Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Orchestrate multi-phase deep research with web search, memory retrieval, pattern matching, and synthesis into structured findings
.claude/skills/ruvnet-deep-research/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-04 | ✗→✓ | ▲ Improved | -3% | 0% |
| case-07 | ✗→✓ | ▲ Improved | 658% | 0% |
| case-12 | ✗→✓ | ▲ Improved | 60% | 0% |
| case-14 | ✗→✓ | ▲ Improved | -59% | 0% |
| case-15 | ✗→✓ | ▲ Improved | -40% | 0% |
Orchestrate multi-phase deep research campaigns that gather, cross-reference, and synthesize information from multiple sources.
When you need to investigate a complex topic thoroughly — spanning web sources, codebase patterns, stored memory, and external documentation — and produce a structured synthesis.
mcp__plugin_ruflo-core_ruflo__memory_search_unified and mcp__plugin_ruflo-core_ruflo__agentdb_pattern-search to check what's already knownWebSearch and WebFetch to gather external information for each sub-questionBash (grep/find), Read to examine relevant source filesmcp__plugin_ruflo-core_ruflo__memory_store with namespace research for each key findingmcp__plugin_ruflo-core_ruflo__agentdb_pattern-store for reusable patterns discoveredresearch — raw findings keyed by topicresearch-synthesis — completed synthesis reportsresearch-sources — source URLs and references| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→fail | 38,004 | 9,018 | -76% | 1 | 1 | 0% | 6,235 | 1,320 | -79% | 0 | 0 | — |
case-02 | fail→fail | 34,481 | 12,602 | -63% | 1 | 1 | 0% | 6,235 | 1,406 | -77% | 0 | 0 | — |
case-03 | fail→fail | 20,323 | 7,932 | -61% | 1 | 1 | 0% | 3,459 | 1,029 | -70% | 0 | 0 | — |
case-04 | fail→pass | 10,251 | 7,763 | -24% | 1 | 1 | 0% | 1,880 | 1,820 | -3% | 0 | 0 | — |
case-05 | fail→fail | 3,068 | 5,017 | +64% | 1 | 1 | 0% | 422 | 727 | +72% | 0 | 0 | — |
case-06 | fail→fail | 7,213 | 6,403 | -11% | 1 | 1 | 0% | 1,133 | 943 | -17% | 0 | 0 | — |
case-07 | fail→pass | 3,946 | 33,200 | +741% | 1 | 1 | 0% | 407 | 3,085 | +658% | 0 | 0 | — |
case-08 | fail→fail | 6,920 | 9,673 | +40% | 1 | 1 | 0% | 685 | 993 | +45% | 0 | 0 | — |
case-09 | fail→fail | 12,293 | 5,112 | -58% | 1 | 1 | 0% | 2,028 | 848 | -58% | 0 | 0 | — |
case-10 | fail→fail | 13,640 | 12,949 | -5% | 1 | 1 | 0% | 2,355 | 2,757 | +17% | 0 | 0 | — |
case-11 | pass→fail | 16,774 | 5,917 | -65% | 1 | 1 | 0% | 2,572 | 922 | -64% | 0 | 0 | — |
case-12 | fail→pass | 8,051 | 9,625 | +20% | 1 | 1 | 0% | 1,420 | 2,277 | +60% | 0 | 0 | — |
case-13 | pass→pass | 10,956 | 9,924 | -9% | 1 | 1 | 0% | 1,868 | 2,306 | +23% | 0 | 0 | — |
case-14 | fail→pass | 12,980 | 3,614 | -72% | 1 | 1 | 0% | 2,218 | 916 | -59% | 0 | 0 | — |
case-15 | fail→pass | 14,731 | 5,766 | -61% | 1 | 1 | 0% | 2,382 | 1,421 | -40% | 0 | 0 | — |
case-16 | fail→pass | 37,421 | 30,621 | -18% | 1 | 1 | 0% | 6,164 | 5,463 | -11% | 0 | 0 | — |
case-17 | pass→fail | 16,952 | 9,059 | -47% | 1 | 1 | 0% | 2,932 | 958 | -67% | 0 | 0 | — |
case-18 | fail→pass | 16,563 | 5,106 | -69% | 1 | 1 | 0% | 826 | 1,459 | +77% | 0 | 0 | — |
case-19 | pass→pass | 13,350 | 5,057 | -62% | 1 | 1 | 0% | 2,413 | 1,344 | -44% | 0 | 0 | — |
case-20 | fail→pass | 10,442 | 4,757 | -54% | 1 | 1 | 0% | 1,662 | 1,273 | -23% | 0 | 0 | — |
case-21 | fail→fail | 6,292 | 3,727 | -41% | 1 | 1 | 0% | 886 | 959 | +8% | 0 | 0 | — |
case-22 | pass→pass | 2,070 | 8,363 | +304% | 1 | 1 | 0% | 319 | 1,561 | +389% | 0 | 0 | — |
case-23 | pass→pass | 1,508 | 2,458 | +63% | 1 | 1 | 0% | 249 | 804 | +223% | 0 | 0 | — |
case-24 | pass→pass | 4,978 | 4,221 | -15% | 1 | 1 | 0% | 971 | 1,235 | +27% | 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. 24 cases were attempted, and 14 counted toward the lift figure. The other 10 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 +25 percentage points is the difference between those two pass rates over the 14 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.