SyncBench quantifies agent out-of-sync recovery from 21 real repositories and finds state-of-the-art LLM agents succeed in under 34% of tasks and ask for help in under 5% of turns.
Claude , 2023
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
citation-role summary
background 1
citation-polarity summary
fields
cs.SE 1years
2025 1verdicts
CONDITIONAL 1roles
background 1polarities
background 1representative citing papers
citing papers explorer
-
SyncMind: Measuring Agent Out-of-Sync Recovery in Collaborative Software Engineering
SyncBench quantifies agent out-of-sync recovery from 21 real repositories and finds state-of-the-art LLM agents succeed in under 34% of tasks and ask for help in under 5% of turns.