OpenHands lets you scope which secrets each agent profile sees

OpenHands released version 1.20.0 on September 17, and the headline change is scoped secrets, per the 1.20.0 release notes.

Instead of giving every conversation the same ambient environment, a saved profile now declares which secrets it can see. Those profiles are selectable from automations.

That matters for anyone running more than one agent off a single machine. Ambient secrets are the default failure mode in agent tooling. One environment variable set, and every process reads it, letting a low-trust automation end up holding the same API key as a production deploy script. Profile-scoped secrets close that off at the cheapest layer available, before you'd need a broker service or per-agent credential vaults.

The release notes don't say whether existing profiles get scoping applied automatically or start wide open until you configure them. They also don't say whether scoping is enforced at the process level or just hidden from the conversation context. That distinction decides how much protection it buys against a compromised agent reading its own environment.