Agent Desktop Ffi

This Skill addresses the significant performance overhead incurred when invoking agent-desktop via CLI subprocesses. In practical development, each operation requires spawning a new process and passing JSON, resulting in extremely low efficiency in high-frequency scenarios. This library provides a C-ABI interface, allowing the dynamic library to be directly linked into languages such as Python, Swift, Node.js, Go, C++, and Ruby, enabling calls with zero process overhead.

Usage is straightforward: you only need to compile and generate a cdylib (.dylib/.so/.dll) using Rust's release-ffi profile, then call ad_init(AD_ABI_VERSION_MAJOR) to validate the ABI version. After that, you can create an adapter via ad_adapter_create or ad_adapter_create_with_session, call ad_snapshot to obtain a UI snapshot, parse the returned JSON to locate element references, and finally use ad_execute_by_ref to perform actions like clicks, completing the observe-act workflow.

It is particularly suitable for agent developers, backend developers, and full-stack developers who need efficient integration of agent-desktop functionality. Especially for teams directly invoking underlying capabilities via ctypes in Python, ffi-napi in Node.js, cgo in Go, etc., it can significantly reduce latency and resource consumption.

It is recommended to strictly match the ABI version number during builds and immediately validate the return value of ad_init upon initialization. Note the memory ownership rules: all strings returned by ad_* functions must be freed by calling ad_free_string. For detailed reference, see documents such as ownership.md and error-handling.md.