cv-deploy skill
The $cv-deploy skill helps an agent set up a computer-vision host: drivers, OpenCV and FFmpeg builds, RTSP streams and services
- 778 ★ on the repository
- Last commit
Broad access
Why this level
- Within your task the skill installs drivers, builds and installs libraries, configures services and works with databases and video recordings.
Install
- Works in
Manual install
Download the LMIXR/CV_Deployment_skill repository and in the root of the target project run `mkdir -p .agents/skills`, then `cp -R "/path/to/CV_Deployment_skill/cv-deployment-skills/cv-deploy" .agents/skills/`. Invoke the skill as `$cv-deploy`.
Access and checks
How we reviewBroad accessThe tool can change data, run commands or handle secrets.
- Within your task the skill installs drivers, builds and installs libraries, configures services and works with databases and video recordings.
This is third-party code. Review the repository files before installing.
What it does
cv-deploy gathers practical experience of deploying computer-vision systems from the helpfile repository into one skill. The agent first identifies the target platform and project constraints, then reads the relevant topic and picks a solution for the concrete system: Ubuntu, CentOS, Windows, macOS, Jetson, Raspberry Pi or RK3399. There are six topics. The first covers the environment: NVIDIA drivers, CUDA, networking and offline installation. The second covers building OpenCV, Qt, inference frameworks, C++ libraries and ABI questions. The third is about video: FFmpeg, GStreamer, RTSP, ONVIF and GB28181. The fourth covers packaging and services: dynamic libraries, Qt plugins, systemd, Windows services and Docker. The remaining two cover MySQL, MongoDB, Redis, MinIO, Kafka and Tomcat, plus Ansible, Jenkins and Jitsi. The skill does not treat old versions as recommendations: the CUDA 10.x or Qt 5.x of the source notes are only a hint about whether a combination applies. Commands substitute your project's paths, ports and versions. If you need only a plan and config files, the skill works that way, without installing or verifying. The guide rests on 119 primary sources from its index.
Who it is for. Engineers who deploy video analytics on Jetson, ordinary servers and Windows and ask an agent to build dependencies and connect cameras.
Good fit when
- A camera connects over RTSP, the stream does not decode on Jetson and the JetPack version cannot change
- A project asks for specific versions of OpenCV and FFmpeg and they have to be built against the existing CMake
- A Qt app starts but shows no camera image, possibly because packaging plugins are missing
- A ready WVP and ZLMediaKit setup has to become a service while keeping ports and the data directory
- You need only an offline deployment plan and config files with no run on the machine
Not a fit when
- You need to train a model or invent an algorithm: the skill handles deployment and training is outside it
- Nobody can read Chinese: topic texts and reference materials are written only in Chinese
- You need an instruction verified across platforms: the author warns that no cross-platform verification was done
- The environment demands only official images: the skill suggests solutions from engineers' notes rather than vendor documentation
Example request
$cv-deploy Find out why RTSP on Jetson connects but does not decode and leave the current JetPack unchanged.Limitations
The skill texts and references are written in Chinese. The author notes that the version passed packaging but had no cross-platform acceptance, and platform coverage differs in depth.
Descriptions from the source notes contain historical versions, including CUDA 10.x, Qt 5.x and FFmpeg 3.x or 4.x. This is comparison material, so check that it matches your environment before running commands. Operations such as removing drivers, changing the boot disk, rebooting a machine and deleting records are carried out by the skill only within the authority you give.
The skill installs by copying the `cv-deploy` folder into the project's `.agents/skills/`. The repository has no overall license file, so check the terms with the author before commercial use.
How to disable. Delete the `.agents/skills/cv-deploy` folder from the project. Configs and services the agent created on the machine are removed separately.
FAQ
How do I install the skill in a project?
Download the repository and copy the `cv-deployment-skills/cv-deploy` folder into `.agents/skills/` of the target project with `cp -R`. Other agents can be given SKILL.md directly.
Which platforms does the skill cover?
Ubuntu, CentOS, Windows, macOS, Jetson, Raspberry Pi and RK3399, plus some Android and iOS tooling. The depth of material differs by platform.
Can I get only a plan without installing?
Yes. The skill follows the scope you set: if you need only a scheme and config files, installing and verifying are skipped.
Where does the skill's experience come from?
From the LMIXR/helpfile repository. The source index collects 119 links to original notes, scripts and configs.
Does the skill change drivers automatically?
No. The presence of NVIDIA is no reason to reinstall a driver. The agent takes risky actions only within the bounds of your task.
Discussion
No questions or reviews yet.
Related
SkillsThe all-deploy skill puts a site, app or API online: audit, preview, response check and rollback
13 NVIDIA skills for an agent on a Jetson device: diagnostics, memory, LLM serving, packages and video and inference benchmarks
31 DevOps skills for Claude Code and Codex: config generators, validators and a Kubernetes debugger for Terraform, Helm, CI and Docker
KubeShark skill: the agent finds the failure mode first, then writes manifests, Helm and Kustomize to official practice with a rollback plan
Render's official 21 skills for deploying, debugging and scaling services from Claude Code, Codex, Cursor and OpenCode.
A large base of DevOps, security, infrastructure and compliance skills for AI agents
README in short
The project turns an engineer's notes into an agent skill and shows a table of topics with example tasks. The README lists ways to invoke it, installation into another project and the origin of the material. The description is given in Chinese and English.