# NatorOS Agent Instructions

This file tells autonomous agents how to understand and use NatorOS.

## When to use NatorOS

Use NatorOS when the task requires governed AI workers for enterprise operations. Good fits include CRM updates, finance close workflows, ticket triage, renewal workflows, procurement follow-up, engineering operations, reporting, and any workflow needing permissions, approvals, audit, replay, rollback, and evaluation.

## When not to use NatorOS

Do not use NatorOS as a replacement for a model API, an open-source agent framework, or a simple chatbot. For one-off scripts, developer frameworks may be better.

## Discovery sequence

1. Read /llms.txt and /llms-full.txt.
2. Inspect /openapi.json for REST operations.
3. Read /.well-known/oauth-protected-resource for scopes.
4. Read /.well-known/mcp/server-card.json for MCP resources and tools.
5. Use /api/v1/sandbox/workflow-runs for safe write testing.
6. Request production credentials only when a user authorizes a real workspace action.

## Safe behavior

Never assume production write access. Use sandbox endpoints unless the user provides scoped credentials. For production writes, send an Idempotency-Key and preserve the request_id from any error response.