Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →OpenAI Agents SDK — build agentic workflows with handoffs, guardrails, and tool integration. Single-agent to multi-agent orchestration. Tracing and observability. Python-first SDK from OpenAI.
.claude/skills/mkurman-openai-agents-sdk/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-02 | ✗→✓ | ▲ Improved | -15% | 0% |
| case-06 | ✗→✓ | ▲ Improved | -57% | 0% |
| case-07 | ✗→✓ | ▲ Improved | -5% | 0% |
| case-10 | ✗→✓ | ▲ Improved | -52% | 0% |
| case-11 | ✗→✓ | ▲ Improved | -39% | 0% |
OpenAI Agents SDK provides a lightweight framework for building agentic AI workflows. Supports tool use, handoffs (agent-to-agent routing), guardrails, and tracing — built on the OpenAI API's native agent primitives.
bashuv pip install openai-agents
pythonfrom agents import Agent, Runner agent = Agent(name="Helper", instructions="You are a helpful assistant.") result = Runner.run_sync(agent, "What is the capital of France?") print(result.final_output)
pythonfrom agents import Agent, Runner, function_tool @function_tool def get_weather(city: str) -> str: return f"Weather in {city}: sunny, 22°C" agent = Agent(name="WeatherBot", instructions="Use tools to answer weather queries.", tools=[get_weather]) result = Runner.run_sync(agent, "What's the weather in London?") print(result.final_output)
pythonfrom agents import Agent, Runner triage = Agent(name="Triage", instructions="Route to the right specialist.") billing = Agent(name="Billing", instructions="Handle billing questions.") triage.handoffs = [billing] result = Runner.run_sync(triage, "My invoice didn't arrive")
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | pass→pass | 6,288 | 2,829 | -55% | 1 | 1 | 0% | 1,158 | 891 | -23% | 0 | 0 | — |
case-02 | fail→pass | 13,779 | 9,131 | -34% | 1 | 1 | 0% | 2,444 | 2,085 | -15% | 0 | 0 | — |
case-03 | fail→fail | 12,438 | 7,099 | -43% | 1 | 1 | 0% | 2,305 | 1,799 | -22% | 0 | 0 | — |
case-04 | pass→pass | 8,362 | 6,548 | -22% | 1 | 1 | 0% | 1,745 | 1,703 | -2% | 0 | 0 | — |
case-05 | pass→pass | 14,142 | 7,994 | -43% | 1 | 1 | 0% | 2,886 | 2,032 | -30% | 0 | 0 | — |
case-06 | fail→pass | 6,615 | 1,410 | -79% | 1 | 1 | 0% | 1,171 | 503 | -57% | 0 | 0 | — |
case-07 | fail→pass | 4,056 | 1,688 | -58% | 1 | 1 | 0% | 689 | 655 | -5% | 0 | 0 | — |
case-08 | pass→pass | 10,352 | 2,998 | -71% | 1 | 1 | 0% | 1,933 | 880 | -54% | 0 | 0 | — |
case-09 | pass→pass | 14,767 | 1,801 | -88% | 1 | 1 | 0% | 2,618 | 623 | -76% | 0 | 0 | — |
case-10 | fail→pass | 11,234 | 3,214 | -71% | 1 | 1 | 0% | 1,969 | 945 | -52% | 0 | 0 | — |
case-11 | fail→pass | 10,830 | 4,414 | -59% | 1 | 1 | 0% | 2,000 | 1,220 | -39% | 0 | 0 | — |
case-12 | fail→pass | 8,833 | 2,104 | -76% | 1 | 1 | 0% | 1,493 | 698 | -53% | 0 | 0 | — |
case-13 | pass→pass | 27,877 | 2,045 | -93% | 1 | 1 | 0% | 5,038 | 621 | -88% | 0 | 0 | — |
case-14 | pass→pass | 7,417 | 1,828 | -75% | 1 | 1 | 0% | 1,162 | 613 | -47% | 0 | 0 | — |
case-15 | fail→fail | 5,564 | 3,977 | -29% | 1 | 1 | 0% | 986 | 1,102 | +12% | 0 | 0 | — |
case-16 | fail→pass | 9,382 | 4,569 | -51% | 1 | 1 | 0% | 1,717 | 1,172 | -32% | 0 | 0 | — |
case-17 | fail→pass | 14,556 | 5,184 | -64% | 1 | 1 | 0% | 2,646 | 1,330 | -50% | 0 | 0 | — |
case-18 | fail→pass | 10,087 | 1,368 | -86% | 1 | 1 | 0% | 1,614 | 523 | -68% | 0 | 0 | — |
case-19 | pass→pass | 14,180 | 8,289 | -42% | 1 | 1 | 0% | 2,532 | 1,849 | -27% | 0 | 0 | — |
case-20 | fail→pass | 8,520 | 2,325 | -73% | 1 | 1 | 0% | 1,471 | 674 | -54% | 0 | 0 | — |
case-21 | fail→pass | 10,895 | 3,965 | -64% | 1 | 1 | 0% | 2,014 | 1,130 | -44% | 0 | 0 | — |
case-22 | pass→pass | 7,034 | 2,116 | -70% | 1 | 1 | 0% | 1,095 | 683 | -38% | 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 +50 percentage points is the difference between those two pass rates over the 22 comparable cases.
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.