Every fresh Claude Code session starts your project at zero. Summaries and /compact
keep your decisions but quietly drop the most expensive thing you learned: the
approaches that didn't work. This kit ends a session with /handoff, writes a
lean verified checkpoint, and teaches the next session to distrust it until it
checks your repo's live state.
1. End your session with /handoff (or paste the zero-install prompt from
PROMPT.md). It writes ~/.claude/handoffs/<project>.md — goal,
exact stop point, verification evidence, next action, and the dead ends — around
30 lines, never a transcript.
2. Open a fresh session and paste one line:
Read ~/.claude/handoffs/<project>.md and continue
3. The fresh session verifies before it touches anything. The handoff header
records branch, HEAD, and dirty counts. The resume ritual compares them against
live git status and reports drift BEFORE editing a file. This is real output
from our test run — the repo had moved, and the session stopped and said so:
No stale-context edits, no confidently wrong resumes. A mismatch means stop and look.
Skill install — copy the skill folder, then end any session with /handoff:
cp -r kit/skill/handoff ~/.claude/skills/handoffNo install — copy the prompt from kit/PROMPT.md and paste it at the end of a session. Works in any harness that can write files.
| File | What it is |
|---|---|
kit/START-HERE.html |
Open this first: install, use, resume, uninstall |
kit/PROMPT.md |
Zero-install paste prompt |
kit/skill/handoff/ |
Self-contained Claude Code skill (/handoff) |
kit/docs/POLICY.md |
The rules that make handoffs actually work |
kit/docs/COMPATIBILITY.md |
What's supported, known limits |
kit/docs/EXAMPLE-HANDOFF.md |
A real generated handoff |
The skill's rules bar secrets, tokens, credentials, private URLs, customer data, and copied file contents from every handoff. No telemetry, no network calls.
The kit is free — grab it on Gumroad to get fixes and new versions by email:
expressive446.gumroad.com/l/handoff-kit
If this kit saves you a re-explain, the paid playbooks go deeper:
| Product | What it does | Price |
|---|---|---|
| Claude Code Setup Playbook | The 3 hooks that stop fake "done", wrong models & wasted tokens | $29 |
| Agent Memory Kit | Give your AI agent a memory that learns who you are | $29 |
| Claude Code Token Tuner | Cut your token bill without losing quality | $19 |
| AI Scout Pipeline | Auto-track the latest AI papers & tools | $19 |
| Loop & Goal Playbook | Make "done" a check that exits 0 | $12 |
| Token Conservation Playbook | 23 ways to cut your Claude Code bill | $12 |
| Reliability Kit | Free CLAUDE.md contract + verify-before-done hook | Free |
| Super Planner Skill | Planning skill for Claude Code | Free |
MIT — see kit/LICENSE.



