Training
Certification Leadership Frameworks Agentic for Business
Community Keynotes Retreat Blog Book A Consultation
Protocol 18 of 18 · Track, Continuity · How it lasts

Working with human tokens after you leave

How to hand off agent context, memory, and project state to the next person, so the work survives a transition without a knowledge tax.

Protocol 17 AI testing vs human testing All 18
Why this matters

The pain it
solves

The whole retreat builds a system that runs on your taste, your judgment, and your eight agents. The question this protocol answers is the one that haunts every founder by month six: what happens when I am not the one running it?

Three reasons it matters: you will hire someone (they need to ramp without reading your mind), you will take a break (the work does not stop, you do), you will pivot (a project paused for six months loses all context, even from your own head).

The teaching

What this
actually is

The question every founder asks by month six

The whole retreat builds a system that runs on your taste, your judgment, and your eight agents. The question this protocol answers is: what happens when I am not the one running it?

Three reasons it matters. You will hire someone, and they need to ramp without reading your mind. You will take a break, and the work does not stop. You will pivot, and a project paused for six months loses all context, even from your own head.

The three layers of a real handoff

Every handoff has these three layers. Most teams only do the first one. That is why most handoffs fail at the boundary.

  • Project state
    What is shipped, what is in progress, what is blocked. Lives in project-status.html (Protocol 13). This layer is easy. Most teams stop here.
  • Agent state
    Every agent's role, context, and skills. Lives in the .claude/agents/ folder (Protocol 03). Hand over the folder, hand over the agent. No retraining required.
  • Tacit knowledge
    Why you made the calls you made. Which trade-offs were deliberate. Which workarounds are temporary and which are load-bearing. This is the layer most teams skip, and the one that makes the difference between a clean handoff and a smouldering crater.

The handoff doc, four sections

One file, docs/handoff.md, kept current as you go. Not written at the end (when the urgency to leave kills the quality of the doc), but maintained throughout.

  • Summary
    Three paragraphs. What this project is, where it is now, what it needs next. Read by anyone in 60 seconds.
  • Active decisions with rationale
    The five biggest in-flight calls. Each one a paragraph. Why we chose what we chose. What we considered. What we rejected.
  • Known landmines
    The things that broke before. The things that look fine but are not. The places to be careful. These are gold.
  • Next-person-should
    If someone fresh started tomorrow, exactly where they would begin. Not abstract advice; the concrete next move.

The discipline

Three rules. The third is the one that catches teams off-guard.

  • Write it as you go, not at the end
    Every epic close, update the handoff doc. Every active decision, document it in the moment. The doc maintained over six months is 10x better than the doc written in the last week.
  • Test the handoff with a fresh person (or a fresh Claude session)
    The doc that makes sense to you almost never makes sense to a fresh reader. Hand it to someone who has never seen the project. Watch them try to operate. The gaps become the next edit.
  • Update CLAUDE.md and AGENTS.md religiously
    These are the agent's version of the handoff doc. Every convention, every rule, every gotcha goes here. A line: code is replaceable, context is not.
Try it yourself 30 minutes

Write your handoff doc and test it on a fresh Claude session in 30 minutes

Pick a real project of yours. Pretend you are leaving tomorrow. You will write the four-section handoff doc, then open a fresh Claude session and ask it to operate the project.

  1. Step 01
    Create docs/handoff.md with four headings

    Summary, Active decisions, Known landmines, Next-person-should. Skeleton first, content second.

  2. Step 02
    Fill the Summary

    Three paragraphs as described in the teaching. The hardest part is the second paragraph (where it is now). Be honest.

  3. Step 03
    Fill Active decisions

    List your five biggest in-flight decisions. For each: what we chose, why, what we considered, what we rejected. A paragraph each.

  4. Step 04
    Fill Known landmines

    List at least three. The deploy that almost killed you. The dependency that breaks under load. The customer who needs a human reply, not an automated one. Be specific.

  5. Step 05
    Test it on a fresh Claude session

    Open a new Claude Code session in the project. Ask: "Read docs/handoff.md and tell me what to do tomorrow." If it can answer, your handoff works. If it cannot, the gap is your next edit.

Outcome

A real docs/handoff.md committed to your project, tested against a fresh reader, and the habit set to update it weekly. Future-you and future-them inherit the full picture, not just the code.

Official resources

Straight from
the source

What you walk out with

By the end of this
protocol

At the retreat

You learn it by
doing it

You pretend you are leaving tomorrow and write the handoff. Then a fresh Claude session reads it and tells you exactly what you forgot. The gaps become the discipline.

Connects to

Other protocols this
compounds with

← Previous, Protocol 17