Owned product · Open source

AHP+ — Agent Handoff Protocol Plus

AHP+ keeps a record inside the project itself of what was done, what was checked, and what is still pending. So when the work moves to another AI assistant, another program, or another computer, nobody has to guess where it left off.

Contribution: Creator and architect of AHP+. I defined how it works, wrote the 1.1 specification, the labelling system that separates checked facts from assumptions, the record format, the terminal tool published on npm, the installers for each platform, and the limits on what it never does on its own.
Focus: Node.js · JSON · npm

Agent Handoff Protocol Plus

01 / The challenge

Understand the business before designing the solution.

One person can now work with several AI assistants: Codex, Cursor, Claude Code, OpenCode, or ChatGPT. Each remembers the conversation its own way, and that memory cannot be checked and does not travel with the project. Switch tools, accounts, or computers and the thread is gone. AHP+ 1.1.0 fixes that by storing the thread inside the repository, next to the code.

Turn that thread into plain files anyone can read and check: where the project stands, which decisions still hold, what was actually tested, and what comes next — without taking control away from the person or replacing Git.

02 / The approach

A clear direction for product, experience, and operations.

01
Every project owns its own memory: AHP+ creates an .ahp/ folder inside the repository and keeps that project’s real state there, never mixed with another.
02
Every entry states how much it can be trusted: checked with a tool, confirmed by a person, only an assumption, already out of date, or contradicted by another source.
03
Handovers between tools get reviewed before they are accepted: the receiver compares project, branch, and commit, and only continues if everything matches.

03 / What I delivered

From product judgment to a solution the team can use.

The open AHP+ 1.1 specification: the rules of the format, written so anyone can implement them.
A terminal tool published on npm as @jossuealcala/ahp-plus, with no third-party dependencies and an Apache-2.0 license.
Installers for Cursor, OpenCode, Codex, Claude Code, ChatGPT, and other agents: each platform asks in its own way, but the meaning stays the same.
Documentation in Spanish and English, plus automated tests on Ubuntu, macOS, and Windows with Node.js 20 and 22.

04 / Signals

Evidence translated into visual reading.

01
Your project
The code, images, tests, and documentation live in the repository.
02
.ahp/ folder
Here AHP+ records where the project stands, what was tested, what the risks are, and what comes next.
03
The next tool
Before continuing, it reads those records and confirms they match the repository.

AHP+ keeps a record of the context; review by a person is still required.

01
VERIFIED
Checked against a file, a command, or a direct source. Safe to decide on.
02
USER_CONFIRMED
The person in charge confirmed it. Also safe to decide on.
03
INFERRED
A reasonable assumption: usable as a hypothesis, never as fact.
04
CONFLICTED
Two sources disagree. It halts any decision that depends on it.

05 / Outcome

What the project put in place.

AHP+ 1.1.0 is published and anyone can install it from npm or GitHub.
It was tested end to end: clean install, all six installers, and a real Codex-to-Cursor handover accepted in another project.
It has a deliberate limit: AHP+ records and checks, but never commits, pushes, merges, or deploys on its own. A person authorizes that.

Principles applied

The project is the source of truth, not the chat.
Nothing counts as done without proof.
A person authorizes what matters.

06 / Command guide

Install and operate AHP+ on any platform.

A step-by-step guide: how to install it on macOS, Linux, or Windows, the 40 ready-to-copy commands, and what to type in the chat of Cursor, OpenCode, Codex, Claude Code, or ChatGPT.

Open Command Atlas

07 / Technology