Establish the current baseline.
Identify the repository, the branch you actually work from, the state of uncommitted changes and the test command. Name the running application and its environment separately from the source code.
Before new work starts, record what already works and what is unfinished. A brief from yesterday may no longer describe the latest files. The current version gives the incoming workflow an anchor.
Create one shared handoff.
A useful handoff includes the goal, current architecture, accepted decisions, active work, known problems and how to verify the application. Existing Markdown documents and attachments can provide that context.
When several AI conversations contributed to the project, reconcile their conclusions. Five separate summaries may disagree about what is finished or who owns the next task. Resolve those conflicts before allocating new work.
Start with responsibility, not a duplicate org chart.
Give a responsible leader the reconciled baseline. Add coding and review responsibilities around the work that remains. An old conversation is not automatically a current assignment.
This avoids a coder reporting completion to a leader that never received the original request. New implementation should have a current brief and acceptance criteria, even when earlier research came from another tool.
A handoff is not a transfer of hidden model memory.
A repository, documents and exported conversation text can carry useful project information. They do not transplant a provider’s internal session state into another provider.
Keep enduring decisions in project context and send the relevant details to each task. That is easier to audit than assuming every new agent has inherited everything the previous AI knew.
A practical first handoff.
- Repository and working branch, including any local changes.
- Current goal and the next small accepted outcome.
- Decisions that must remain in force.
- Known defects and work already in progress.
- How to run the app and its checks.
- Service names and environments, without passwords or secret keys.
This is a project preparation guide. It does not claim a one-click import of private sessions from every AI service.