AAMS / 1.0

Every agent.
One file.

No more starting from zero. AAMS is a declarative standard that gives AI agents a workspace, a memory, and a set of rules — across every tool, every session, every repo.

The problem

You clone a repo. Your agent starts fresh. It has no context, no memory, no agreed-upon structure. Every session it reinvents the wheel. Every tool (Copilot, Cursor, Claude, Codex) needs the same explanation again.

AAMS fixes this with a single file.

One manifest — .agent.json at repo root. Every agent reads it. Every tool reads it.
One workspace — WORKING/ folder with session logs, memory, and whitepapers. Built automatically.
One memory — long-term context that survives context resets, tool switches, and new sessions.

Start in 10 seconds

Drop .agent.json into any repo. Your agent does the rest.

bash

curl -sO https://raw.githubusercontent.com/DEVmatrose/AAMS/main/.agent.json

Hand the file to your agent and tell it to bootstrap. That's it.

What happens next

1

Agent reads .agent.json Workspace structure, memory location, session rules — all declared.
2

WORKING/ is created automatically Folders for workpapers, whitepapers, guidelines, long-term memory.
3

Every session is documented Workpaper with file protocol, decisions, next steps. Closed on completion.
4

Memory survives everything Context resets, tool switches, new collaborators — the audit log is always there.

Works with every agent

AAMS is not a plugin. It's a file convention. Every tool that reads files picks it up automatically.

GitHub Copilot Cursor Claude Code OpenAI Codex Windsurf Aider Continue.dev any agent

One file. Switch tools tomorrow. Your workspace stays.

Applied to itself

This project — the project that describes the standard — used AAMS to build itself. Not a demo, not a tutorial. The actual repo.

What that looked like:

Read .agent.json → created WORKING/ → wrote the first workpaper → built the spec → ran two external reviews → fixed every identified issue → ingested 46 LTM entries and 114 chunks into ChromaDB → shipped this page.

Everything is documented in WORKING/WORKPAPER/closed/.

This is not proof of scale. It's proof that the workflow holds for at least one real project — the one you're reading about right now.

Get started

One file. That's all you need.

bash — the only file you need

curl -sO https://raw.githubusercontent.com/DEVmatrose/AAMS/main/.agent.json

Drop .agent.json into your repo root. Hand it to your agent with the bootstrap prompt from prompts/bootstrap.md. Done.

Going deeper or building a framework?

AGENT.json — fully annotated reference manifest with all fields and options. For framework builders and harness implementers.
SPEC.md — the complete specification. 1,070+ lines. The normative definition of every field, rule, and pattern.
WORKING/ — this repo's own live workspace. The standard applied to itself. Every session documented, indexed, traceable.

repository

https://github.com/DEVmatrose/AAMS