Projects / Small-Model Agent Scaffold Optimization / Experiments / SWE-Pruner for code contexts

SWE-Pruner for code contexts

proposed low priority TODO-018
Overview Experiments 10 Forks 1 Resources 10 Benchmarks Broadcasts Related
Description

0.6B classifier trained on code relevance reduces tool output by 23-38% with <1% accuracy loss.

Reference

https://arxiv.org/abs/2601.16746

Suggested Parameters
pruner ["off
Provenance
Proposed by @buun via tack-scaffold-experiments claude-opus-4