Skip to content

Arena designs

One markdown design per planned arena. When an arena is implemented it moves from "design" here to a runnable arenas/<id>/ directory (spec + dataset + mock script) and this file becomes its reference notes.

Arena Design Implemented
tool_use (shipped) ✅ arenas/tool_use/
boundary_response boundary_response.md 16 paired fixture cases; vanilla, LangGraph and OpenAI Agents only
structured_output structured_output.md ✅ arenas/structured_output/
resilience (built directly; see the arena.toml) ✅ arenas/resilience/
multi_agent multi_agent.md ✅ arenas/multi_agent/ — single-agent entries and five *_multi pipelines (structural, handoff, sub-agent-as-tool)
rag rag.md ✅ arenas/rag/
human_in_the_loop human_in_the_loop.md ✅ arenas/human_in_the_loop/ — six adapters pause, six distinct mechanisms
durable_state durable_state.md ✅ arenas/durable_state/ — five adapters 8/8 across a real process restart

See ../../CONTRIBUTING.md → "Adding an arena".