developer tools
The Coding Agent Kept Working After the Repository Moved
SWE-Shift resumes identical saved histories with a relevant mid-run change; every tested agent lost success relative to the unchanged continuation.
Summary
SWE-Shift resumes identical saved histories with a relevant mid-run change; every tested agent lost success relative to the unchanged continuation.
Software workspaces do not hold still while an agent reasons. SWE-Shift creates 120 scenarios across 12 change families and nine intervention positions, pairing an unchanged continuation with one that receives a relevant update from the same saved history and files. All 17 evaluated coding-agent systems performed worse after the shift. The paper reports one model falling from 67.35 percent clean success to 24.79 percent shifted success; tool or session changes and stale evidence produced zero paired success across all systems. These are benchmark results, not forecasts for every repository. They isolate a neglected capability: preserving valid work while refreshing state, evidence and requirements after change.
Why it matters
SWE-Shift resumes identical saved histories with a relevant mid-run change; every tested agent lost success relative to the unchanged continuation.
Limits and context
- Software workspaces do not hold still while an agent reasons.
- These are benchmark results, not forecasts for every repository.
Key claims
SWE-Shift resumes identical saved histories with a relevant mid-run change; every tested agent lost success relative to the unchanged continuation.
Qualification: Software workspaces do not hold still while an agent reasons.
Evidence: source-2026-10-26-011
Sources
- arXiv preprint 2610.32196, original v1arXiv · primary research
Corrections
No corrections have been recorded for this story.