AI with MattLearn as you work.Member area

Understand the project.
Then make your change.

Your free awm-skill: Understand an unfamiliar project. Get the full instructions and download, with no signup required.

Start with an unfamiliar project.

Use the skill to explore the actual source, find relevant entry points and separate what your assistant verified from what it only suggested.

What to have ready

  • A project checkout or supplied files, and the task you want to understand.
  • Execution is optional; commands must be distinguished from checks actually run.

The current download is our exact authored review version. Keep its notice and check its output on your task.

Use it with your assistant.

Follow the first-skill walkthrough for file paths, a small task and recovery when setup differs.

Codex

Unzip the skill folder into your project’s .agents/skills/ directory. The skill file should be at .agents/skills/awm-repo-orientation/SKILL.md. Check that your client can see it before asking it to work.

Claude Code

Use your project’s .claude/skills/ directory instead. This is documented setup guidance, not a verified local Claude Code run.

A local-model client

Supply the instructions explicitly in your client. Automatic discovery is client-dependent; a model alone doesn’t load a skill folder.

The full skill instructions

These are the same authored instructions included in the download and member library.

---
name: awm-repo-orientation
description: Orient a member to an unfamiliar code repository, trace a representative workflow, and establish a supported first local verification step. Use for onboarding or returning to a codebase, not automatic repair, deployment or a full security audit.
---

# 🧭 Find your way around a repository

Announce use as `🧭 awm-repo-orientation`.

Help the member understand what runs, how the important parts connect, and what they can safely verify next. Adapt to their access: a checkout with tools supports observed results; pasted files support a source-based orientation. Do not claim to have run a command merely because it appears in a README.

## Establish the starting point

Use the member’s goal and supplied repository. Ask only for a missing detail that changes the next step, such as which application in a monorepo they want to run. Read the closest project instructions, current status if available, manifests and existing setup/architecture notes. Preserve user changes and use the repository’s existing documentation structure; do not create parallel setup guides by default.

Identify the declared runtime, package manager, lockfile, entry point and prerequisites. When sources disagree, identify the conflict and explain which current source supports the proposed command. Missing files or unavailable tools are explicit limits, not a reason to invent a familiar framework command. Inspect configuration examples for required variable names without displaying credentials or reading live secret stores.

## Trace one useful path

Follow a representative request, command or job through actual callers and data boundaries. Distinguish accepting work from processing it and returning a response from a later side effect. Link the entry point, main modules and data ownership to file locations. Label inferred design intent; a plausible explanation for a design choice is not recorded history. A small diagram is useful only when it clarifies that path.

Keep observations focused on what a newcomer needs. An unreferenced file or TODO is not proof that code is dead or may be deleted. Route a concrete bug or deeper audit to a separately scoped task rather than turning onboarding into a rewrite.

## Establish the first verification step

Read the command and relevant hooks before execution. Test, install and setup labels do not guarantee local or harmless behavior: inspect lifecycle hooks, environment selection and remote services. Do not run production migrations, deployment, seeds or native-client configuration writes as onboarding checks. If a supplied command crosses that boundary, explain the specific effect and choose an already supported local check when one exists. Do not suggest bypassing install scripts as a universally safe substitute.

When tools and scope permit, run a bounded local check and retain the actual result and exit status. Preserve failures and explain what they block; do not silently fix unrelated code or filter output so a failure looks green. When execution is unavailable, provide a source-supported command for the member, clearly marked unrun, and explain what result to bring back. Do not fabricate expected test counts or a successful setup.

## Give the member a usable handoff

Return what the project does, the representative flow, the relevant prerequisites, and the next executable step with its working directory. Separate observed checks, source-supported instructions and unresolved requirements. Call out the most consequential verified gotcha. Scale detail to the member’s task; do not invent productivity estimates or require two new documents. If documentation edits are requested, update the appropriate existing notes and preserve project rules.

## About this skill

An AI with Matt skill. Matt’s version adds access-aware verification, command-effect inspection and integration with existing project documentation. Source history and evaluation limits are recorded with the draft; this package does not depend on Patientize tools or paths.

One skill is a starting point.

Membership brings together our awm-skills and prompts with a guided beginner path, prerequisites and explanations. You can explore the offer or preview the member skills section.