Get up to speed on an unfamiliar codebase
Inheriting someone's project, or returning to your own repo from six months ago: have the structure and call graph explained before you touch anything.
What you end up with
- A queryable map of the project so you know where things live
- Answers like 'who calls this function' instead of grep-and-guess
- Your first change lands with tests behind it
The sequence
1. Index the structure before letting AI edit anything
Ask an assistant to edit without context and it only sees the files you pasted; the change won't fit the rest of the project. Turning the whole repo into a queryable structure is the prerequisite for everything after.
2. Make the first change with a terminal coding agent
Pick an agent that reads the repo, runs tests and sees failures. Deliberately make the first change small — a bug fix or a new flag — to test whether it actually understands this project.
opencodeOpen source★ 219.6KAn open-source AI coding agent that automatically writes code, fixes bugs, and runs tests in the terminal to help you complete coding tasks.claude-codeOpen source★ 152.7KUse natural language in the terminal to command AI to write code, fix bugs, and handle Git, improving development efficiency.3. Give the agent your project's rules
Write your style, directory conventions and prohibitions into a file the agent always reads. This is the difference between occasional lucky output and consistently following your rules.
Only need one piece of this?
You don't have to build the whole chain. Every tool above is listed on its own, with a summary, pros and cons, and a starting point.