Seeed reCamera Pro Development Skill for Codex

中文说明

Develop applications for Seeed reCamera Pro through natural-language conversations with Codex.

After installing this skill, you can describe what you want to build—such as an AI camera application, an ONNX model conversion, microphone capture, speaker playback, or an RTSP inference stream—and Codex will use the reCamera Pro-specific workflow, tools, and hardware knowledge included in this repository.

Choose your agent branch

Clone the branch for the agent you use. Each branch contains the same reCamera Pro development workflow with agent-specific installation and usage instructions.

Agent	Branch	User skill directory
OpenAI Codex	main	~/.agents/skills/recamera-rknn-dev
Claude Code	claude-code	~/.claude/skills/recamera-rknn-dev
GitHub Copilot	github-copilot	~/.copilot/skills/recamera-rknn-dev
Cursor	cursor	~/.cursor/skills/recamera-rknn-dev
Gemini CLI	gemini-cli	~/.gemini/skills/recamera-rknn-dev

This main branch is the Codex edition.

What you can ask Codex to do

Convert AI models

Ask Codex to convert an ONNX model into an RKNN model for the RV1126B NPU. The skill keeps RKNN-Toolkit2 and RKNN Runtime aligned at version 2.3.2, inspects the model contract, handles FP16 or INT8 conversion, and preserves conversion metadata.

Example:

Use the reCamera Pro skill to convert my model.onnx into an RKNN model for RV1126B.

Build native AI applications

Ask Codex to create or modify C/C++ applications that use the RKNN Runtime on reCamera Pro. The skill understands the aarch64 target, cross-compilation requirements, sysroot layout, runtime library path, ABI checks, and deployment constraints.

Example:

Build a reCamera Pro application that captures camera frames and runs my RKNN object-detection model.

Use the camera, microphone, and speaker

The skill contains verified information about the reCamera Pro camera node, ALSA audio device, microphone capture, speaker playback, media-device conflicts, and supported formats.

Example:

Add microphone recording and speaker playback to my reCamera Pro application.

Develop GStreamer and RTSP pipelines

Ask Codex to inspect the available GStreamer plugins, prepare cross-compilation dependencies, consume or publish RTSP streams, and integrate inference results into a multimedia application.

Example:

Create an RTSP inference application for reCamera Pro and explain which dependencies need to be cross-compiled.

Diagnose the development environment

Codex can inspect the host, toolchain, sysroot, target libraries, camera and audio devices, GStreamer plugins, ELF dependencies, and RKNN version compatibility before building.

Example:

Check whether my computer is ready to cross-compile applications for reCamera Pro and fix the missing development dependencies.

Supported platform

This skill is designed specifically for:

• Seeed reCamera Pro
• Rockchip RV1126B
• aarch64 Linux
• RKNN-Toolkit2 2.3.2
• RKNN Runtime 2.3.2

It is not intended for the SG2002/riscv64 reCamera platform.

Install

Ask Codex to install it

Send Codex this request:

Install the reCamera Pro development skill from:
https://github.com/Seeed-Projects/recamera-pro-development-skill.git

Install manually

git clone --branch main --single-branch https://github.com/Seeed-Projects/recamera-pro-development-skill.git
cd recamera-pro-development-skill
./scripts/install_skill.sh

The skill is installed as:

~/.agents/skills/recamera-rknn-dev

Codex detects skills in this user-level Agent Skills directory. If it does not appear in the current session, restart Codex.

Use

You can mention the skill explicitly:

Use $recamera-rknn-dev to build an application for reCamera Pro that detects people and publishes the annotated video over RTSP.

You can also describe the task naturally after the skill is installed:

I have an ONNX detection model. Convert it for reCamera Pro and create the corresponding C++ camera application.

Codex will load the relevant conversion, cross-compilation, camera, audio, or streaming guidance automatically. Device connections, file transfers, and on-device execution are performed only when you explicitly request and authorize them.

Included knowledge

The repository packages reusable Codex instructions, scripts, references, a native application template, and the Seeed-qualified RKNN Runtime 2.3.2 cross-link library. The technical commands are intentionally kept inside the skill so users can work primarily through natural language.

Before public release, the repository owner must add the applicable project license and any required notices for redistributed binary components.