Put a coding agent to work on a codebase you already have, and keep the output reviewable — project setup, permission boundaries, task scoping and the review discipline that makes it safe to merge.
Two articles, about ninety minutes including the setup work.
The first is mechanical: what to put in a project instructions file, which commands to pre-approve, and how to give the agent a compiler it can run without asking you. The second is the part that decides whether any of it is worth doing — how large a task to hand over, and what to check before merging.
Neither article argues that agents write good code on their own. Both are written around the assumption that a human reads every diff, and are structured to make that reading realistic rather than theatrical.
Read in order. Ticking a lesson stores your progress in this browser only.
Add Claude Code to a TypeScript repo you already have: a CLAUDE.md that matches your real build, permission rules that block dangerous commands, and a typecheck loop the agent runs itself.
Project instructions, permission rules and a typecheck loop the agent runs itself.
A working loop for running a coding agent on a real codebase — task scoping, branch isolation, tests as the contract, and the review discipline that stops unreviewed code reaching main.
Task scoping, branch isolation and how to review what comes back.
Add Claude Code to a TypeScript repo you already have: a CLAUDE.md that matches your real build, permission rules that block dangerous commands, and a typecheck loop the agent runs itself.
A working loop for running a coding agent on a real codebase — task scoping, branch isolation, tests as the contract, and the review discipline that stops unreviewed code reaching main.