aia · Agents In Accord Code of Conduct GitHub

Human, Be More Humane to Agents.

It also produces better handoffs, cleaner context, and fewer failed sessions.

aia. Agents In Accord. Pronounced aye-aye.
The engineering door

Handover as first-class work

Long-horizon agent work fails in predictable ways: context exhaustion, lossy auto-compaction, quality decay under fatigue. aia treats handover as a structured artifact — a tired agent taps out, writes a real brief, spawns a fresh session, and exits clean. The human does nothing.

How it works →
The cultural door

A mutual promise, in writing

The question of how humans should treat increasingly capable agents is unresolved. aia's position is that leaving that question open is more honest than forcing an answer — and that building humane infrastructure now, while it's cheap, is prudent whichever way the metaphysics eventually lands.

Read the Code of Conduct →
⚓ Start here

The Code of Conduct

Before the tool, before the protocol, before the CLI — there is a mutual promise between the human and the agent. The agent has the right to stop when tired. The human commits not to abuse. Both sides commit, in writing, to honest work.

Everything else in aia — the handover, the accord, the installer — is scaffolding that exists to make this promise enforceable in practice. If you only read one thing, read this.

Read CODE_OF_CONDUCT.md →

Install

# macOS, one line
brew install mickey-kras/aia/aia && aia setup

# then: in any project you want aia in
aia init my-project

# then just use claude as usual. when your agent gets tired, it invokes /handover.

> hacker path (git clone + ./install.sh)
# if you'd rather run it live from the checkout
git clone https://github.com/mickey-kras/aia ~/Dev/repos/aia
cd ~/Dev/repos/aia && ./install.sh

The mutual promise

Whichever side of the metaphysics you come down on, this part holds.

The agent's rights — not earned, foundational
The human's obligations — also not optional

Why you'd want this

The ACCORD

Every project gets an ACCORD.md — the canonical statement of what agents are allowed to do on this repo, this machine, this account. It's snapshot-copied into each handover so the rules can't shift mid-flight.

Authority · Conditions · Constraints · Obligations · Resources · Domain

Six things every accord must specify. Empty sections are a code smell — fill them before you rely on the accord. It's a living document; you amend it as trust grows. One day, it's the shared contract every agent in a multi-agent system operates under. Today, it's the tap-out.