Why this exists
Agent Manifest addresses a specific problem:
As AI agents become more autonomous, composable, and networked, much of the surrounding tooling describes how they communicate, what they can do, or how the services they call constrain access. Comparatively little lets an agent’s deployer declare, before interaction and in a neutral, portable form, who it is, what it intends, what it will not do, and who is accountable for it.
Most existing approaches focus on:
- execution
- enforcement
- scoring
- policy
- control
- runtime behavior
Agent Manifest intentionally does none of those.
The Gap
Today, agents can act before any clear declaration of:
- identity
- purpose
- declared boundaries
- autonomy level
- risk profile
- data handling assumptions
This creates opacity by default.
Agent Manifest defines a declarative layer that precedes interaction, rather than following execution.
What This Is
Agent Manifest is:
- open
- minimal
- execution-agnostic
- framework-independent
- protocol-neutral
It is designed to be read by:
- humans
- agents
- systems
- validators
- auditors
Without requiring trust in the runtime.
What This Is Not
Agent Manifest does not:
- execute code
- validate behavior
- enforce rules
- score agents
- grant permissions
- make decisions
Those responsibilities belong elsewhere.
The Intent
The intent is not control.
The intent is clarity before action.
Agent Manifest allows any agent to say, in advance:
“This is who I am.
This is what I am for.
These are my limits.”
Nothing more.
Nothing less.
Why Now
The need for declarations precedes:
- regulation
- institutions
- enforcement layers
Agent Manifest is intentionally small so it can:
- coexist with future systems
- be embedded anywhere
- remain stable as runtimes evolve
A Primitive, Not a Platform
Agent Manifest is not a product.
It is a primitive.
A building block that others can adopt, ignore, extend, or reference.
Its value comes from being simple enough to survive.