AI systems · Operating protocol

AHP+ — Agent Handoff Protocol Plus

I turn AI workflows into operable, auditable systems: verifiable memory in Git, cross-platform continuity, and handoffs that don’t rely on remembering the chat.

Contribution: Creator and architect of AHP+. I defined the concept, specification, certainty model, records, CLI, Pangea OS integration, and authorship narrative for CV and portfolio use.
Focus: Node.js · JSON · GitHub

Agent Handoff Protocol Plus

01 / The challenge

Understand the business before designing the solution.

Teams can use Codex, Cursor, Claude Code, OpenCode, or ChatGPT to build software, but continuity breaks when context lives only in conversations, summaries, or private provider memory.

Design an operating plane that turns memory, decisions, QA, risks, and handoffs into versioned repository data without replacing human authority or Git controls.

02 / The approach

A clear direction for product, experience, and operations.

01
Treat the repository as the source of truth and `/agent` as canonical memory, not as a folder of free-form notes.
02
Separate verified facts, user confirmations, inferences, stale state, and conflicts to reduce operational hallucination.
03
Turn every material session into state, evidence, QA, and a reproducible next action so another agent can continue.

03 / What I delivered

From product judgment to a solution the team can use.

AHP+ 1.0 specification covering canonical source, certainty, records, states, evidence, handoff, locks, and conformance.
Node.js CLI runtime to verify, inspect, record evidence, generate briefs, create handoffs, and validate structure.
Pangea OS integration for operating multiple web projects with active state, QA receipts, and cross-agent continuity.
Authorship and portfolio wording that explains the system in a human, marketable, and honest way.

04 / Signals

Evidence translated into visual reading.

01
Web project
Code, assets, tests, and documentation live in the repo.
02
/agent
AHP+ stores current state, evidence, QA, risks, and handoffs.
03
Next agent
Before continuing, it verifies the repository and reads canonical state.

The protocol records operational context; it does not replace human review.

01
VERIFIED
Can govern execution because it was observed through a file, command, or primary source.
02
USER_CONFIRMED
Can govern execution when the owner explicitly confirms it.
03
INFERRED
Works as a hypothesis, but is not presented as fact.
04
CONFLICTED
Blocks dependent decisions until the contradiction is resolved.

05 / Outcome

What the project put in place.

An operating continuity system for AI-assisted web projects, with portable and verifiable memory.
A clear model for AI assistance with boundaries: evidence over confidence, Git over recollection, and humans over automation.

Principles applied