Policies

Policies are deterministic decision rules that determine whether an action is allowed. They are evaluated at execution time and return one of three outcomes: allow, deny, or requires_exception.

Overview

Policies define whether an action may proceed, not how it proceeds. They must be deterministic, explainable, and auditable.

Policy Outcomes

Policies return one of three outcomes:

  • allow: The action is permitted
  • deny: The action is not permitted
  • requires_exception: Human approval is required

Policy Characteristics

Policies:

  • Are evaluated at execution time with explicit inputs
  • Are versioned and immutable once published
  • Can define exception routes (who can approve, via which channel)
  • Are recorded in decision traces with their version/hash

Policy Inputs

[Content to be filled - how to define inputs, types, validation]

Policy Logic

[Content to be filled - rule sets, evaluation logic, examples]

Exception Routes

[Content to be filled - how policies reference approval routes]

TraceMem is trace-native infrastructure for AI agents