AxiomCore / EXPLORE
Get started

Declarative software for humans + agents.

What a coding agent needs beyond a file tree

Use explicit contracts to connect UI actions, backend operations, validation, and tests without confusing declared relationships with observed behavior.

By AxiomCore engineering ·

A repository tells a coding agent where files live. It does not necessarily tell it which interface a consumer is allowed to rely on, which operation a button calls, or which callers a changed field might affect. These are engineering questions even when an agent writes the code.

AxiomCore approaches this through declared contracts and inspectable application evidence. The aim is to give the developer and agent a shared model they can check, rather than expecting either to reconstruct every relationship from naming conventions.

Three different questions

Consider a task application with a title input and a submit action. There are at least three questions worth keeping separate:

  1. What is declared? A model defines the title field. An operation accepts an input. A UI action calls that operation.
  2. What is allowed? Types, validation, and authority constrain the values and effects accepted at the boundary.
  3. What happened? A particular submission succeeded or failed during an observed execution.

A static relationship answers part of the first question. It cannot establish the third. A mock success also cannot prove that an external production service implements the same behavior.

Start with a small, inspectable example

Open the AxiomCore Playground. It loads frontend and backend sources for a task application and compiles them in the browser using WebAssembly. The preview uses a local mock backend; its runtime events belong to that browser session.

Run the example, add a task, and open the Inspector’s graph and runtime views. Look for the model, operation, and UI relationships associated with that interaction. Then compare what the graph declares with what the runtime actually recorded. The distinction matters more than the number of nodes on the screen.

For a local workspace, the documented Inspector entry points are:

axiom inspect overview .
axiom inspect overview . --format json
axiom inspect serve .

The JSON form lets tooling consume the same evidence model. The local dashboard provides an interactive view. See the maintained Inspector guide for prerequisites and current command behavior.

Make an agent’s review task specific

Instead of asking only whether a change looks correct, ask the agent to identify the changed public fact, the supported relationships leading to consumers, and the evidence still missing. That produces a review you can challenge with source, contracts, tests, and runtime observations.

A useful review prompt is: “Which declared consumers depend on this field, what changed about accepted values, and what execution evidence would we need before releasing it?” This is a suggested workflow, not a claim that every agent automatically integrates with Inspector.

Keep unknowns visible

Inspector cannot infer an external implementation body from an endpoint name. It also should not treat requested extension permissions as an application grant. Evidence gaps and bounded traversals belong in the result, especially when an agent turns that result into prose.

AxiomCore is alpha. Use the support matrix to choose a supported path. For an existing service, incremental adoption starts with one boundary and one consumer; moving the whole application to Acore is a separate choice.

Continue with application context for AI coding agents or a validation-change review.

AxiomStudio · Early access

Build with agents, first.

Join the AxiomStudio waitlist. We'll email you when early access opens.

We'll only email you about early access.

You're on the list.

We'll email when it opens.