In this article, we present a temporal multi-agent logic called Updatable Strategy Logic (USL) which subsumes the main propositions in this area, such as ATL-ATL*, ATLsc and SL. These logics allow to express the capabilities of agents to ensure the satisfaction of temporal properties. USL mainly differs from SL in two ways. Semantically, the notion of strategy composition is extended to enable an agent to update (or refine) its own strategy without revoking it. Syntactically, a new operator, called unbinder, is introduced: it allows an agent to explicitly revoke a strategy, whereas it is implicitly done according to SL semantics. We show that USL allows to express the notion of "sustainable capability" for an agent, i.e. a capability that still holds even after it has been employed. This makes USL strictly more expressive than SL. We also show that the model-checking problem for USL is decidable (but non-elementary as for SL), and that it is PSPACE-complete for its memory-less version.