# Agent OS Project Brief

## What This Project Is

Agent OS is a tool-agnostic, plain-files operating layer for LLM agents. Markdown files are the shared memory, operating contract, and handoff surface.

## Runtime

- Server runtime root: `/opt/agent-os/`
- Scheduled jobs: local scheduler or cron on the server
- Agent web output: `/opt/sites/<short-name>/`
- Agent status logs: `/opt/agent-os/logs/<agent>/last-run.json`

## Required Output Contract

Every scheduled or autonomous agent should write:

- `last-run.json` at `/opt/agent-os/logs/<agent>/last-run.json`
- `last-output.md` at `/opt/sites/<short-name>/last-output.md`

Keep `last-output.md` concise: heading, timestamp, and useful summary.

## Repository Rules

- Gitea or the configured remote is the recoverable source of truth.
- Runtime state, logs, caches, credentials, databases, uploads, and generated exports stay out of commits.
- Review and validate changes before deployment.
- Copy successful live fixes back to the edit clone before committing.

## First-Run Checklist

1. Fill in `identity.md`.
2. Fill in stable infrastructure facts in `brain.md`.
3. Record non-negotiable constraints in `memory/constraints.md`.
4. Replace this project brief with environment-specific status.
5. Add real skills under `skills/<skill-name>/`.