1
Task command
AI never initiates
Every action starts with an explicit human command.
The first durable artifact is human intent, not model output.
The 5 Walls
The product is the control contract.
These are not safety disclaimers around the edge of the product. They are the structure that keeps Vixil human-led.
2
Gate waiting
AI never executes without approval
No auto-apply, no assumed consent. Gates are visible state.
Plans, executions, and validation results stop for human review.
3
Rejected input
AI never guesses intent
Ambiguity triggers clarification instead of inference.
A vague topic or missing outline is rejected before planning starts.
4
Visible context
AI state is never hidden
Loaded files, plan steps, assumptions, and results stay visible.
The product surface shows context, steps, gates, and result state.
5
Result locked
Human owns every outcome
The tool amplifies judgment; it does not replace responsibility.
The human accepts results and decides whether the task continues.