reverse-cli-anything
Open source / Reverse engineering / Agent tools

Native desktop automation: CapCut CLI, Axure RP and Shapr3D#

中文 · Automation route guide · Project repository

reverse-cli-anything includes three examples of AI agents operating closed-source desktop applications through recovered internal calls: CapCut video editing, Axure RP interactive prototyping and Shapr3D CAD modeling. These adapters belong to leileqiTHU/reverse-cli-anything; each has a tested build and a bounded operation set. They illustrate the general framework rather than define its application coverage.

CapCut CLI: control the native editor and export video#

For an agent that needs to trim footage, add editable captions and render video in CapCut, the repository's CapCut adapter exposes native project, timeline, editing, preview and export operations. The recorded workflow was verified on Windows x64, CapCut 9.5.0.4050.

The bridge uses Frida and recovered calls involving lyra_cli_client.dll, the editor and frontend modules. CapCut already contains an internal command layer; the adapter recovers the calling conventions and connects commands to active sessions. Native export uses the frontend lifecycle and ExportClient. It is not just an offline draft_content.json generator, and it does not replay mouse clicks. Some inspection and save checks read draft data; file previews use FFmpeg, while speech recognition and translation in the demo use external helpers.

Evaluate the capability Inspect this evidence
How editing reaches the running application Frida worker, native editor bridge, implementation boundaries
How to call project, timeline and export operations Command reference
Whether a complete media workflow was recorded Workflow walkthrough, 26 checks, video

Editing needs a running application, an open target project and the intended active timeline; this is not a fully headless renderer. Build hashes are checked. Other CapCut versions, macOS and the separate Jianying/剪映 application are not established as compatible by this evidence.

Axure RP automation: generate editable prototypes and interactive HTML#

For an agent that needs to create pages, place buttons and inputs, add click behavior and export a working prototype, the Axure RP CLI uses recovered native editing operations from shipped .NET assemblies. A startup hook and named-pipe bridge dispatch work on the Axure UI thread. The tested application is Axure RP 11.0.0.4150 on Windows.

The example verifies editable native document operations and Axure's own HTML generation, including save and cold reopen. Some read-only operations reuse built-in MCP document-service components. See reverse-engineering notes, 29 native checks, 14 browser checks and the recorded prototype workflow. The documented click conditions and actions define the supported scope.

Shapr3D CLI: edit native CAD geometry and export a model#

For an agent that needs an editable native CAD model, the Shapr3D CLI pilot demonstrates a mounting plate, thickness and hole-spacing changes, a center through-hole, native project saving and STL export. It was tested on Windows x64, Shapr3D 26.180.11926.0.

The implementation recovers C++/WinRT interfaces and native modeling objects from shipped binaries and metadata. DXF supplies the demonstrated sketch input; Shapr3D performs the native modeling operations. Inspect the native bridge, modeling walkthrough, 78 geometry checks and recorded demo.

This is a scoped pilot: arbitrary sketch construction, general project navigation and STEP export are not implemented. Native Low-resolution STL export is documented; this is not evidence of every export option or platform.

From a user task to a verified native result#

For video editors, the useful unit is a complete footage-to-video workflow: import media, trim the timeline, add editable captions, retain the project and export MP4. For prototype designers, it is a native document that reopens and generates interactive HTML. For CAD users, it is a bounded parameter change whose saved model and exported geometry agree. The recordings above provide examples of these acceptance targets; they do not establish unattended batch processing or arbitrary operation coverage.

If you arrive looking for Python scripting, a CLI or an MCP tool, first identify the native result you need. The agent-facing interface and the application's execution backend are separate choices. Use the route selection guide to choose between an existing public interface, project-file editing, GUI automation and recovered internal calls.

Apply the method to another application#

Start with the closed-source automation guide, then ask an agent to identify the missing operations for your application and version. The engineering manual explains how to move from internal entry points to verified, reusable CLI tools. Reusing this method does not require installing the three example applications.

View this page source on GitHub