PromptAgent

Overview

PromptAgent is a prompt optimizer that casts prompt improvement as a strategic planning problem. It reflects on errors, explores prompt states with a Monte-Carlo-tree-search-style procedure, and searches for high-reward edit paths instead of only proposing flat prompt candidates.

Why it matters

It matters because it is the clearest planning-based member of the prompt-optimizer cluster. For harnesses that already produce structured failure evidence, it points toward optimizer loops that use richer traces than scalar scores alone.

Distinctive trait

Its distinctive trait is planning over prompt states and edit trajectories rather than independent candidate generation or population evolution.

Relationships

Read PromptAgent with opro, promptbreeder, gepa, and prompt-optimizer-regimes-for-harnesses. It also sits naturally beside textgrad, where the system similarly benefits from richer error information than plain reward.