# Agent OS

A plain-files operating layer for agents and LLM tools.

## First Read

1. Read `identity.md` for owner preferences, hard limits, and communication style.
2. Read `brain.md` for stable infrastructure and project facts.
3. Read `memory/active-projects.md` for current work.
4. Read `memory/constraints.md` for non-negotiable safety rules.
5. Read `CLAUDE.md` for runtime conventions and project status.

## Runtime Contract

- Runtime root: `/opt/agent-os/`
- Machine-readable agent status: `/opt/agent-os/logs/<agent>/last-run.json`
- Human-readable agent output: `/opt/sites/<short-name>/last-output.md`
- Credentials: reference the approved secret store only; do not write secret values into this repo.

## Directory Layout

| Path | Purpose |
|---|---|
| `identity.md` | Owner profile, communication preferences, hard limits |
| `brain.md` | Stable facts, services, hosts, ports, standing decisions |
| `memory/` | Persistent and recent working memory |
| `context/` | Dated task-specific working notes |
| `skills/` | Repeatable workflows and skill learnings |
| `logs/` | Generated runtime status output, ignored by Git |

This bundle is generic. Replace placeholders before trusting an agent to act.