Grow the Harness, Not the Context: From Strategy-Free Scaffolds to Reusable Specialist Agents
Laizhen Li, Xitong Gao and colleagues at the Shenzhen Institutes of Advanced Technology (Chinese Academy of Sciences) propose Growing Harness, which learns an agent's control logic as executable harness code from task failures, so the model is called only for task-specific reasoning.
Ask this paper
Starting point. A strategy-free scaffold exposes fixed model and tool interfaces but contains no task-solving controller.
Failure-guided repair. Function-level execution traces localize each failure to a small code region, an optimizer repairs a window of failures together, and a held-out gate rolls back repairs that break earlier capability.
Results. Across BrowseComp-Plus and WebArena-Verified with deployment models from 4B to 120B, it has the highest mean success in five of six settings and is 0.7 points behind the best in the sixth.
Cost. Against a tool-calling agent it cuts LLM calls by 76.0% to 91.8% and inference cost by 74.4% to 98.6%.
Small models keep working. On WebArena-Verified its success stays between 44.7% and 45.3% across model sizes, while the tool-calling agent falls to 6.7% with the 4B model.
Abstract
Large language model (LLM) agents often handle streams of related tasks, yet standard harnesses repeatedly ask the model to reconstruct the same control decisions inside each task's context. We study whether task feedback can instead turn recurring control into reusable executable code, while reserving LLM calls for task-specific semantic reasoning. We introduce Growing Harness, a failure-guided training paradigm that learns the agent harness itself from a strategy-free scaffold that exposes fixed model and tool interfaces but encodes no task-solving controller. Function-level execution traces localize each failure to a bounded code surface, an optimizer repairs a window of failures jointly, and a success-first held-out gate rolls back repair sequences that harm prior capability. Accepted edits accumulate in one shared harness, allowing its control structure to emerge from task feedback. Across BrowseComp-Plus and WebArena-Verified with three deployment models from 4B to 120B parameters, Growing Harness achieves the highest mean success in five of six benchmark-model settings and trails the best mean by 0.7 pp. in the sixth. Relative to a Tool-Calling agent, it reduces LLM calls by 76.0-91.8% and deployed-agent inference cost by 74.4-98.6%. On WebArena-Verified, its success remains 44.7-45.3% across model scales, whereas Tool-Calling falls to 6.7% with the 4B model. Ablations show that trace-local edits, joint repair, and gate-based rollback each improve final success. These results show that persistent program growth can move recurring control out of model context and into low-cost code, yielding reusable specialist agents that remain effective with smaller deployment models.