AI with MattLearn as you work.Member area

Your first skill.
One small, checked task.

A skill is a set of instructions you can give your AI assistant. Start with the free project skill, which helps you find your way around an app before changing it. Use your usual AI coding tool. This page explains the setup; it does not install anything for you. If you want to try a working change first, start with the project estimate tool below.

Try a useful project tool before setting up a skill. You only need a browser.

Start with files you can inspect.

Start with your own project and one small outcome you want to understand. You need an assistant that can receive instructions and permitted project files or excerpts. The teaching project is optional practice if you do not have suitable material ready.

The skill and sample are separate ZIPs. No account or payment is required. Node.js 22+ is needed only if you choose to run the sample commands; reading its source does not require Node.

Inspect the skill and your project.

  1. Find the actual skill folder.

    After extraction, find awm-repo-orientation/SKILL.md. Your unzip tool may add a wrapper directory. Use the folder that directly contains SKILL.md, and keep the package notice alongside your review files.

  2. Open your project.

    Confirm the intended folder, its README and relevant files are available to your assistant. If you chose optional practice, the sample folder is project-orientation. Confirm that you can find src/index.js, src/tasks.js and test/tasks.test.js. For that sample, open its folder in your client or supply those source excerpts through its supported method.

  3. Choose your client below.

    Use one relevant branch. Installing into a local project folder and supplying instructions directly are different methods; neither gives the assistant access to files you haven’t supplied.

Codex

Local project in a Codex client.

Unzip the skill folder into your project’s .agents/skills/ directory. For the free skill, check .agents/skills/awm-repo-orientation/SKILL.md.

Expected file shape

project-orientation/
  .agents/skills/
    awm-repo-orientation/
      SKILL.md
  src/
    index.js
    tasks.js
  test/
    tasks.test.js

This file tree is a sample illustration. In your own project, replace the outer folder and source files with its actual structure; do not create these sample files. The skill folder sits inside the project’s .agents/skills/ folder. Avoid an extra repeated awm-repo-orientation/ layer between that folder and SKILL.md.

Confirm what is available

Ask the client to show that it can read the intended skill before giving it your real task. Client versions and project context can differ.

Ask which skill name and source files it can actually access. Compare any quoted instruction or file detail with the original. A statement that it “loaded successfully” alone is not evidence that the intended instructions were used.

Official OpenAI skill documentation

Claude Code

Local project sessions; account/cloud skill loading can differ.

For a local project, place the unzipped skill folder in .claude/skills/. For the free skill, check .claude/skills/awm-repo-orientation/SKILL.md.

Expected file shape

project-orientation/
  .claude/skills/
    awm-repo-orientation/
      SKILL.md
  src/
    index.js
    tasks.js
  test/
    tasks.test.js

This file tree is a sample illustration. In your own project, replace the outer folder and source files with its actual structure; do not create these sample files. The skill folder sits inside the project’s .claude/skills/ folder. Avoid an extra repeated awm-repo-orientation/ layer between that folder and SKILL.md.

Confirm what is available

Confirm the intended skill is available in your current local session. The website has not executed Claude Code or inspected your installation.

Ask which skill name and source files it can actually access. Compare any quoted instruction or file detail with the original. A statement that it “loaded successfully” alone is not evidence that the intended instructions were used.

Official Claude Code skill documentation

A local-model client

Your chosen client controls how instructions reach the model.

Open the skill instructions and supply them explicitly through your client. There is no universal local-model skill directory.

Supply the instructions explicitly

Open the complete free instructions, read them and copy them into your client’s supported instruction/context field, or attach the file if the client supports that. Then supply your project source and your task. There is no universal local-model installation folder.

If the client cannot access files, paste a small permitted excerpt and describe which file it came from. If its context limit cannot fit the necessary material, reduce the task or use a suitable client rather than assuming the missing files were read.

Confirm what is available

Ask your client to confirm the instructions it received. Automatic discovery and file access depend on the client, not just the model.

Ask which skill name and source files it can actually access. Compare any quoted instruction or file detail with the original. A statement that it “loaded successfully” alone is not evidence that the intended instructions were used.

Try one small request.

After supplying the intended skill and your project source, replace the bracketed goal in this reusable starting request:

A prompt is different: supply a prompt directly with your task context. It does not require a skill-folder installation.

Check your own project first.

Open the files the assistant cites. Compare prerequisites and commands with this project’s README, configuration and existing tests. Run only appropriate checks in an environment you can use; record actual output, failures and anything you could not run. A new project may use a different language, structure or test runner.

Keep each project’s instructions, task context and results separate. When switching projects, confirm the open folder and accessible files again; reuse the method, not assumed paths or earlier results.

Follow the reusable project checklist

Optional practice: inspect the teaching sample.

  1. Follow the cited files.

    The supplied entry point imports unfinishedTitles from src/tasks.js. Open both files yourself. An explanation that cites a different or nonexistent file needs correction.

  2. Check the actual rule.

    The sample keeps tasks where done !== true and returns their titles. A missing done flag therefore still counts as unfinished. Compare the assistant’s explanation with the source.

  3. Separate reading from execution.

    If you have a suitable isolated Node.js environment, run node src/index.js and node --test --test-reporter=tap from the sample folder. The saved example has four passing tests; your own run is verified only after you inspect its actual output. Suggested commands are not passed checks.

Compare with the complete worked example and its saved results.

If something doesn’t work

I can’t find SKILL.md after unzipping

Inspect the extracted folders. Find the single skill folder directly containing SKILL.md. Keep the original download; move only the intended skill folder into the client’s project location. Do not rename an unrelated file to make the path look correct.

The skill exists, but my client can’t see it

Check the project you actually opened, the exact folder spelling and whether a wrapper folder added an extra level. Confirm the session has permission to read those files. Consult the linked documentation for that client/version and mode; account/cloud sessions may load skills differently. Don’t mark your setup as checked while discovery is unresolved.

The assistant talks about files I didn’t supply

Ask it to identify which files it inspected and which details it inferred. Supply the missing permitted source or narrow the task. Do not accept invented paths as proof of access.

My local client has no skill feature

Use its supported method to supply the instructions as context. If that is unavailable or too limited for the task, choose a different supported client or a smaller task. A model alone does not discover project folders.

The instructions were supplied, but the result is wrong

Point to the specific source mismatch, ask for a correction and repeat the relevant check. Loading a skill does not guarantee a correct answer. Keep unknowns and unrun checks visible.

Record what you observed.

Use the existing setup checklist to record your own checks. Reading this guide, copying a request or downloading a file does not mark any step complete. Saved setup and task observations stay in your browser; they are separate from account access.

Member destinations require an active membership in the production build. The guide, free skill and teaching sample remain public.