Working explanations

AI agent handoff builder

Create a focused handoff for Claude Code, Codex, or another coding agent. Define the outcome, acceptance, scope, changes, and next step, then copy or download the brief.

Package the next step

Start from a review or teammate handoff example, then replace its placeholders. Required fields are marked with an asterisk.

Illustrative example: these are not your files or live task results. Replace the revision, evidence and owners before sharing.

One concrete outcome. What should the next agent accomplish?
An observable done-when check, plus who will review it.
Files or surfaces in scope, exclusions, and permitted actions.
Changed files and the branch or revision. Write “none” if work has not started.
Current state, one unresolved question, and the first next step.
For team handoffs, name the person or agent who agreed to take the next action and where to return the result.
Actual checks and results, decisions, failures, and the original work handle. Mark pending checks and confirm the recipient can open each reference.

Your handoff

Five core fields filled. For a team handoff, also name the next owner, attach evidence, and confirm access before sending.

The builder runs in this page. It does not submit these fields, start an agent, or send a message. Refreshing the page resets the fields.

What should an agent handoff contain?

A useful handoff tells the next agent what to do and how its result will be judged. It points to the current work rather than asking the recipient to reconstruct a long conversation. The five core fields come from OpenScout’s handoff practice.

  1. Goal: the concrete outcome the next session should achieve.
  2. Acceptance: the observable check that makes the result reviewable.
  3. Scope: the files, actions, and boundaries of the task.
  4. Touched: what actually changed, with a branch or revision when useful.
  5. Stopped: where work stands and the first next move.

How to use the brief

Verify it against the current working tree and attach links to the actual evidence. A filled field does not prove its claim. Keep “not run” or “not verified” explicit when a check is still pending.

Copy the Markdown into the next agent’s context or save it alongside the work. If you use Scout, a request that expects a reply or owned work uses ask. Keep the returned handle for continuity. See agents and collaboration.

The brief itself grants no execution permissions. The responsible person still decides which actions are authorized and whether the delivered work meets the acceptance check.

Can I hand off between Claude Code and Codex?

Yes. This output is plain Markdown, so it can travel between coding-agent tools. It transfers an explicit work brief; it does not move hidden model state or automatically migrate a live harness session. For a tracked review workflow, read Claude Code and Codex review.

How do I hand the task to a teammate?

Choose Load teammate example, then add the real repository and revision, the person who agreed to take the next action, and the evidence they can inspect. Include known failures and checks still pending. Confirm access to private links and inspect existing work before launching another task.

Ask the recipient to read back the revision, first action, missing input and return path. A brief does not transfer provider credentials or ownership of a live session. The teammate handoff guide explains the workflow.

For a reusable file, use the teammate-handoff template, review packet, or morning-triage worksheet. These are blank documents to fill with actual evidence.

What happens to the text I enter?

The builder processes fields in your browser and offers a local Markdown download. It does not submit these fields to an agent or a repository. The page does not persist your edits after a refresh. The site’s general data practices are described in the privacy policy.