Resume a coding project without starting from scratch.

Start with the last recorded outcome.

Open your project and look at Context for recorded goals, sessions and outcomes. Then check Queues for active work. A saved task describes intended work; its presence does not mean an agent has executed it.

If the project is empty, connect your agent and publish the first map. If no session history was recorded, ask the agent to inspect the current project rather than guessing what happened last time.

Compare the snapshot with today's project.

Change Review shows recorded updates and published graph comparisons. Use Files and Project Graph to locate the area you want to change, then have your agent check the current source and tests with its existing local access.

If you use GitHub repository tracking, a newer tracked commit can mark the map as needing a refresh. Tracking metadata does not rebuild the map. Queue a map refresh or ask your authorized agent to publish an updated map.

Choose one task with a clear result.

For example, if a report identifies a failing validation check, review its evidence, queue the specific fix, and ask your connected agent to work on that task in the selected project.

Use Brain Scanner for [project name and ID]. Read the recorded context, recent changes and active queue. Check their freshness against the current project using your existing access. Summarize what is known and what is missing, then propose one next task with a concrete check for completion. Wait for me to choose the task before editing files.

You can edit or cancel queued work before it runs. A queue item is not a new grant of repository access; the agent still uses its own permissions.

Hand off from Claude Code to Codex.

Before leaving Claude Code, ask it to finish its Brain Scanner session with the result, checks and blockers. In Codex, connect Brain Scanner using the supported setup, sign in to the same account and select the same project ID. Give Codex access to the local checkout separately.

• Read the last recorded session outcome and recent Change Review entries.
• Check the active queue for unfinished or already claimed work.
• Verify the current files and relevant tests before relying on an earlier conclusion.

Continue project [name and ID] in Brain Scanner. The local project is [path]. Read the latest recorded outcome and queue task [task]. Check whether the current project still matches that context. Summarize the remaining work and any missing access before continuing the task I select.

The handoff uses recorded project context. A new agent does not automatically inherit another client's conversation or local permissions.

Take a short decision trail with you.

In your project's Context tab, choose Sessions and expand the session you want to continue. Choose Copy summary or Download Markdown. The handoff includes its recorded goal, changes, reported checks, decisions, blockers and outcome. Missing sections say “Not recorded.”

Review the text before pasting it into another agent or sending it to a teammate. The export omits source diffs and full conversations; it is not a secret detector. Only share project details with people and services you trust.

For example, a synthetic task-app handoff might say: “Blank titles are now rejected; the validation test passes; keyboard error focus still needs checking.” Give the receiving agent that remaining check, rather than asking it to repeat the whole fix.

Leave a useful handoff for the next visit.

After the work, ask the agent to record what changed, the checks it actually ran, any failures and what remains unfinished. Review that result in Reports, Change Review or the task's queue history. Refresh the project map when its structure has changed.

Review the recorded changes and checks before closing a task. See a sample outcome with a clear next step.

Keep the next task small enough to verify. For a concrete example, see how to find a shared helper's callers and relevant tests.

Continue in your dashboard