ReAct Agent Optimisation — Fewer Steps, Better Reasoning

I will optimise your existing ReAct agent — improving reasoning quality through system prompt refinement, reducing average steps-to-completion through better tool design, and improving error recovery through reasoning loop modifications. ReAct agents that are deployed without optimisation consistently take more steps than necessary, select suboptimal tools when multiple tools could serve a purpose, and fail to recover effectively from tool errors — each of these failure modes is addressable through targeted system prompt changes, tool description improvement, and reasoning loop modification.

The optimisation covers reasoning quality analysis on 30+ agent runs, system prompt refinement testing, tool description improvement for ambiguous selections, error recovery prompt improvement, step reduction optimisation, and a before/after performance comparison.