How I Use Claude Code Every Day: Plan, Act, Verify
Over the last few posts we built up the pieces — CLAUDE.md, skills, context, tools, subagents, and hooks. Today they come together. This is how I actually use Claude Code every day: the plan, act, verify loop, the permission modes that decide how much Claude does on its own, and the session controls that let me stop, resume, and undo without fear.
Plan, Act, Verify — the daily loop
Almost every task follows the same three beats. First Claude plans: it explores the code and tells me how it would approach the change. Then it acts: it edits the files. Then we verify: it runs the tests or the build to confirm the change works. The official docs show this same shape in recipe after recipe — for a refactor you find the old code, apply the change, then run the tests. Keep this loop in mind and you never trust a change you have not checked.
Plan mode: look before you leap
For anything bigger than a one-line fix, I start in plan mode. The docs put it simply: Claude reads files and explores, then writes a plan, but it does not edit your source. Edits stay blocked until you approve the plan. Enter it by pressing Shift+Tab or by starting the prompt with /plan.
# start a session in plan mode
claude --permission-mode plan
# or press Shift+Tab mid-session to cycle into it
# the status bar shows: plan mode onPermission modes: how long a leash
- default (Manual) — Claude reads freely but asks before it edits a file or runs a command. Best for sensitive work.
- acceptEdits — Claude creates and edits files without asking, plus common filesystem commands. Best when you review changes as they land.
- plan — Claude explores and proposes, but edits nothing until you approve. Cycle all three with Shift+Tab.
Sessions: resume and rewind
# continue the most recent session in this folder
claude -c # same as: claude --continue
# open a picker to choose an older session
claude -r # same as: claude --resume
# undo a wrong turn (or press Esc twice when the input is empty)
/rewindClaude Code saves every conversation locally, so a task can span many sittings. Rewind is the other half: it captures your code before each prompt, and the rewind menu lets you restore your code, your conversation, or both. One caution from the docs — rewind tracks direct file edits, not files changed by bash commands, so keep using git for permanent history.
Next post: the finale — how the whole agentic loop ties together.