RobustSGPO: Search-Space Control for Agent Harness Evolution
Zibo Zhao, Jijun Shi, Ruiming Tang, Wenwu Ou and Kun Gai (Wuhan University and Kuaishou Technology) add explicit control over edit scope and restart point to semantic-gradient prompt optimization for agent harnesses, and measure it over 7,350 candidate attempts.
Ask this paper
What the base method leaves unspecified: Semantic-gradient prompt optimization improves harnesses from execution feedback, but its local update rule does not decide how large an edit to request or which operation to apply.
Three additions: RobustSGPO specifies the requested edit, constructs and checks the patch before accepting it, and continues search from either the incumbent or a retained snapshot rather than always from the current best.
Permission scheduling beats a fixed ceiling: Periodic 1 to 2 to 3 edit-permission scheduling exceeds fixed maximum permission by 0.28 test-score points.
Held-out results: Completion on 30 held-out tasks rises from 60.0% to 80.0% and test quality from 3.77 to 4.14 under a 20-million-token budget.
Retention is a real tradeoff: Category retention reduces source-task degradation after a task-family shift, while random retention reaches a higher destination endpoint, and retention carries measurable overhead. Evaluated in the AgentX brainstorming workflow over 120 tasks and 95 runs.
Abstract
Semantic-gradient-based prompt optimization (SGPO) improves agent harnesses using execution feedback, but its local update rule leaves the choice of edit scope and operation unresolved. We introduce RobustSGPO, which specifies the requested edit, constructs and checks the patch, and continues search from either the incumbent or retained snapshots. We evaluate permission scheduling, cumulative controls, and task-family transfer in the AgentX brainstorming workflow using 120 tasks, 95 runs, and 7,350 candidate attempts. Periodic $1\to2\to3$ scheduling exceeds fixed maximum permission by 0.28 test-score points. RobustSGPO increases completion on 30 held-out tasks from 60.0% to 80.0% and improves test quality from 3.77 to 4.14 under a 20-million-token budget. Category retention reduces source-task degradation after a shift, whereas random retention reaches a higher destination endpoint. Search-space control benefits quality through executable edits and alternative starting points, with measurable retention overhead.