Skip to content
AX Framework
Principle 01

Explicit Capabilities

A product should declare what an agent can do with it, in a machine-readable form, rather than leaving the agent to infer capabilities from the interface.

Why it matters

Agents plan before they act. A plan built on guessed capabilities fails at the first wrong guess, often after the agent has already spent time and tokens or made partial changes.

Explicit capabilities also make the product's intent clear. A product that publishes "search, compare, add to cart, check out" is telling agents what it supports and, by omission, what it does not.

Examples

Good exampleIllustrative

A store publishes a manifest listing each action with a one-line purpose, an input schema, and whether authentication is required.

json
{
  "name": "add_to_cart",
  "description": "Add a purchasable product variant to the active cart.",
  "input": { "$ref": "#/schemas/AddToCartInput" },
  "auth": "session",
  "sideEffects": "modifies_cart",
  "idempotent": true
}
Bad exampleIllustrative

The only way to learn that a product can be gift-wrapped is to notice a checkbox that appears after selecting a size.

Implementation guidance

  1. 1Start from user tasks, not from your API surface. List the five to ten jobs an agent is most likely to be asked to do.
  2. 2Publish one entry point that links to every agent-facing capability.
  3. 3Describe each capability in a sentence a non-expert would understand.
  4. 4Remove capabilities from the manifest the moment they stop working.

Evaluation criteria

  • An agent can list supported tasks without rendering any page.
  • Every listed capability has a description, input schema, and auth requirement.
  • No capability that agents commonly use is missing from the manifest.