| case-17 | ✗→✓ | ▲ Improved | -6% | 0% |
| case-20 | ✗→✓ | ▲ Improved | -47% | 0% |
Derived from oh-my-claudecode (MIT, Yeachan Heo). Adapted for the EvoNexus Engineering Layer.
Parallel external research. When a task needs multiple unrelated external lookups (SDK A + Framework B + Standard C), spawn @scroll-docs instances in parallel instead of serial.
@scroll-docs directly@scout-explorer instead@scroll-docs agents in parallel — one per topicworkspace/development/research/[C]external-context-{topic}-{date}.mdmarkdown## External Context — {topic} ### Lookup 1: {topic A} [from @scroll-docs] ### Lookup 2: {topic B} [from @scroll-docs] ### Lookup 3: {topic C} [from @scroll-docs] ## Synthesis [Combined understanding] ## Conflicts [Where sources disagree] ## Recommendation [Next step based on combined research]
@scroll-docs (parallel workers)@compass-planner (consumer of synthesized research)@apex-architect (consumer for architecture decisions)| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | pass→fail | 18,141 | 5,604 | -69% | 1 | 1 | 0% | 3,136 | 1,433 | -54% | 0 | 0 | — |
case-02 | fail→fail | 18,141 | 24,425 | +35% | 1 | 1 | 0% | 2,940 | 3,796 | +29% | 0 | 0 | — |
case-03 | pass→pass | 32,190 | 21,941 | -32% | 1 | 1 | 0% | 5,929 | 4,239 | -29% | 0 | 0 | — |
case-04 | fail→fail | 10,110 | 19,755 | +95% | 1 | 1 | 0% | 1,625 | 3,633 | +124% | 0 | 0 | — |
case-05 | fail→fail | 3,965 | 3,562 | -10% | 1 | 1 | 0% | 579 | 974 | +68% | 0 | 0 | — |
case-06 | fail→fail | 3,981 | 3,842 | -3% | 1 | 1 | 0% | 611 | 1,052 | +72% | 0 | 0 | — |
case-07 | fail→fail | 6,136 | 9,479 | +54% | 1 | 1 | 0% | 350 | 912 | +161% | 0 | 0 | — |
case-08 | fail→pass | 22,699 | 16,689 | -26% | 1 | 1 | 0% | 3,331 | 3,004 | -10% | 0 | 0 | — |
case-09 | pass→pass | 16,406 | 23,246 | +42% | 1 | 1 | 0% | 3,023 | 4,468 | +48% | 0 | 0 | — |
case-10 | pass→pass | 17,889 | 15,711 | -12% | 1 | 1 | 0% | 2,882 | 3,079 | +7% | 0 | 0 | — |
case-11 | pass→fail | 25,686 | 9,049 | -65% | 1 | 1 | 0% | 4,205 | 896 | -79% | 0 | 0 | — |
case-12 | fail→pass | 19,745 | 17,333 | -12% | 1 | 1 | 0% | 3,282 | 3,398 | +4% | 0 | 0 | — |
case-13 | fail→fail | 25,021 | 9,143 | -63% | 1 | 1 | 0% | 4,276 | 996 | -77% | 0 | 0 | — |
case-14 | pass→pass | 35,331 | 16,182 | -54% | 1 | 1 | 0% | 6,198 | 3,233 | -48% | 0 | 0 | — |
case-15 | fail→fail | 22,900 | 24,303 | +6% | 1 | 1 | 0% | 3,846 | 4,347 | +13% | 0 | 0 | — |
case-16 | fail→pass | 15,808 | 18,000 | +14% | 1 | 1 | 0% | 2,852 | 3,615 | +27% | 0 | 0 | — |
case-17 | fail→pass | 24,176 | 19,746 | -18% | 1 | 1 | 0% | 3,865 | 3,649 | -6% | 0 | 0 | — |
case-18 | pass→pass | 25,082 | 18,999 | -24% | 1 | 1 | 0% | 4,674 | 3,297 | -29% | 0 | 0 | — |
case-19 | pass→pass | 28,867 | 29,673 | +3% | 1 | 1 | 0% | 4,559 | 4,221 | -7% | 0 | 0 | — |
case-20 | fail→pass | 38,373 | 17,532 | -54% | 1 | 1 | 0% | 6,185 | 3,297 | -47% | 0 | 0 | — |
case-21 | fail→fail | 28,620 | 7,714 | -73% | 1 | 1 | 0% | 4,161 | 855 | -79% | 0 | 0 | — |
case-22 | fail→fail | 40,096 | 5,370 | -87% | 1 | 1 | 0% | 6,185 | 967 | -84% | 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 17 counted toward the lift figure. The other 5 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 17 comparable cases. 3 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.