Switching AI Tools Mid-Project Should Not Mean Starting Over

  • You've spent 20 minutes building context in Gemini. Research done, decisions made, draft outlined. Now you need Claude for the next step.
  • You paste a vague summary, Claude gets confused, and you spend the next 15 minutes re-explaining what Gemini already understood.
  • A micro-handoff fixes that. You move only what Claude needs, in a format it can use immediately.

5 Prompts. Use Them in Order.

A micro-handoff is a structured method for transferring AI session context from one tool to another using a 4-part brief. It covers project context, decisions made, current output, and the next task. Here is the full method.

STEP 1: Ask Gemini to summarise your session

Use in: Gemini (send as your last message in the thread)

Prompt:

Summarise our session in exactly 4 parts and label each one:

1. WORKING ON: What project or task we were focused on.
2. DECISIONS MADE: Any choices, directions, or constraints we agreed on.
3. OUTPUT PRODUCED: The main content, draft, or result from this session.
4. NEXT STEP: What still needs to be done after this session ends.

Keep each part to 3–5 lines. Be specific. No filler.

What you should get back:

  • A clear 1–2 line description of the project
  • A short list of decisions made in the session
  • A description of the draft, plan, or content that was created
  • One or two clear actions still pending

Action: Send this at the end of your Gemini session. Copy the full reply. Keep the tab open.

Note: If Gemini gives a long response, that is fine. You will trim it in Step 2. Right now just get the structured output.

STEP 2: Ask Gemini to compress it for handoff

Use in: Gemini (follow-up in the same thread)

Prompt:

Now compress that summary into a handoff brief. Format it exactly like this:

PROJECT: [one line]
DECIDED: [bullet list, one item per line]
OUTPUT: [2–3 lines describing what was produced]
NEXT: [one clear instruction]

Nothing else. No preamble. No closing line.

What you should get back:

  • PROJECT: Single line describing the task
  • DECIDED: Bullet list of key decisions
  • OUTPUT: Short description of the work produced
  • NEXT: One actionable instruction

Action: Copy this compressed reply. This is the only thing you paste into Claude.

Note: Do not paste the full Step 1 summary. Just this compressed version.

STEP 3: Open Claude and send the context brief

Use in: Claude (open a new chat)

Prompt:

I am handing off work from another AI session. Read this brief and confirm you understand it before I give you a task.

--- HANDOFF BRIEF ---
[Paste your Step 2 output here]
--- END BRIEF ---

Do not do anything yet. Just confirm you have read the brief and tell me back: the project, the decisions already made, and the next step I listed.

What you should get back:

  • Claude repeats the project in its own words
  • Claude lists the decisions already made
  • Claude states the next step you wrote
  • Claude does NOT start doing the task yet

Action: Open a new Claude chat. Paste this prompt with the Step 2 brief inside it. Send. Read Claude's confirmation before continuing.

Note: Always start a new Claude session for a handoff. Pasting into an existing thread mixes old context. A clean session means the brief is the only context Claude has.

STEP 4: Correct anything Claude got wrong

Use in: Claude (same chat)

Prompt:

One correction before we continue:

[Write your correction here. Example: "The tone is clinical-warm, not neutral. The audience is procurement leads, not marketing teams. Everything else is correct."]

After reading this, confirm the updated understanding in the same format as before.

What you should get back:

  • Claude acknowledges the correction
  • Claude restates the corrected detail accurately
  • Claude confirms it is ready for the task

Action: If Step 3 was accurate, skip this step. If Claude missed or misread anything, send this correction prompt before moving to Step 5.

Note: Be direct and specific. "The tone is clinical-warm" is more useful than "you got the tone wrong". Claude needs the correct value, not just the error.

STEP 5: Give Claude its first task

Use in: Claude (same chat)

Prompt:

Using the project context from the handoff brief, [describe the task clearly].

Constraints:
- [Any tone, length, or format rules]
- Keep the decisions already made. Do not re-open them.

Output format: [Describe what you want back. E.g. a short draft, a list of options, a revised version]

What you should get back:

  • Claude produces output that fits the project context
  • Claude does not re-ask questions already answered in the brief
  • Claude follows the format you specified

Action: Replace each placeholder with your actual task, constraints, and output format. Send.

Note: The phrase "keep the decisions already made, do not re-open them" is important. Without it, Claude may suggest alternatives to things you already decided in Gemini.

How to Use Summary

  • Step 1. End your Gemini session with the Step 1 summary prompt.
  • Step 2. Follow up with the Step 2 compression prompt. Copy the output.
  • Step 3. Open a new Claude chat. Paste the Step 3 framing prompt with the brief inside it.
  • Step 4. Correct any mistakes Claude made in its confirmation.
  • Step 5. Send your task using the Step 5 format.
Harshal Saraf

Harshal Saraf

Creative Director + Orchestrates AI Workflows

Helping founders and agencies work smarter. As a Creative Director, he builds brand identities and orchestrates AI workflows for businesses. He also writes about productivity, the YourLife OS framework, and publishes Oh So AI, a weekly newsletter.