Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →CrewAI — multi-agent AI framework. Role-based agents with defined goals, tools, and memory. Hierarchical and sequential task execution. Human input delegation and process orchestration.
.claude/skills/mkurman-crewai/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-09 | ✓→✗ | ▼ Worse | -16% | 0% |
| case-02 | ✓→✓ | ＝ Same ✓ | -30% | 0% |
| case-03 | ✓→✓ | ＝ Same ✓ | -10% | 0% |
| case-04 | ✓→✓ | ＝ Same ✓ | 41% | 0% |
| case-05 | ✓→✓ | ＝ Same ✓ | 4% | 0% |
CrewAI enables role-based multi-agent AI systems. Agents have defined goals, tools, backstories, and memory. Tasks are assigned to specific agents with expected outputs. Supports sequential and hierarchical execution.
bashuv pip install crewai
pythonfrom crewai import Agent, Task, Crew researcher = Agent( role="Research Analyst", goal="Find latest developments in AI agents", backstory="Expert at finding relevant information", ) writer = Agent( role="Technical Writer", goal="Write clear summary of findings", backstory="Skilled at explaining technical topics", ) task1 = Task(description="Search for latest AI agent frameworks in 2025", expected_output="List of frameworks with key features", agent=researcher) task2 = Task(description="Write a 3-paragraph summary", expected_output="Markdown report", agent=writer) crew = Crew(agents=[researcher, writer], tasks=[task1, task2]) result = crew.kickoff() print(result)
pythonfrom crewai_tools import SerperDevTool researcher = Agent( role="Research Analyst", tools=[SerperDevTool()], )
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→fail | 8,261 | 8,038 | -3% | 1 | 1 | 0% | 1,681 | 1,963 | +17% | 0 | 0 | — |
case-02 | pass→pass | 5,148 | 1,932 | -62% | 1 | 1 | 0% | 848 | 590 | -30% | 0 | 0 | — |
case-03 | pass→pass | 3,717 | 1,538 | -59% | 1 | 1 | 0% | 585 | 524 | -10% | 0 | 0 | — |
case-04 | pass→pass | 3,112 | 2,829 | -9% | 1 | 1 | 0% | 484 | 684 | +41% | 0 | 0 | — |
case-05 | pass→pass | 4,166 | 2,583 | -38% | 1 | 1 | 0% | 687 | 716 | +4% | 0 | 0 | — |
case-06 | pass→pass | 4,940 | 2,125 | -57% | 1 | 1 | 0% | 887 | 684 | -23% | 0 | 0 | — |
case-07 | pass→pass | 4,477 | 1,846 | -59% | 1 | 1 | 0% | 751 | 602 | -20% | 0 | 0 | — |
case-08 | pass→pass | 4,404 | 2,054 | -53% | 1 | 1 | 0% | 628 | 627 | -0% | 0 | 0 | — |
case-09 | pass→fail | 4,514 | 2,190 | -51% | 1 | 1 | 0% | 748 | 632 | -16% | 0 | 0 | — |
case-10 | pass→pass | 4,783 | 1,996 | -58% | 1 | 1 | 0% | 767 | 614 | -20% | 0 | 0 | — |
case-11 | pass→pass | 5,065 | 4,638 | -8% | 1 | 1 | 0% | 819 | 1,074 | +31% | 0 | 0 | — |
case-12 | pass→pass | 3,297 | 1,834 | -44% | 1 | 1 | 0% | 525 | 606 | +15% | 0 | 0 | — |
case-13 | pass→pass | 2,006 | 1,822 | -9% | 1 | 1 | 0% | 285 | 574 | +101% | 0 | 0 | — |
case-14 | pass→pass | 2,217 | 1,742 | -21% | 1 | 1 | 0% | 327 | 582 | +78% | 0 | 0 | — |
case-15 | pass→pass | 2,524 | 1,987 | -21% | 1 | 1 | 0% | 349 | 581 | +66% | 0 | 0 | — |
case-16 | pass→pass | 3,928 | 1,448 | -63% | 1 | 1 | 0% | 616 | 538 | -13% | 0 | 0 | — |
case-17 | fail→fail | 5,403 | 3,489 | -35% | 1 | 1 | 0% | 1,038 | 908 | -13% | 0 | 0 | — |
case-18 | pass→pass | 8,707 | 4,938 | -43% | 1 | 1 | 0% | 1,595 | 1,103 | -31% | 0 | 0 | — |
case-19 | pass→pass | 14,363 | 9,925 | -31% | 1 | 1 | 0% | 2,315 | 2,104 | -9% | 0 | 0 | — |
case-20 | pass→pass | 7,595 | 7,221 | -5% | 1 | 1 | 0% | 1,536 | 1,772 | +15% | 0 | 0 | — |
case-21 | fail→fail | 9,971 | 7,013 | -30% | 1 | 1 | 0% | 1,876 | 1,649 | -12% | 0 | 0 | — |
case-22 | pass→pass | 10,308 | 6,572 | -36% | 1 | 1 | 0% | 1,990 | 1,516 | -24% | 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. The headline lift of -100 percentage points is the difference between those two pass rates over the 22 comparable cases. 1 case got worse with the skill loaded, and it is included in that figure.
Other measured skills in the registry, with their headline benchmark lift.