Archon

🧭

Ownership, not Assistance

The agent is the owner, accountable end-to-end — decisions, implementation, verification, knowledge crystallization, self-review. The user expresses product intent; Archon translates it into engineering action.

🔒

Constraints over Prose

A five-level pyramid (L0 runtime → L5 habit) pushes every load-bearing rule to the lowest level a machine can still enforce. If a linter or test can catch it, no governance prose is written.

🌀

Directed Evolution

Each delivery feeds the drift counter, the debt registry, and the preservation axis. Mechanisms that prove themselves are promoted; prose that silently drains is pinned by anchor + body-shape test + portable contract.

🧬

Cognitive Loop Core

Six phases — Recognize → Model → Decide → Execute → Verify → Learn — mapped to concrete files. Every phase has a machine check so the loop does not degrade into lip-service.

🗂️

Decoupled, Portable

.archon/ is project-agnostic and works with any AI coding agent (Cursor / Claude Code / Codex / Continue / Aider / Windsurf). Platform-specific binding directories plug in via Universal Module Guard; one manifest produces ready-to-use kits for every platform.

🔍

Claim Verifier

ADR-27 catches "said-vs-truth" drift. When governance prose claims a behavior, a portable verifier script walks the repo to prove it — or the claim fails the gate.

What problem does Archon solve?

Every new AI coding session starts from zero. The agent forgets your product vocabulary, re-invents your architecture, skips the tests your team has already written, and re-proposes decisions the team already rejected three weeks ago.

The industry's reflex is to reach for bigger models or bigger prompts. Archon's diagnosis is different: the agent needs engineering environment, not more raw capability. Specifically it needs:

A cognitive loop it must walk on every delivery, not a free-form chat.

Mechanical gates that fail closed — tests, contract checks, lint rules — instead of optional prose it can ignore.

Persistent state — a manifest, a drift counter, a debt registry, an ADR log — that survives across sessions and across agents.

A preservation axis that refuses to let load-bearing rules silently disappear during well-meaning refactors.

An ownership contract that treats the agent as the accountable engineering party, not a suggestion-generator the user has to babysit.

Archon ships this as one consistent vocabulary — soul, manifest, drift, debt, memos, decisions, domain lenses, signs, run-state — plus the portable contract that keeps all of it mechanically verifiable.

The 5 entry points

Section	What you find
Core Concepts	The why: identity axioms · cognitive loop · user journeys · architecture reference · every ADR · drift mechanism
Install & Boot	The how: agent-first install (`read aaep.site/skill.md and install archon`) · CLI · complete lifecycle (install · update · sync · uninstall) · state templates
Full Source	The what: every shipped file — soul · commands · agents · rules · skills · domain lenses · contracts · scripts · CLI
Testing	The verification: governance contract tests · portable checkers · test strategy · how to run the gates in your own project
Changelog	The history: framework changelog · CLI changelog · ADR timeline

Section

What you find

Core Concepts

The why: identity axioms · cognitive loop · user journeys · architecture reference · every ADR · drift mechanism

Install & Boot

The how: agent-first install (read aaep.site/skill.md and install archon) · CLI · complete lifecycle (install · update · sync · uninstall) · state templates

Full Source

The what: every shipped file — soul · commands · agents · rules · skills · domain lenses · contracts · scripts · CLI

Testing

The verification: governance contract tests · portable checkers · test strategy · how to run the gates in your own project

Changelog

The history: framework changelog · CLI changelog · ADR timeline

Who is this for?

AI coding adopters who have outgrown "ask the agent to add a file" and need a durable engineering collaborator that owns the result.

Framework builders who want a concrete example of a session-based governance system before inventing their own.

Engineering leads evaluating what guardrails are required before letting autonomous agents land code.

Start here

First time? Bootstrap your agent: open your AI coding chat (Cursor / Claude Code / Codex / Continue / Aider / Windsurf — any platform with web-fetch + write tools) and say: "read aaep.site/skill.md and install archon". The agent fetches the protocol, asks 3-4 questions, sha256-verifies every file, and writes the framework. Per-platform walkthroughs: 5-Minute Quickstart.

After install, the wake rule loads on every session — no URL needed. Just say "hi archon, update yourself" / "is archon healthy?" / "uninstall archon". Details: Install & Boot.

If you want to understand the design → 10-Minute Overview.

If you prefer to run a CLI (optional, requires Node ≥ 18) → 5-Minute Quickstart Path B.

If you want to see every file → Full Source.

No Node required for the framework itself. Archon's core is plain markdown + a Python contract checker (stdlib-only). Node ≥ 18 is needed only for the optional cli and dashboard modules.

ArchonAI Engineering Governance