Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →majiayu000/agentic-systems-design
Use when designing tool-use agents, multi-step reasoning systems, planning loops, multi-agent architectures, or agent evaluation frameworks.
Passed 23 of 25 test cases with the skill, and 18 of 25 without it.
and 4 more in the full report
5 fixed18 passed either way2 still failing0 broke
Tokens, turns and time describe resource use, not a dollar price.
One of 5 test cases the skill fixed. Same model, same prompt, with and without the skill.
A developer needs to build an assistant that answers simple operational questions using 2 REST API endpoints in a direct linear sequence. Which agent architecture should be selected?
| Checked | Without skill | With skill |
|---|---|---|
| Selects a single-agent ReAct architecture. | Missed | Passed |
The instructions your model reads when the skill loads.
| Architecture | Use When | Complexity | Latency |
|---|---|---|---|
| Single-agent ReAct | 1-5 tools, linear reasoning | Low | Low |
| Plan-and-execute | Multi-step tasks needing upfront planning | Medium | Medium |
| Tree-of-Thought | Tasks with branching solutions, math/logic | Medium | High |
| LATS (Language Agent Tree Search) | Complex search + evaluation loops | High | Very High |
| Multi-agent supervisor | Specialized sub-tasks, delegation | High | Medium |
| Multi-agent debate | Tasks needing verification, fact-checking | High | High |
| Multi-agent chain | Sequential pipeline, each agent transforms output | Medium | Medium-High |
Decision rule: Start with single-agent ReAct. Escalate to plan-and-execute if the agent frequently fails mid-task. Use multi-agent only when a single model cannot hold all required expertise in context.
The default pattern. Model alternates between reasoning (think) and acting (tool call).
import anthropic
client = anthropic.Anthropic()
def react_agent(question: str, tools: list[dict], max_steps: int = 10) -> str:
system = """You are a helpful agent. For each step:
1. Think about what you need to do next
2. Use a tool if needed
3. When you have enough information, provide the final answer"""
messages = [{"role": "user", "content": question}]
for step in range(max_steps):
response = client.messages.create(
model="claude-sonnet-4-5-20250929",
max_tokens=4096,
system=system,
test cases passed, vs 18 / 25 without
More skills from the same source and nearby in the registry, with their headline benchmark lift.