Put Claude Code Sessions Online

How to choose between local resume, Remote Control, and Claude Code on the web so your AI work keeps running even when you step away from your computer.

The short version

There are three different things people mean when they say “put my Claude Code session online”:

  1. Resume a local session later. Good for your own computer. Bad if your laptop is closed, lost, or out of battery.
  2. Control your local Claude Code from another device. Good when your machine stays awake and online.
  3. Run Claude Code from the web against a GitHub repo. Good when you want work to continue without depending on your laptop.

Those are different workflows. The mistake is treating them like one feature.

If the work must survive you stepping away from your computer, make it a remote task against a repo. If the work depends on local files, local secrets, or a running local app, keep it local.


Choose the right surface

Local vs online Claude Code
Criterion Local session Terminal or app Remote control Local machine stays on Web task Best pick Runs away from your laptop
Needs your laptop to stay awake Yes Yes No
Good for repo documentation
Good for long unattended work Only if machine stays online
Good for local app visual QA
Good for secret-heavy setup Maybe Maybe Avoid

Use the web task when the job can be described as:

  • Read this repo
  • Make this change
  • Open a pull request
  • Run the existing checks
  • Leave me a summary

Do not use it as the first place you test a messy, unknown project. The cloud surface is powerful when the repo is ready. It is frustrating when the repo has no setup notes, no commands, and no project context.


Step 1: Make the repo understandable

Before you put a session online, give the agent the same orientation you would give a teammate.

At minimum, your repo should have:

  • README.md with what the project is and how to run it
  • CLAUDE.md with project rules, design-system rules, and permissions
  • tools.md with framework, package manager, commands, testing, and deployment facts
  • A working default branch
  • A clean enough build or test command that failures mean something

The most important part is the split:

  • CLAUDE.md tells Claude how to behave in this project.
  • tools.md tells Claude what tools and commands exist in this project.

If those two files are missing, the online task starts by guessing. That is where drift begins.


Step 2: Start with a safe remote task

Your first online task should be boring. Boring is how you test the loop.

Good first tasks:

  • “Read the repo and add missing setup notes to README.md.”
  • “Audit token naming against CLAUDE.md and return a report. Do not edit files.”
  • “Add alt text to the three image examples in this guide.”
  • “Find broken internal links in the guides folder and open a PR with fixes.”

Bad first tasks:

  • “Set up authentication.”
  • “Refactor the whole component library.”
  • “Deploy this to production.”
  • “Rotate API keys.”
A safe first online task
  1. 01

    Pick repo

    Use a repo with README, CLAUDE.md, and tools.md.

  2. 02

    Scope task

    One folder, one outcome, one verification command.

  3. 03

    Run online

    Let Claude work remotely instead of relying on your laptop.

  4. 04

    Review PR

    Read the diff, run checks locally if needed, then merge.


Step 3: Write the prompt like a handoff

Use this structure:

You are working in [repo name].

Goal:
[one sentence]

Read first:
- ./CLAUDE.md
- ./tools.md
- [any specific file or folder]

Scope:
- You may edit [files/folders]
- Do not edit [files/folders]

Verification:
- Run [command]
- If the command fails, explain whether it failed before or after your change

Output:
- Open a PR
- Include a short summary, changed files, and verification result

This is not ceremony. It is how you keep cloud work from turning into a vague autonomous wander.


Step 4: Decide what moves online next

After the first task works, move repeatable work online before creative work.

Good candidates:

  • Weekly documentation cleanup
  • Token naming audits
  • Broken-link checks
  • Screenshot inventory
  • Component README updates
  • Draft PRs for small guide edits

Keep these local until you have a stronger trust loop:

  • Secret handling
  • Billing and auth changes
  • Production deployments
  • Large visual redesigns
  • Anything that needs you to judge taste in the browser

The pattern is simple: cloud for mechanical work, local for context-heavy judgment.


Exercise

Move one safe Claude Code task online

25 min
  1. Prepare the repo handoff

    Pick one repo. Confirm it has a README.md, CLAUDE.md, and tools.md. If one is missing, create the smallest useful version before you start the online task.

    • The repo says how to install, run, and verify the project
    • Claude has project behavior rules in CLAUDE.md
    • Tool and command facts live in tools.md
  2. Run a low-risk web task

    Ask Claude Code on the web to do one documentation, audit, or small cleanup task. Keep the editable scope narrow. Ask for a PR or a clear report.

    • The task finishes without needing your local laptop
    • The diff is small enough to review in under 10 minutes
    • The summary includes what changed and how it was verified

Finished this lesson?

Mark it complete to track your progress through "Claude Code".

Lesson
3 / 12
Progress
25%
Read time
20 min
Free to try Cancel anytime
The guides alone saved me a full day of work every sprint.
Senior Design Systems Lead
Enterprise SaaS
Pro
Full access to everything.
$39 /month
  • All guides, prompts, and templates
  • Starter kits and templates
  • New content every week
  • Priority support