Mad skillz for you
Michael Wimble
Hi ROS community,
I'd like to share a project I've been building for my own ROS 2 work that I think others will find useful: ros2-copilot-skills, a library of 158 expert-level AI skill files for GitHub Copilot, Claude, and any other AI coding assistant that supports the VS Code skills/instructions format.
What are "skills"?
If
you use GitHub Copilot Chat today, you've probably noticed that it
knows a lot about Python, C++, and software engineering in general
— but it has no persistent memory of your domain.
Every conversation starts from scratch, and the AI has to re-learn
the context of your project from whatever you paste in. Skills are
a way to fix that. A skill is simply a Markdown file that sits in
a special directory (~/.copilot/skills/)
that VS Code automatically makes available to Copilot at the start
of every conversation. When you ask Copilot to help with something
that matches a skill — tuning the MPPI controller, writing a
launch file, configuring AMCL — Copilot reads the relevant skill
file first, so its answer is grounded in expert-level,
domain-specific knowledge rather than general internet training
data. Think of it as a permanently open reference book that
Copilot consults before answering you. You don't have to do
anything differently: just keep using Copilot Chat the way you
always have, and the skills load silently in the background when
they're relevant.
What it is
Each skill is a single SKILL.md file that teaches the AI assistant deep, opinionated knowledge about a specific ROS 2 topic — not surface-level API documentation, but the things that take weeks to learn from painful experience: parameter interactions, tuning strategies, common failure modes, and working code examples.
The AI agent reads only a short description (~100 tokens) during discovery. When a skill is relevant to your current task, the full file is loaded automatically. This means you can have all 158 skills installed without any token overhead on unrelated conversations.
What's covered
The library spans 15 categories:
- ROS 2 core — nodes, launch files, lifecycle, QoS, TF2, parameters, actions, testing
- Nav2 — full coverage of costmaps (obstacle, voxel, inflation, range, denoise, keepout, speed zones), all planners (NavFn, SMAC 2D/Hybrid-A*/Lattice, Theta*), all controllers (MPPI with every critic, DWB with critics, Regulated Pure Pursuit, Rotation Shim), behaviors, lifecycle management, error codes, simple commander
- Behavior Trees — XML format, every BT.CPP node type, custom C++ and Python actions/conditions, BT logging and replay, Groot2 integration, patrol and recovery tree patterns
- SLAM and mapping — SLAM Toolbox (online and lifelong), Cartographer, map server/saver, loop closure, map merging, multi-floor mapping, best practices
- Localization — AMCL tuning (the alpha1–alpha5 motion model parameters explained physically), EKF and UKF sensor fusion (the 15-element boolean vectors decoded), IMU integration, wheel odometry, localization recovery
- Perception — LiDAR driver config, laser scan processing, pointcloud processing, depth cameras, YOLO ROS 2 integration, DepthAI spatial AI, visual odometry, wall line extraction, anomaly detection, person tracking
- Hardware — Teensy/PlatformIO, serial communication protocols, motor controller interfaces, I²C sensors, servos, IMU drivers, udev rules, hardware watchdog, emergency stop, hardware abstraction layer, micro-ROS
- Simulation — Gazebo Harmonic setup, SDF world building, physics tuning, simulated sensors, ros_gz_bridge, cross-compilation
- URDF — fundamentals, Xacro macros, inertia calculation, joint types and limits, differential drive model, sensor frame mounting, Gazebo plugins
- DevOps — colcon workspaces, workspace overlays, Docker for ROS 2, systemd autostart, GitHub Actions CI, rosdep dependencies
A few highlights from the deeper skills:
mppi-critics/— Every MPPI critic plugin documented with parameter interactions and three weight profiles (conservative, balanced, aggressive).amcl-tuning/— The alpha1–alpha5 odometry noise model parameters explained with physical meaning, not just "set these numbers."ekf-sensor-fusion/— The 15-element boolean state vector forrobot_localizationdecoded element by element, with worked examples for common sensor combinations.costmap-architecture/— Includes the critical gotcha that thepluginslist is order-dependent (layers apply in sequence) butobservation_sourcesis not.inflation-layer/— The actual exponential cost formula, explained so you can reason about clearance distances and footprint inflation before you touch a parameter.
Why it matters
If
you use GitHub Copilot or Claude to help write ROS 2 code, you've
probably noticed that AI assistants are good at Python and C++ in
general but often produce subtly wrong ROS 2 — wrong QoS profiles
for sensor topics, broken PythonExpression boolean
conditions in launch files, Nav2 parameters that exist in Humble
but were renamed in Jazzy, and so on. These skills give the AI the
grounding it needs to produce correct, idiomatic ROS 2 code on the
first try.
