Search papers, labs, and topics across Lattice.
The paper introduces task-conditioned tool-output pruning to reduce the input token length for coding agents by extracting only the relevant evidence block from tool outputs. They construct a benchmark dataset from SWE-bench and synthetic data to train a LoRA-fine-tuned Qwen 3.5 2B model. The resulting model achieves 0.86 recall and 0.80 F1 while pruning 92% of input tokens, significantly outperforming larger zero-shot models and heuristic baselines.
Coding agents can achieve higher recall and F1 scores by selectively attending to just 8% of tool output tokens, rather than processing the entire observation.
Coding agents repeatedly consume long tool observations even though only a small fraction of each observation matters for the next step. We study task-conditioned tool-output pruning: given a focused query and one tool output, return the smallest verbatim evidence block the agent should inspect next. We introduce a benchmark of 11,477 examples built from SWE-bench repository interactions and synthetic multi-ecosystem tool outputs, with a manually curated 618-example test set. We fine-tune Qwen 3.5 2B with LoRA and compare it against larger zero-shot models and heuristic pruning baselines. Our model reaches 0.86 recall and 0.80 F1 while removing 92% of input tokens, outperforming zero-shot Qwen 3.5 35B A3B by 11 recall points and all heuristic baselines by a wide margin.
