Coding Agents Explained
What a coding agent does in your project, and how notes, permissions and skills steer it.
A coding agent reads, edits and tests code in your project from plain goals. A notes file, such as CLAUDE.md, gives it standing rules. Permissions, skills, hooks and helper agents steer it. Work in small steps and run the tests.
- Coding agent
- An AI that works inside a code project.
- CLAUDE.md
- A notes file the agent reads at the start of each session.
- Hook
- A small script that runs before or after the agent acts.
A coding agent is an AI that works inside a code project. It reads files, edits them and runs commands. Claude Code is one example. The docs describe it as a tool that understands your whole project. You describe a goal in plain language. It plans, changes several files and checks its work. This lesson explains the main ideas, even if you are not a programmer.
What it can do
- Write tests for code that has none.
- Fix a bug when you paste an error message.
- Build a feature from a plain description.
- Prepare a commit with a clear message.
- Update many files in one go.
Where it runs
Claude Code runs in a terminal, in code editors, in a desktop app and on the web. They all use the same engine. Your project notes and settings work in each place.
How you steer it
- A notes file. A file named CLAUDE.md sits in your project. The agent reads it at the start of each session. Put team rules there: style, tools and what to avoid.
- Permissions. The agent asks before it does risky things. You decide what it may do alone.
- Skills. Save repeat jobs, such as a review checklist, and use them by name.
- Hooks. Small scripts that run before or after the agent acts, such as a formatter.
- Helper agents. The agent can start sub-agents to work on parts of a task at the same time.
- MCP. Connect it to your tickets, docs and chat.
# Project notes - Use plain, short function names. - Run the tests before you say a task is done. - Never edit files in the billing folder without asking. - Our tools: npm for packages, Prettier for formatting.
A safe way to work
Good habits
- Work on a copy or a branch, not on your only version.
- Ask it to explain changes you do not understand.
- Keep tasks small. Big vague tasks give vague results.
- Read the commands it wants to run before you approve them.
What to expect from the first week
Expect the agent to be fast at routine work and shaky on big, vague tasks. Give it a small job first, such as adding a test. Watch how it reads your files and what commands it runs. Then raise the size of the tasks step by step. Trust grows from evidence, not from hope.
If you are not a programmer, you can still use these ideas. The same habits apply to any tool that acts for you. Small steps, clear notes, checks and backups keep you safe.
Teams often share one notes file in the project. That way every person gets the same rules and the agent behaves the same way for all of them. Update the notes in the same way you update any shared document, with a short review.
- ✓A coding agent reads, edits and tests code in your project from plain-language goals.
- ✓A notes file, such as CLAUDE.md, gives it standing rules.
- ✓Permissions, skills, hooks and sub-agents help you steer it.
- ✓Work in small steps on a copy, and always run the tests.
- ✓Read commands before approving them, and keep backups.
Write a project notes file
- Pick any project, even a small one.
- Write five short rules: style, tools, tests, risky areas and how to say done.
- Give the agent one small task and see if it follows the notes.
- Add one new rule for each mistake you saw.
Q1What does a CLAUDE.md file do?show answer
It holds standing rules the agent reads at the start of every session.
Q2Why ask for a plan first?show answer
You can fix a wrong direction before any files change.
Q3What should you keep before letting an agent run commands?show answer
A backup or a branch, so any damage can be undone.