Continue your work in a new AI chat, in one paste.
Session Handover writes a handover at the end of a long chat. You paste the handover into a new chat. The AI then continues your work. It keeps the decisions you made, and it does not repeat what already failed.
This handover comes from a chat about a team offsite. The people and places are invented. Your handover uses the same headings in the same order.
### Instructions for the new chat
> You are continuing work from a previous session. Before you act:
> 1. Restate the goal and the next action in two sentences.
> 2. List anything in this handover that conflicts with what you can see now.
> 3. Ask me to confirm. Then start.
> Treat [USER-DECIDED] items as fixed unless I reopen them. Treat [AI-PROPOSED], [ASSUMED], and [UNVERIFIED] items as open.
### 1. Goal and finish line
- Goal: Plan a two-day offsite for a 12-person design team.
- Finished when: the venue is booked and the agenda is sent to the team.
### 2. Current state
- Budget is $9,000 for venue, food, and travel. Sam confirmed it in the chat.
- A shortlist of three venues is drafted. Nobody has checked their dates yet.
- [ASSUMED] The team can travel on a weekday.
### 3. Decisions
| Decision | Tag | Reason |
| --- | --- | --- |
| Stay within two hours' drive of the office | [USER-DECIDED] | Sam wants no flights in the budget. |
| End day two at 3 pm | [AI-PROPOSED] | People get home before dark. |
### 4. Do not repeat
| Attempt | Why it failed | Evidence |
| --- | --- | --- |
| Lakeside Lodge | It sleeps 10. The team is 12. | The venue's capacity page |
### 5. Open items
1. [OPEN] Which week suits everyone? Sam must answer. Blocks the booking.
2. [OPEN] Does anyone need a special diet? Sam asks the team. Blocks the catering order.
### 6. Next action
Email the three shortlisted venues for prices and dates in the chosen week. Input: the shortlist in References. Expected result: a reply from each venue. Depends on open item 1.
### 7. User rules and preferences
- "Keep the agenda light. No more than three sessions a day."
- "No team-building games."
### 8. References
| Name or path | What it contains | Current or outdated |
| --- | --- | --- |
| venue-shortlist.xlsx | Three venues, with capacity and price | Current |
| [UNVERIFIED] Hilltop House rules | Says outside catering is allowed | Not checked |
### Check before use
- [AI-PROPOSED] End day two at 3 pm.
- [ASSUMED] The team can travel on a weekday.
- [UNVERIFIED] Hilltop House allows outside catering.
Read Check before use first. It lists every item that you did not confirm.
What the new chat replies
This reply is an example. Your words will differ.
Goal: plan a two-day offsite for a 12-person design team. Next action: email the three shortlisted venues for prices and dates.
One conflict: I cannot see venue-shortlist.xlsx. Please attach it.
Two questions block the work: the week, and special diets. Do you confirm the goal and the next action?
Five tags
Each item shows how sure the AI is
You do not need to learn the tags. Read Check before use and correct what is wrong. The new chat treats your decisions as fixed. It treats every other item as open, and it asks you first.
Tag
Meaning
In “Check before use”
USER-DECIDED
You said it, or you approved it in clear words.
No
AI-PROPOSED
The AI suggested it. You did not confirm it.
Yes
ASSUMED
The AI treated it as true. Nobody confirmed it.
Yes
UNVERIFIED
The AI is not sure the chat said it.
Yes
OPEN
Nobody has decided it.
No. It is in “Open items.”
The handover never copies a password or an account number. It writes where the secret is stored.
Optional · Claude only · Works on the free plan
Install it once in Claude
A skill is a small instruction file that Claude loads when you need it. Install this skill once. After that, type write a handover at the end of any long chat. You do not paste the prompt again.
Claude Code picks it up in the session you already have open, unless the command above had to create the skills folder. Restart it in that one case. On Windows, run the command in Git Bash or WSL.
Install it on claude.ai using the steps in the first tab.
Start a new Cowork session. The skill is already there.
Cowork loads the skills your account has enabled, and it reads them when a session starts. A session you already had open will not see it.
Questions
Before you start
Why not ask the AI for a summary?
A summary lists the topics you discussed. It drops your reasons. The new chat then suggests the idea you already rejected, and it runs the fix that already failed.
Does it cost anything?
No. The tool is free and needs no signup. The skill works on the free Claude plan.
Claude has memory. Why do I need a handover?
Memory helps Claude recall facts about you. A handover covers one piece of work. It lists what you decided, what failed, and what is still open. You can read it, correct it, and paste it into any AI tool.
What if my chat is already full?
The AI needs one more message in the old chat to write the handover. If the chat accepts no more messages, the tool cannot run there. Write the handover before that point.
Does it copy my passwords?
No. It writes where the secret is stored, and it leaves the secret out.
Yes. Change it and share it. Keep the credit to Luminary AI.
For developers
The skill works in Claude on the web, in Claude Code, and in Cowork.
In a coding chat with a shell, it runs git status, git diff --stat, and your test command before it writes. It trusts the repository over its memory. With no shell, it tags every repository fact UNVERIFIED.
A coding chat gets one more section, “9. Coding only”: environment, branch, last commit, uncommitted changes, test counts, and known bugs.
It never copies an API key or a token. It writes the variable name or the storage location.
A short “sounds good” to a long proposal confirms the direction only. The details stay AI-PROPOSED.
Before it prints, it reads its own handover as the new chat would. Each question it cannot answer goes into “Open items.”
