Guide · Agent skill
How to use the Coding in a Repository agent skill
Updated · written with AI from the project's own documentation and checked against our catalogue data
The Coding in a Repository skill, published in Leon's skills directory under skills/agent/coding, teaches an agent to investigate, change and verify code inside a real project folder. It is aimed at developers who want an agent to make edits to an existing repository and check them with the project's own tests. After reading this guide you will know what the skill does, how to install it, how to ask for work in plain language and which requirements and safety points to check first.
Coding in a Repositoryleon-ai/leon · MIT Download the skill Quick start
- 1Download the ZIP. Download the coding skill ZIP from this page. It contains the SKILL.md file.
- 2Unzip into a skills folder. Unzip into ~/.claude/skills/ for a personal install, or into the project's .claude/skills/ folder.
- 3Or upload in the Claude apps. In the Claude apps, go to Settings → Capabilities → Skills and upload the ZIP.
- 4Check the requirements. Confirm Leon's file, search and shell tools are available, the project's toolchain is installed, and Git is present if the repo is versioned.
- 5Make a specific request. Describe the change, the outcome you need and any scope limits, such as analysis only or no commit.
- 6Review the result. Read the reported checks and the final diff before deciding what to keep or commit.
What the coding skill makes the agent do
An agent skill is a folder containing a SKILL.md file with instructions that an AI agent loads when a task matches them. This skill, named coding, tells the agent how to work on code in a repository using Leon's file, search and shell tools.
In practice the agent first establishes the working directory and reads the project's instructions and nearby code. It reproduces a bug before fixing it, makes targeted edits, and runs the project's configured tests and checks before reporting back. It also respects scope: analysis alone does not authorise edits, and commits, pushes or publishing happen only when you ask.
- Reads applicable AGENTS.md files from the repository root down to the target directory
- Inspects Git status and diffs and preserves unrelated changes you already have
- Uses targeted edits rather than overwriting files from partial views
- Runs baseline and final checks, and reports which passed or failed
Install the coding skill
You can install the skill in two ways. For Claude Code, download the ZIP from this page and unzip it into ~/.claude/skills/ for a personal install, or into a project's .claude/skills/ folder to share it with that project. In the Claude apps, you can instead upload the ZIP under Settings → Capabilities → Skills.
The download contains SKILL.md. Once it is in one of these locations, the agent can load it when a coding request matches its description.
- Personal install: unzip into ~/.claude/skills/
- Project install: unzip into the project's .claude/skills/ folder
- Claude apps: upload the ZIP under Settings → Capabilities → Skills
Trigger the skill with natural requests
You do not need a special command. Describe the change you want in the repository you are working in, and mention the outcome you need to see. The skill's description covers investigating, implementing and verifying code changes, so requests that combine those steps fit well.
Be specific about scope. If you want analysis only, say so, because the skill will not edit files, commit or push without authorisation. If you want a commit, ask for it explicitly.
- "Investigate why the export test fails in this repo, reproduce it, then fix it and run the tests."
- "Add a retry to the upload client in this project. Follow AGENTS.md and run the project's test suite."
- "Review the uncommitted changes and tell me what is safe to keep. Do not commit anything."
- "Fix the off-by-one bug in the pagination helper, run the affected tests, and suggest a commit message."
What the skill needs to run
The skill depends on Leon's tool set. Its instructions name file.read, file.patch, file.write, shell.executeCommand and shell.startSession, so it is not a standalone coding tool that works on its own. The project itself must have its own toolchain, package manager and test runner installed, because the agent runs those commands. If the repository is under version control, Git must be available for the status and diff checks.
The skill also runs scripts and is not limited to reading files, so review what the agent is about to run when working on sensitive code.
- Leon's file, search and shell tools (file.read, file.patch, file.write, shell.executeCommand, shell.startSession)
- The target project's own toolchain, package manager and test runner
- Git, if the repository is under version control
- Runs scripts: yes. Uses network: no
Safety notes for scripts and uncommitted work
The skill is designed to protect uncommitted work. It records and preserves unrelated changed and untracked files, and it does not reset, clean, stage or commit your work without authorisation. Even so, edits and checks can contain mistakes, so review the final diff before you accept it.
Because the skill runs project commands, avoid unattended use on repositories where important uncommitted work exists. Withhold publication of credentials or private data until outgoing content has been cleaned and rechecked. Git ignore rules do not remove secrets that are already tracked or committed.
- Review the diff before accepting changes
- Commit or back up important uncommitted work before a long unattended run
- Do not paste secret values into requests; the skill is instructed never to print them
Tips for better results
Point the agent at the right folder and say which files or features matter. Keep project rules in AGENTS.md files, because the skill reads them before running commands or editing. If you have a specific test command, name it, since the skill uses the project's configured runner and will not invent a separate test suite.
For larger tasks, state the acceptance criteria up front. The skill records criteria and evidence, and reports any requirement it could not verify, so you know what still needs checking.
- Put build, lint and test rules in AGENTS.md so the agent follows them
- State acceptance criteria and the checks that must pass
- Ask for the diff and suggested commit message at the end, then commit yourself
What people use it for
- Reproducing and fixing a failing test in an existing project, then running the affected tests
- Making a targeted change across a few files while following the project's AGENTS.md rules
- Reviewing uncommitted work and identifying which changes are safe to keep, without committing
- Preparing a suggested commit message after changes have passed the project's checks
Frequently asked questions
How do I install the coding agent skill?
Download the ZIP from this page and unzip it into ~/.claude/skills/ for a personal install, or into a project's .claude/skills/ folder. In the Claude apps, you can upload the ZIP under Settings → Capabilities → Skills instead.
What does the coding skill require?
It requires Leon's file, search and shell tools (file.read, file.patch, file.write, shell.executeCommand and shell.startSession), the target project's own toolchain, package manager and test runner, and Git if the repository is under version control. It is not a standalone coding tool.
Is the coding agent skill free and what is its licence?
The skill is listed under the MIT licence in the Leon repository. This page does not state any price, so check the directory listing for any cost details.
Will the skill commit or push my changes?
No, not unless you ask. The skill treats analysis as not authorising edits, commits, pushes or publication, and it will not reset, clean, stage or commit your work without authorisation.
Does the coding skill use the network?
The listing states it does not use the network, but it does run scripts. Review the commands it runs when working on sensitive code.
Why isn't my coding skill following my project rules?
Make sure your project rules are in AGENTS.md files at the repository root or in the target directory. The skill reads these before running commands or editing, so rules placed elsewhere may be missed.
New agent skills, repos and AI tools — every week
The best new skills and open-source projects, the fastest-growing repos, and one free prompt. Five minutes to read. Unsubscribe anytime.
Our verdict
Choose this skill if you are a developer or team using Leon's agent framework and want an agent to make and verify changes in an existing repository while respecting AGENTS.md and Git state. Pick something else if you do not use Leon's tool set, want a standalone coding tool, need unattended runs on repositories with important uncommitted work, or are looking for a general writing or design helper.
Download the skill