When Agent Skills Go Stale: Runtime Behavior and Selective Maintenance in Evolving Environments
Previously useful agent skills can cut performance by more than 10 percentage points after an environment changes.
Across four task settings and two models, stale skills could also raise execution cost by nearly 40% compared with using no skill. Task success sometimes held steady while execution savings declined. The researchers’ selective maintenance method replays checks from successful runs without additional model calls, repairs affected skill content and rebuilds when local fixes are insufficient. Compared with reusing stale state, maintenance improved accuracy by up to 67 percentage points and reduced execution steps by up to 37%.
ArXiv · AI/CL/LG's note
Across four task settings and two models, stale skills could also raise execution cost by nearly 40% compared with using no skill. Task success sometimes held steady while execution savings declined. The researchers’ selective maintenance method replays checks from successful runs without additional model calls, repairs affected skill content and rebuilds when local fixes are insufficient. Compared with reusing stale state, maintenance improved accuracy by up to 67 percentage points and reduced execution steps by up to 37%.
ArXiv · AI/CL/LG's note
score 5