We propose an agent-based computational model for a financial system consisting of a network of banks with interconnected balance sheets comprising fixed assets (e.g. loans to agents outside the network), liquid assets (e.g. cash or central bank reserves), general collateral (e.g. government debt), unsecured interbank loans and reverse-repos to other banks as assets, as well as deposits, unsecured interbank loans and repos from other banks as liabilities. Importantly, we allow banks to use reverse-repo assets as collateral for obtaining repo loans from other banks, that is to say, rehypothecation. Banks need to satisfy liquidity, collateral, and solvency constraints. If the first two constraints are violated because of internal or external shocks, solvent banks attempt to restore them by rebalancing their assets, which might lead to the propagation of the shock because of fire-sale effects (if fixed assets are sold) or liquidity hoarding (if secured or unsecured loans are recalled). Insolvent banks, as well as banks that failed to restore the liquidity and collateral constraints after rebalancing, are removed from the network using a resolution algorithm that includes a netting step (i.e. removal of closed cycles of liabilities) and a novation step (i.e. redistribution of repo assets and liabilities to remaining banks). We show analytically that this proposed resolution algorithm has several desirable properties, most importantly the order-independence of the novation step, and we investigate the stability properties of the network through a series of numerical experiments.