An SDK for building AI agents that refactor your code

What it does The OpenHands Software Agent SDK is a set of APIs for orchestrating autonomous agents that perform software engineering tasks. You point an agent at a workspace—your local disk or an ephemeral Docker/Kubernetes container via the Agent Server—and it can execute terminal commands, edit files, and track tasks. The same engine powers the OpenHands CLI and Cloud offerings.

The interesting bit The SDK ships with a public skills marketplace that auto-detects modern package tooling: if your repo contains uv.lock or deno.json, the agent can load corresponding uv or deno skills without manual configuration. That saves you from writing yet another “how to install dependencies” prompt.

Key highlights

• Supports both standalone local execution and remote client-server architectures over WebSocket
• Ships with pre-built tools for terminal access, file editing, and task tracking
• Public skills marketplace auto-loads uv and deno guidance when it detects lockfiles or config markers
• Includes examples for CI/CD GitHub workflows, not just local scripts
• Repo badge advertises a 77.6 SWE-bench score

Verdict Worth a look if you are building autonomous coding tools and need a structured alternative to ad-hoc agent scripts. Skip it if you just want a chat wrapper around an LLM—this is designed for agents that actually mutate files.