Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Go in depth harness — fan-out web searches, fetch sources, adversarially verify claims, synthesize a cited report.
.claude/skills/sickn33-go-in-depth/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-06 | ✗→✓ | ▲ Improved | -65% | 0% |
| case-01 | ✓→✗ | ▼ Worse | -86% | 0% |
| case-02 | ✓→✗ | ▼ Worse | -84% | 0% |
| case-03 | ✓→✗ | ▼ Worse | -85% | 0% |
| case-07 | ✓→✗ | ▼ Worse | 142% | 0% |
Go in depth harness — fan-out web searches, fetch sources, adversarially verify claims, synthesize a cited report. Run the "go-in-depth" workflow.
When the user wants a deep, multi-source, fact-checked research report on any topic. BEFORE invoking, check if the question is specific enough to research directly — if underspecified (e.g., "what car to buy" without budget/use-case/region), ask 2-3 clarifying questions to narrow scope. Then pass the refined question as args, weaving the answers in.
Phases:
Workflow({ name: "go-in-depth" })Workflow({ name: "go-in-depth", args: { query: "best hybrid cars under $30k in the US for families" } })Workflow({ name: "go-in-depth", args: { query: "how does the transformer architecture handle positional encoding?" } })Workflow({ name: "go-in-depth", args: { query: "efficacy of intermittent fasting for long-term weight loss in adults" } })scripts/workflow-script.js
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | pass→fail | 33,999 | 5,560 | -84% | 1 | 1 | 0% | 6,187 | 892 | -86% | 0 | 0 | — |
case-02 | pass→fail | 39,607 | 9,398 | -76% | 1 | 1 | 0% | 6,185 | 1,008 | -84% | 0 | 0 | — |
case-03 | pass→fail | 37,292 | 5,012 | -87% | 1 | 1 | 0% | 6,192 | 948 | -85% | 0 | 0 | — |
case-04 | pass→pass | 8,057 | 3,130 | -61% | 1 | 1 | 0% | 1,337 | 1,019 | -24% | 0 | 0 | — |
case-05 | pass→pass | 10,311 | 3,587 | -65% | 1 | 1 | 0% | 1,768 | 1,034 | -42% | 0 | 0 | — |
case-06 | fail→pass | 21,191 | 4,887 | -77% | 1 | 1 | 0% | 3,506 | 1,238 | -65% | 0 | 0 | — |
case-07 | pass→fail | 2,765 | 8,630 | +212% | 1 | 1 | 0% | 386 | 935 | +142% | 0 | 0 | — |
case-08 | pass→pass | 3,111 | 12,658 | +307% | 1 | 1 | 0% | 414 | 1,686 | +307% | 0 | 0 | — |
case-09 | pass→pass | 4,952 | 8,993 | +82% | 1 | 1 | 0% | 850 | 1,964 | +131% | 0 | 0 | — |
case-10 | pass→fail | 36,579 | 6,858 | -81% | 1 | 1 | 0% | 6,171 | 903 | -85% | 0 | 0 | — |
case-11 | pass→fail | 34,287 | 6,077 | -82% | 1 | 1 | 0% | 3,887 | 948 | -76% | 0 | 0 | — |
case-12 | fail→fail | 32,125 | 4,923 | -85% | 1 | 1 | 0% | 5,902 | 776 | -87% | 0 | 0 | — |
case-13 | pass→fail | 37,531 | 5,543 | -85% | 1 | 1 | 0% | 6,166 | 724 | -88% | 0 | 0 | — |
case-14 | pass→fail | 27,494 | 7,289 | -73% | 1 | 1 | 0% | 4,544 | 899 | -80% | 0 | 0 | — |
case-15 | pass→fail | 40,815 | 6,714 | -84% | 1 | 1 | 0% | 6,165 | 865 | -86% | 0 | 0 | — |
case-16 | fail→fail | 26,237 | 7,784 | -70% | 1 | 1 | 0% | 3,262 | 1,060 | -68% | 0 | 0 | — |
case-17 | pass→fail | 34,357 | 7,203 | -79% | 1 | 1 | 0% | 6,124 | 820 | -87% | 0 | 0 | — |
case-18 | pass→fail | 22,284 | 5,926 | -73% | 1 | 1 | 0% | 3,167 | 797 | -75% | 0 | 0 | — |
case-19 | pass→fail | 39,114 | 5,477 | -86% | 1 | 1 | 0% | 6,170 | 783 | -87% | 0 | 0 | — |
case-20 | fail→fail | 19,108 | 5,644 | -70% | 1 | 1 | 0% | 3,343 | 857 | -74% | 0 | 0 | — |
case-21 | pass→fail | 29,213 | 6,156 | -79% | 1 | 1 | 0% | 4,766 | 990 | -79% | 0 | 0 | — |
case-22 | pass→fail | 36,089 | 6,709 | -81% | 1 | 1 | 0% | 6,166 | 889 | -86% | 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 5 counted toward the lift figure. The other 17 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 -59 percentage points is the difference between those two pass rates over the 5 comparable cases. 14 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.