Advanced · Expert · 6 rounds · ~8 min · free with an account
Building Simple Agents (ReAct)
Thought → Action → Observation — the foundation of every modern agent, with halt conditions that don't burn $500.
Prompt Arcade · 6 rounds · free with an account
Building Simple Agents (ReAct)
An agent loops: think, act with a tool, observe the result, repeat — until a clear stop condition. Good agents have narrow tools, explicit stop rules and a budget.
- Order it
- Spot it
- Build it
- Match it
- Spot it
- Build it
Five games are open to everyone — start with Prompting Mindset 101. The other 31 need a free account (no card) so your scores and progress are saved.
What you'll learn
- The ReAct loop: Think → act → observe → decide. Repeat until the stop rule fires.
- Agent design flaws: Narrow tools, a budget, confirmation for irreversible steps, a clear stop.
- Stop conditions: Define success, a budget, and what to do when stuck.
- Observations: Read each observation and decide: retry differently, stop, or escalate.
- Tool descriptions: Agents choose tools by their descriptions. Be precise.
- Human in the loop: Automate the reversible; confirm the irreversible.
More advanced games
Self-Consistency VotingRun the same prompt 3-5 times → take the majority answer. +12-18% accuracy on reasoning tasks (Wang et al. 2022).Few-Shot Pattern RecognitionShow 2-4 examples → the model maps the relational structure, not the surface details (Chase & Simon's chess masters, applied).Constraint CalibrationToo few constraints → drift. Too many → brittle. The 5-constraint sweet spot and how to find it.