Search papers, labs, and topics across Lattice.
To address the instability and unconstrained mutation scopes that plague semantic-gradient prompt optimization (SGPO) in multi-step agent scaffolds, RobustSGPO introduces explicit search-space controls via automated patch verification, snapshot retention, and dynamic permission scheduling. Controlling the granularity of harness edits and enabling branching from retained checkpoints prevents optimization collapse and mitigates catastrophic forgetting across task transfers. Evaluated across 7,350 candidate iterations in the AgentX workflow under a 20-million-token budget, this approach boosted held-out task completion from 60.0% to 80.0% while periodic permission scheduling outperformed fixed-permission baselines by 0.28 points.
Giving agent harness optimizers unconstrained edit scope reliably degrades performance; enforcing graduated, periodic permission scheduling ($1\to2\to3$) instead yields an immediate 20% jump in held-out task completion.
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.
