Coding AI

Install both CLIs, sign in, and fix one failing test in a scratch repo you can check with a single command.

How each tool loads its instruction file, one file that serves both, and a worked AGENTS.md for a Next.js app.

When to plan before editing, and how to cut a feature into tasks that each end in a check.

Read an agent's diff in the order that finds risk first, with a checklist to copy.

Branches, commits, worktrees, and PRs with an agent, and the git actions that stay with a human.

Turn the rules an agent keeps forgetting into checks that run every time.

Split research and review into separate contexts, and run sessions side by side without collisions.

Connect an agent to context that lives outside the repo, and decide which servers earn a place.

Package a workflow you repeat into instructions the agent loads only when it needs them.

Run an agent without a human in the loop, in scripts and CI, with permissions to match.

The three tools compared on the same terms once you know how each one works.

A terminal agent against an in-editor assistant, and which work each one suits.