Learning Loop — a GitHub Copilot CLI plugin

An iterative, top-down study loop you can use in any folder from the terminal:

topic → learn → quiz → lab → repeat

It's resumable (progress lives in learnlog.md), artifacts stay in your working directory (learn/, lab/), and the whole thing installs once as a Copilot CLI plugin instead of being symlinked into every project.

Who it's for

Developers who want a lightweight, hands-on, structured way to learn a topic and practice it — active learning instead of passively watching videos.

What's in the plugin

• ulearn coach (agents/ulearn.agent.md) — the orchestrator. It tracks progress in learnlog.md, decides the next step, and runs it.
• Step skills — topic, learn, quiz, lab.
• Utility skills — ucode (understand a codebase), lpp-lint (validate an LPP module).

See AGENTS.md for the design and how the pieces compose.

Prerequisites

• GitHub Copilot CLI installed and logged in (copilot, then /login if needed).

Install

1. Clone this repo:

   git clone https://github.com/asyrovprog/vscode_prompts.git
   cd vscode_prompts

2. Install it as a plugin (point at the repo directory):

   copilot plugin install ./

3. Verify it loaded. Start Copilot CLI and run:

   /plugin list          # learning-loop should appear
   /agent                # the "ulearn" coach should be listed
   

   The step skills (topic, learn, quiz, lab) and utilities (ucode, lpp-lint) are internal — they're marked user-invocable: false, so the coach uses them but you don't select them directly. The only thing you pick is the ulearn agent.

Note: installing from a local path (or a git repo/URL) currently works but is deprecated — a future CLI release will support only marketplace installs (copilot plugin install <name>@<marketplace>). Publishing this plugin to a marketplace is on the roadmap; until then, the local-path install above is the way to use it.

Updating: after pulling changes, re-run copilot plugin install ./ (plugin components are cached). Mid-session you can run /skills reload.

Uninstall: copilot plugin uninstall learning-loop.

That's it — the coach and skills are now available in every folder. No per-project setup.

Walkthrough

Open a terminal in a folder where you want your learning notes to live (a fresh ~/learning folder is fine), then start Copilot CLI:

mkdir ~/learning && cd ~/learning
copilot

1. Start a session. Pick the coach with /agent → ulearn, or just ask:

I want to learn Semantic Kernel

2. Learn (~20 min). The coach runs the learn step: it generates compact, top-down materials (intuitions, diagrams, a few web references) and writes a copy to learn/learn01.md. Read it. When you're done, type:

next

3. Quiz (6–10 questions). The coach runs the quiz step and prints the quiz. Answer compactly, e.g.:

1:A,2:BC,3:Y

It scores you immediately (writing answers to learn/quiz01_answers.md). Score < 70%? Type explain for a recap and a retry. Otherwise:

next

4. Lab (~25–30 min). The coach runs the lab step: it asks C# or Python, proposes a couple of concepts, and scaffolds a LeetCode-style exercise under lab/iter01/ with TODO stubs and failing tests. Fill in the TODOs, then:

check        # runs the tests and reports pass/fail (won't fix anything for you)
explain      # optional: hints tied to each TODO

When the lab passes:

next

5. Repeat. The coach proposes the next topic based on your progress and starts iteration 02.

Stop anytime. Everything is recorded in learnlog.md. Next time, just open the folder, start copilot, select the ulearn coach (or say "resume my learning"), and it picks up exactly where you left off — on this machine or another.

Commands at a glance

Command	What it does
next	Complete the current step and move to the next
prev	Go back to the previous step
explain	Quiz: explain wrong answers · Lab: hints tied to TODO IDs
check	Lab only: run the tests and report pass/fail

The skills are internal

You don't invoke the step skills directly — they're marked user-invocable: false. The ulearn coach calls them as it runs the loop. Your single entry point is the ulearn agent (via /agent or natural language).

Where things live

• The tool (this plugin): installed once via copilot plugin install.
• Your learning data: learn/, lab/, and learnlog.md in each learning folder you work in. Keep those folders in their own git repo if you want to sync across machines — they're independent of this plugin.

This repo's .gitignore excludes generated artifacts so they never get committed here.

Notes

• This was originally a set of VS Code Copilot slash-prompts (hence the repo name vscode_prompts); it now targets GitHub Copilot CLI only, packaged as a plugin. VS Code support has been dropped. (The instructions inside stay model- and tool-agnostic — they avoid hardcoded tool IDs and model names — but the packaging is CLI-specific.)
• The prompt modules use Literal Prompt Programming (LPP) — a small shared vocabulary between the author and the model. See lpp_spec.md.
• Future: publish to a plugin marketplace for one-command install.

License

MIT — see LICENSE.