Resources Handoff
Free Claude skillHand your work to a fresh session without re-explaining it
Handoff turns an in-progress conversation into a structured brief someone with zero context can act on. The decisions you made, the things you ruled out, the approaches that already failed and why.
Free, just your email.
What it does
- Writes for zero shared context. No "as we discussed," no dangling pronouns, no assumed background. Every line stands on its own.
- Records the dead ends. What was tried, the root cause it failed, what to do instead. This is the part that stops the next session burning an hour repeating you.
- Separates decisions from defaults. Only choices that were actually debated or iterated on get written down, each one with its reason attached.
- Marks gaps instead of filling them. Anything it doesn't know gets labelled Unknown. A stated gap is useful. An invented fact costs you a day.
- Saves to a file, not the chat. A handoff exists to be shared, so it comes back as a named markdown file you can drop straight into the next session.
The structure it produces
- Goal
- One sentence on what the work is for.
- State
- What exists right now — exact file paths, what's built, what's tested and what isn't.
- Decisions
- Each choice with the reason behind it.
- Constraints
- Hard requirements and explicit nos.
- Dead ends
- What failed, why it failed, what to try instead.
- Next steps
- Ordered, concrete actions.
- Access needed
- Tools, files and permissions — named, never their values.
Empty sections get dropped rather than left as a bare header. Secrets never make it in: the skill names the access required and omits the keys.
Summary vs. handoff
Ask any model to "summarise where we got to" and you get narrative — readable, and useless to someone who wasn't there. Here's the same two hours of work written both ways.
A normal summary
We spent some time working on the client onboarding automation and made good progress. We looked at a few different approaches for handling the intake form and settled on one that seems to work well. There were some issues with the API that we worked around. The next step is to finish the remaining pieces and test everything end to end before handing it over to the client.
The same work, as a handoff
Goal
Automate client intake so a signed proposal creates the project workspace without anyone touching it.
State
Form capture and workspace creation are built and working end to end in intake/main.py. Notification step is written but untested — no test has ever run against a real webhook.
Decisions
- Polling the form API on a 5-minute timer instead of webhooks — their webhook signing is undocumented and support hasn't replied.
- One workspace per signed proposal, not per client — clients come back for repeat projects and were colliding on the old scheme.
Constraints
- Client is on the legacy plan. No API write access, read-only.
- Must run on the existing scheduler. No new infrastructure.
Dead ends
- Webhook subscription — the signature check fails because their docs describe v1 signing and the endpoint sends v2. Not a config problem. Don't retry until support confirms the version.
Next steps
- Test the notification step against a real submission.
- Add a retry on the workspace creation call — it has failed twice on timeout.
Access needed
Form API read token, workspace admin account.
Everything above is actionable. Nothing above needs a follow-up question.
When to reach for it
- A session is getting long and you'd rather start clean than keep paying for the whole history.
- You're switching tools or models mid-project and the context doesn't travel with you.
- Someone else is picking it up — a contractor, a teammate, or you in three weeks, which is functionally the same person.
- You're parking the work and want the reasoning captured while you still remember it.
What you need
Claude with skills enabled — Claude.ai, Claude Desktop, or Claude Code.
Work that's actually in progress. The skill reads the real record of what happened. It won't manufacture a project history that doesn't exist.
Chat search helps, but isn't required. Inside a Claude Project it pulls the actual conversation history rather than working from memory, which makes the brief noticeably sharper.
How to install
Download handoff.skill above.
Open Claude settings. Go to Skills → Add → Upload a skill.
Select the file you downloaded.
Say "/handoff" at the end of a working session and it will trigger the skill.
Stop re-explaining your own project
Drop the .skill file into Claude, ask for a handoff, get a briefing file back that the next session can actually use.
Free, just your email.