The Method

Lytos defines the framework in which AI coding agents operate, rather than trying to make them play roles. The method has five pillars, each materialised as a folder of plain markdown in your repository. Every pillar answers a specific question an agent has when it opens your project.

The five pillars

Pillar	Folder	What it tells the agent
Manifest	manifest.md	Who we are, what we’re building, and what principles guide our decisions
Skills	skills/	How to perform recurring tasks — code review, testing, deployment, etc. — in the agentskills.io open format
Rules	rules/	The non-negotiable quality criteria the agent must respect before handing back code
Issue board	issue-board/	The kanban flow, with every issue a markdown file whose YAML frontmatter is the source of truth
Memory	memory/	What the team has learned — architectural decisions, patterns, past bugs — so sessions don’t start blank

Two companion pieces

Orchestrator	A light coordination model for multi-issue sprints — who picks up what, how dependencies are respected, when to merge
Sub-agents	How Lytos handles the sub-agent question — the short answer is: one well-contextualised agent usually beats a cast of personas

Works with your existing tool

Lytos doesn’t replace your AI assistant — it configures your project so any compatible tool reads the same context. Compatibility lists the current integrations: Claude Code, Cursor, Codex, with Copilot, Gemini CLI and Windsurf coming next.

Where to start

• If you’re new to the method, read the Manifest first — it’s the entry point every session opens.
• If you want to understand the “why”, the Philosophy section covers the design principles.
• If you want to try it on a project right now, Quick Start runs you through lyt init in two minutes.