CanMV K230 Skills

中文 | English

Production-oriented agent skills for CanMV K230 workflows.

This repository currently ships two focused skills:

• canmv_k230_control: host-side board control over the CanMV IDE serial protocol
• canmv_k230_coding: offline coding knowledge and templates for CanMV K230 MicroPython

The goal is simple: help an agent control a real K230 board when needed, and help it write correct K230 code without relying on generic Python guesses.

Why This Repo

Most agent-skill collections are broad catalogs. This repo is intentionally narrow and practical:

• built around real CanMV K230 workflows
• optimized for single-file board development
• grounded in bundled references and scripts instead of vague instructions
• suitable for project-local installation

If you want a large curated directory of community skills, see awesome-agent-skills.

Skill Catalog

Skill	What it does	Best used for	Includes
canmv_k230_control	Talks to a CanMV K230 board from the host using the IDE serial protocol	probing ports, running a local .py, stopping execution, following logs, fetching frame buffers, inspecting MTP files	CLI script, serial-protocol notes, usage notes, agent config
canmv_k230_coding	Gives an agent offline K230 API knowledge for writing and reviewing code	machine, network, socket, Sensor, Display, MediaManager, RTSP, image, nncase, Ai2d, YOLO	7 reference files, 6 starter templates, agent config

Quick Install

Install one or both skill folders into the local skill directory used by your agent environment.

Common project-local layouts:

.codex/skills/<skill-name>
.agents/skills/<skill-name>
.opencode/skills/<skill-name>

Example:

.codex/skills/canmv_k230_control
.codex/skills/canmv_k230_coding

After installation, each skill folder should contain its own SKILL.md.

Which Skill Should I Use?

Use canmv_k230_control when the agent needs to interact with a real board:

• detect the correct serial port
• run the current local file without replacing main.py
• stop a stuck session
• monitor logs or capture frame-buffer output
• read or upload files through Windows MTP helpers

Use canmv_k230_coding when the agent needs to write or explain K230 code:

• generate single-file CanMV scripts
• stay within confirmed CanMV APIs
• reuse known patterns for media, network, image, and AI tasks
• avoid generic desktop-Python assumptions

Example Prompts

Examples for canmv_k230_control:

Use $canmv_k230_control to probe the connected K230 board, find the right COM port, run app.py, and follow logs for 15 seconds.

Use $canmv_k230_control to stop the current board session on <PORT> and fetch one frame buffer image.

Examples for canmv_k230_coding:

Use $canmv_k230_coding to write a single-file CanMV K230 RTSP demo with Sensor, Display, MediaManager, and RTSP.

Use $canmv_k230_coding to review this K230 script and check whether the media lifecycle order is correct.

Repository Layout

skills/
  canmv_k230_control/
    agents/
    references/
    scripts/
    SKILL.md
  canmv_k230_coding/
    agents/
    assets/templates/
    references/
    SKILL.md

Notes

• This repository publishes reusable skills, not a full end-user application project.
• canmv_k230_control is the operational skill. It is most useful when a real board is connected.
• Windows MTP helpers in canmv_k230_control are Windows-specific by design.
• canmv_k230_coding is offline-first and intentionally narrow. It does not try to be a full CanMV documentation mirror.
• The skills are focused on current behavior. They do not add compatibility layers for unrelated runtimes by default.