Skills
My reusable agent skills.
Skills
| Skill | Purpose |
|---|---|
| Differential regression review | Compare base and pull request behavior with temporary end-to-end tests. |
| Draw.io pitch | Research the subject and visual direction, then create a dense A5 diagram and PNG. |
| Draw.io iterate | Complete at least four PNG review passes: an 800% diagram-code increase, then corrections only. |
| Narrated PR walkthrough | Create an animated HTML walkthrough by default, or a narrated MP4 video when requested. |
Use drawio-pitch for the initial proposal, then drawio-iterate to refine it.
The skills do not prescribe a visual style or PNG export tool.
Differential regression review requires Git and the target project's test dependencies.
Narrated PR walkthrough uses HTML by default. The video option requires both
draw.io skills, GPT-6 Astra subagents, a draw.io PNG export tool, Docker,
FFmpeg, and FFprobe. The skill works best with a Claude Opus 5.5 or higher
subagent for the HTML option. The video option uses the model requirements in
its instructions and the linked draw.io skills.
Install
Copilot marketplace
In Copilot's plugin marketplace settings, add aymenfurter/skills as the
source. Then install aymenfurter-skills from the aymenfurter-skills
marketplace. The plugin includes all four skills and their supporting files.
From Copilot CLI:
copilot plugin marketplace add aymenfurter/skills
copilot plugin install aymenfurter-skills@aymenfurter-skills
To use a local checkout before publication, run these commands from the repository root:
copilot plugin marketplace add .
copilot plugin install aymenfurter-skills@aymenfurter-skills
The marketplace catalog is in
.github/plugin/marketplace.json.
The root plugin.json uses the Agent Plugins 1.0 format.
It loads each skill from an immediate subdirectory of skills/.
Individual skills
From a local checkout, use GitHub CLI 2.90.0 or later:
gh skill install . --all \
--from-local --agent github-copilot --scope user
To install one skill, replace --all with its name.
Use either the plugin or individual skill installation to avoid duplicate
copies.
Diagram samples
Recommended model: GPT-6 Astra. It delivered the best result in this comparison.
Both models refined the same AI Engineer Coach draft through four
drawio-iterate passes.
GPT-6 Astra
AI Engineer Coach architecture: local analysis, hosting environments, and optional model calls
GPT-5.6 Sol reference output
Kept for comparison only. This version has clipped labels and omits some secondary connectors that are present in the Astra result.
AI Engineer Coach architecture reference produced by GPT-5.6 Sol
Source and license
Differential regression review is copied unchanged from aymenfurter/differential-regression-review.
MIT License. Each skill directory also includes the license for standalone distribution.