HarnessMM: Recursive Self-Improvement for Harness Token Efficiency
Yuhua Jiang, Chang Xu, Feifei Gao and Li Zhao (Tsinghua University, Microsoft Research Asia) present HarnessMM, which cuts harness token cost by reusing decisions that earlier tool feedback has already fixed instead of calling the model again.
Ask this paper
Decision reuse. Executable Decision Graphs store exact actions, literal arguments, causal dependencies and guards. A controller at the model-tool boundary reuses a resolved decision when its guard holds and calls the original actor for anything new.
Self-improvement loop. Recursive self-improvement refines companion harness configurations from training feedback while actor weights stay fixed; validation selects the deployed configuration.
Codex result. On HumanEval, Codex keeps 90.82% success while tokens per task fall from 66.72k to 47.02k, a 29.54% saving.
DeepSeek result. On the DeepSeek harness, success rises from 71.43% to 85.71% while tokens per task fall from 96.28k to 11.62k, an 87.93% saving. An MBPP ablation shows 26.91% lower cost than single-use graphs.
Abstract
Agent harnesses repeatedly invoke reasoning after tool feedback, even when an earlier call has already fixed the next action. We introduce HarnessMM (Harness Minus Minus), a method for improving harness token efficiency through guarded decision reuse and recursive self-improvement (RSI). RSI iteratively refines companion harness configurations from training feedback, with validation selecting the configuration for deployment. A controller at the model-tool boundary reuses resolved decisions to replace later reasoning callbacks while actor weights remain fixed. Executable Decision Graphs store exact actions, literal arguments, causal dependencies, and public guards. Guarded Decision Reuse activates these persistent decisions from observed receipts; Certified Callback Substitution executes them through the native interface and records removal of the registered slow callback. New semantic decisions, unsupported evidence, and selector abstention invoke the original actor again. The representation restricts what feedback may change, while complete-task measurements establish quality and cost. On HumanEval, HarnessMM preserves Codex success at 90.82\% while reducing Tokens/task (k) from 66.72 to 47.02 (29.54\% saving). For DeepSeek Harness, success rises from 71.43\% to 85.71\% while Tokens/task (k) fall from 96.28 to 11.62 (87.93\% saving). A separate MBPP ablation measures 26.91\% lower cost than single-use graphs and 19.52\% lower cost than retained callbacks. These results establish guarded decision reuse as a practical way to reduce repeated reasoning in existing harnesses.