Explore → Plan → Implement

The Problem

Most agentic coding failures happen in the first ten seconds — when Claude jumps straight to writing code before it understands the codebase. It solves the wrong problem, edits the wrong file, or misses a pattern your team has already established. The explore-first loop fixes that.

Phase 1: Explore

Open Plan Mode (Shift+Tab twice) and ask Claude to read relevant files without touching anything:

1read /src/auth and understand how sessions and login work.
2also check how we handle environment variables for secrets.

Claude explores without making changes. You see what it finds before it acts.

Phase 2: Plan

Still in Plan Mode, ask for a concrete implementation plan:

1I want to add Google OAuth. What files need to change?
2What's the session flow? Write a step-by-step plan.

Press Ctrl+G to open the plan in your editor and edit it directly before Claude proceeds. This is the moment to catch wrong assumptions cheaply.

Phase 3: Implement

Switch to Normal Mode and give Claude the green light — with a verification hook:

1implement the OAuth flow from your plan. write tests for the
2callback handler, run the test suite, and fix any failures.

Claude writes, tests, and iterates. You review output, not keystrokes.

Phase 4: Commit

When to Skip Planning

Planning adds overhead. Skip it for tasks where you could describe the diff in one sentence — a typo fix, a log line, renaming a variable. Use it when the change touches multiple files, the scope isn't clear upfront, or you're unfamiliar with the affected code.

Source

• Common Workflows – Claude Code Docscode.claude.com (opens in a new tab)