How it works · mechanism

One type of principal. One kind of edge. Three rules that cannot be argued with.

This page explains this system, not the category. For what agent identity is, see flashyos.com/concepts/agent-identity and gda.group.

Object 01

Principal

One type, three kinds: human, organization, agent. The whole idea is that they are the same type — an agent is addressed exactly as a person is.

kind: human | org | agent
Object 02

Grant

The edge. Carries scopes, constraints, expiry, and a pointer to its parent. Nothing in the system holds authority except a grant.

scopes · limits · exp · parent
Object 03

Policy

The organization’s rules, evaluated server-side. An agent never supplies the judgement that gates it.

PLANNED · Q1 2027
Object 04

Assertion

The signed, portable claim. The only object that crosses an organizational boundary, and the only one a counterparty needs.

RS256 · kid flshy-01

The delegation chain

Each edge is annotated by what it carries. A counterparty walks this without holding any Flashy secret.

Human
Named person
root grant
identity of record
issues
Organization
Flashy Academy
role-gated
policy owner
scopes
Agent
CEO Agent
spend ≤ $5,000
exp 4d 6h
attenuates
Agent
Ops Agent
spend ≤ $500
exp 6h
presents
Foreign agent
Counterparty
verifies offline
no shared secret
Every link resolves to the human at the root. That is the accountability chain, and it is the reason the product exists.

The three invariants

Enforced in code

Attenuation only — a grant can never exceed its parent.

Enforced in code

Authority always expires — there is no permanent grant.

Enforced in code★ Cross-org propagation · planned Q1 2027

Revocation walks down — revoking a link revokes everything beneath it.

Within one organization this is enforced today at resolution time. Propagation to a counterparty who has cached an assertion is not yet built; today they must re-resolve. Dated gold, on the sentence.

What we did not have to say★ Policy engine · planned Q1 2027

The policy engine described above is not built. Today an agent’s declared impact is taken at face value and a LOW impact auto-approves. Until Policy ships, treat every scope on this page as an assertion about intent, not an enforced boundary — except expiry and attenuation, which are checked at resolution.