Authority matrix reference - Harmonia — Agents for Humans

Capability	Agent	Human operator	Deterministic application	External provider
Analyze and recommend	Yes, within role	Yes	Validates	Supplies evidence only
Approve strategy	No	Yes	Authenticates and binds digest	No
Select and persist next stage	No	No	Yes	No
Approve an effect	No	Yes	Authenticates and binds digest	No
Create a scheduled command	No	Requests through UI	Yes	No
Execute an authorized effect	No	No	Yes	Performs requested API operation
Record receipt	No	No	Yes	Supplies response evidence
Verify result	No	No	Yes, by independent read-back	Supplies observed state
Grant policy exception	No	Explicit governed action only	Enforces policy	No

AgentCore Memory, AgentCore Runtime sessions, telemetry, and calendar entries confer no authority. Tenant membership and payload identity are checked at decision time. See Authority boundaries.

Agent-specific reasoning scope

Agent	May decide	Must not decide
Nimi	Evidence-supported moments and analytical angles	Strategy, copy, approval, or effects
Ryan	Strategy, briefs, priorities, channels, cadence, CTA, and KPI recommendations	Strategy approval, calendar dates, copy, or effects
Temi	Operational sequencing inside the approved strategy and immutable planning snapshot	Strategy changes, external calendar mutation, approval, or effects
Noni	Exact platform-native draft and one issue-bound revision	Strategy, item selection, approval, scheduling, or publishing
Dara	Accept or request bounded draft correction	Rewriting, final effect approval, or workflow mutation
Maya	Reference-only component composition	Authoritative content, decisions, or callbacks
Nova	Read-only interpretation of current tool evidence	Commands, approvals, retries, schedules, or mutations

The complete runtime capability inventory is here.

Last modified on September 14, 2026