Read-first by default
We start by inspecting and mapping context before proposing broad write access.
Trust
AI work needs boundaries before speed.
How HILLS Lab handles client context, private repositories, permissions, AI workflows, and approval boundaries.
Least privilege
Repository, ticket, document, and chat access should be scoped to the task and reviewed before expansion.
No training by default
Client data is not used for model training unless explicitly agreed in writing.
Approval boundaries
Sensitive write actions, public changes, data export, and production-impacting steps stay behind human review.
Source-backed memory
Durable context should point to a source, owner, scope, freshness status, and replacement path.
Private stays private
We do not publish client code, internal paths, operational details, secrets, or private repo history.