Get Started
Home
Topics
Search
Library
Research questionHow can teams identify which design choices reliably improve repository-scale refactoring agents without breaking behavior?A single refactoring can require consistent edits across many interdependent files, so agent success may reflect several coupled design choices. Teams also need evidence that changes preserve program behavior and can be audited across runs.
AI
AI Agents
Code Generation & Program Synthesis
Evaluation & Benchmarks
Multi-agent Systems
Retrieval-Augmented Generation
Latest papersRecent research connected to this question, newest first.RefactorPlatform: An Open-Source Harness for Controlled Evaluation of Repository-Scale Refactoring AgentsThe evidence is RefactorPlatform, an open-source harness that fixes the environment and varies model backbone, retrieval, prompt specificity, and execution regime, including baseline, retrieval-augmented, and multi-agent configurations. It supports isolated workspaces, run logs, diffs, transcripts, AST-based verification, and exportable telemetry, using model access through OpenRouter and GitHub Copilot CLI. Its demonstration covers 100 multi-file RefactorBench tasks across four model families, so the findings are limited to the evaluated tasks and configurations.research paper · Sep 4, 2026
Related questions
How can repository-level coding-agent benchmarks detect review-constraint failures beyond passing functional tests?How can coding agents maintain accurate repository context as codebases evolve without repeated discovery?How can coding agents reliably implement systems-level requirements and detect the defects they introduce?How can scientific agents choose domain-specific procedures that make analyses defensible?