Add stop conditions before connecting multiple AI steps

The automation has explicit stops for missing sources, sensitive data, uncertainty, unusual requests, and failed checks.

1. 01

   Draw the smallest sequence from trigger to draft to review to final action.

2. 02

   At each handoff, name the expected input and a validation that can fail closed.

3. 03

   Stop on missing authority, prohibited data, unsupported claims, malformed output, or unusual consequence.

4. 04

   Test the stops with fictional edge cases before connecting a live system.

• Every step has an owner.
• Each handoff is validated.
• High-impact actions require approval.
• A safe manual fallback exists.