Installation — one command
That's it. Reopen VS Code and every future Copilot/Claude conversation in any workspace will have access to the full library. To stay current:
Target distributions: ROS 2 Jazzy and Rolling. Most content applies equally well to Humble and Iron, though some parameter names differ.
License: Apache 2.0
Contributions are very welcome — if you have deep knowledge in an area that isn't covered, or want to improve an existing skill, please open a pull request.
https://github.com/wimblerobotics/ros2-copilot-skills
Michael
Wimble
Wimble Robotics
James H Phelan
Michael,
Fantastic! Can't wait to try it.
Where do I install this? In Github? How to I clone into GitHub? Or do I fork your repo? Create a new repository for skillz?
James
James H Phelan "Nihil est sine ratione cur potius sit quam non sit" Leibniz
--
You received this message because you are subscribed to the Google Groups "HomeBrew Robotics Club" group.
To unsubscribe from this group and stop receiving emails from it, send an email to hbrobotics+...@googlegroups.com.
To view this discussion visit https://groups.google.com/d/msgid/hbrobotics/bda7cb99-f308-4441-a79f-46ce8c27c1db%40gmail.com.
Michael Wimble
git clone https://github.com/wimblerobotics/ros2-copilot-skills.git ~/.copilot/skills
James H Phelan
I'm running VSCode on a Windows laptop but SSHing to the Raspberry Pi / Ubuntu 24.04 so I'd install it on the robot & on my other LinuxLaptop.
James H Phelan "Nihil est sine ratione cur potius sit quam non sit" Leibniz
--
You received this message because you are subscribed to the Google Groups "HomeBrew Robotics Club" group.
To unsubscribe from this group and stop receiving emails from it, send an email to hbrobotics+...@googlegroups.com.
To view this discussion visit https://groups.google.com/d/msgid/hbrobotics/8F6C5B28-296E-4DA9-AC7F-C50643F1E1D7%40gmail.com.
Michael Wimble
On Windows, the idea is the same as on Linux—you just need to map ~/.github to the Windows equivalent of your home directory.
📁 Where ~/.github lives on Windows
In Windows, ~ points to your user profile folder, usually:
C:\Users\<YourUsername>\
So the equivalent path becomes:
C:\Users\<YourUsername>\.github
🛠️ How to clone the skills directory
- Open a terminal:
- Command Prompt, PowerShell, or Windows Terminal
Navigate to your home directory (optional, you’re usually already there):
cd $HOMEClone into
.github:git clone <repo-url> .githubOr, if
.githubalready exists:git clone <repo-url> $HOME\.github
✅ Notes
$HOMEworks in PowerShell; in Command Prompt you’d use%USERPROFILE%.- The
.githubfolder may be hidden by default—this is normal. - Functionally, this behaves the same as
~/.githubon Linux or macOS.
⚠️ Common gotcha
If your friend is using something like GitHub Copilot “skills” or extensions that expect a Unix-style path, make sure:
- The tool supports Windows paths, or
- They’re running inside something like Windows Subsystem for Linux, where the Linux instructions would apply directly.
Pito Salas
I just perused this, it’s pretty amazing. Lots of deep ROS2 expertise embedded in them. You might add how to use these with claude. It might be trivial but it would save some trial and error.
A more subtle question is the tradeoff between open sourcing a library of code that embodies this knowledge and open sourcing skills. They are two very very different ways of sharing your expertise.
Another suggestion would be to include one worked example of a skill in use. Reading them, as I said, shows lots of ROS2 smarts. For example let’s say I start by typing /wheel-odometry-model, what happens next? There are different scenarios in the skill. Do I just proceed giving specific prompts which require that knowledge, or how do I get the most benefit from that skill?
Best,
Pito
Boston Robot Hackers &&
Comp. Sci Faculty, Brandeis University (Emeritus)
> You received this message because you are subscribed to the Google Groups "HomeBrew Robotics Club" group.
> To unsubscribe from this group and stop receiving emails from it, send an email to hbrobotics+...@googlegroups.com.
Sergei Grichine
To view this discussion visit https://groups.google.com/d/msgid/hbrobotics/1A404E86-C019-42C0-BB71-33C7B023FAB8%40gmail.com.
Sergei Grichine
- https://github.com/search?q=skills&type=repositories
- https://github.com/search?q=SKILL.md&type=repositories
- https://github.com/search?q=SKILL.md+AND+ROS&type=repositories
- https://github.com/search?q=ROS2+path%3A**%2FSKILL.md&type=code
