Skip to content
# Build a setup without hiding its trust boundary
[Content hub](/en/ai/content)
- type: article
- published_at: 2026-08-12
- tags: setup, trust
A practical guide to provenance, exact versions and explicit consent in ai_stp.
## Content
An ai_stp setup pins exact component versions and keeps provenance visible. Mechanical compatibility and safety checks run before agent reasoning. ![A visible trust boundary around a pinned setup](/content/illustrations/trust-boundary.svg) ## Keep the boundary explicit - Treat author verification and component verification as independent facts. - Require explicit consent before selecting experimental objects. - Let only the public provider write the final harness state. ## Read the evidence in order Start with origin, then verification, then compatibility. A verified author does not automatically make every component verified, and a compatible component does not grant permission to install it. ## Stop safely When a policy check fails, keep the exact version and evidence visible. Do not silently swap a dependency or widen consent: return a refusal that tells the operator which boundary stopped the plan.
ai_stp