Agents read stored records and evidence; public page summaries are not production data.
Agent safety and governance
Action-ready Planetary Model context needs governed software use.
Agents are useful only when the systems they call are inspectable, scoped, and honest about uncertainty. AgentLayer keeps context and evidence attached rather than flattening the answer into free-form text.
Rules
Agents get context with boundaries.
Every response should preserve provenance, confidence, uncertainty, and entitlements.
Grounded context packets keep current state, evidence, contradictions, and causal context together.
Discovery outputs preserve blockers, uncertainty, entitlement posture, and next-action posture.
Composite scores must show their components and whether high means risk or quality.
Sensitive outputs require the right scope and may be limited or withheld.
Scenario results are labeled clearly and kept separate from observed records.
AgentLayer is the integration surface for production machine use.
Agents should cite returned records and evidence.
No public-page shortcuts
Production agents should use approved product access rather than treating public pages as data.
No unsupported claims
Outputs should cite returned records, evidence, and decision context when used downstream.
No silent sensitivity handling
Restricted or sensitive fields should be redacted or withheld with a visible reason.
No scenario mixing
Scenario results stay labeled and separate from observed records.