CAPO: Constraint-Aware Prompt Optimization for LLM Agents

2026-08-17

AI Summary

This study addresses the challenge of satisfying multiple operational constraints for LLM agents in the absence of domain-specific supervised data. We propose CAPO and DCAPO, frameworks grounded in primal-dual algorithms that integrate pooled rewriting with adaptive weighting mechanisms. By training a feedback-conditioned rewriter to automatically optimize system prompts, our approach eliminates the need for post-training task models. Experimental results demonstrate that this method reliably converges to feasible operating points in agent benchmarks, significantly enhancing task accuracy while ensuring strict adherence to generation constraints. Ultimately, this work achieves adaptive optimization of multi-constraint prompts in unsupervised scenarios, offering a robust solution for aligning agent behaviors with complex operational requirements without reliance on labeled supervision.

Abstract

Large language models (LLMs) are increasingly deployed as agents that rely on system prompts to use tools and complete tasks. Such deployments impose distinct operational requirements, including appropriate tool use, concise prompts and solution paths, and compliance with safety and formatting policies. For many practitioners, however, assembling domain-specific supervised data to post-train models to meet these requirements is infeasible. We introduce CAPO (Constraint-Aware Prompt Optimization), a primal-dual method that combines pool-based rewrites with adaptive constraint weighting to optimize system prompts under explicit operational constraints. Across agentic benchmarks, CAPO more reliably reaches empirically feasible operating points while improving task performance. CAPO also generalizes beyond agentic settings, achieving strong results on assistant-style evaluations with output-format and safety/privacy constraints. We further introduce DCAPO (Dynamically Trained CAPO), which trains a feedback- and dual-conditioned rewriter with pool-based GRPO while keeping the task agent frozen. Across task agents of different sizes, DCAPO produces a feasible prompt in every evaluated domain and matches or improves the task accuracy achieved by the evaluated baselines. A surrogate analysis characterizes how finite-pool and discrete-rewrite errors enter the inexact primal-dual procedure.

Problem

• LLM Agents
• Prompt Optimization
• Operational Constraints
• System Prompts

Innovation

• Constraint-Aware Prompt Optimization
• Primal-Dual Method
• Pool-Based Rewrites
• Adaptive Constraint Weighting
• DCAPO