fabric/testing

Helpers for testing an application’s agents with scripted models, and approvers that trust any reviewer. Production code needs nothing here. A store backend’s checks are in fabric/store/conformance.

Values

call

</>

pub fn call(
  definition: tool.Definition(input, output),
  id: String,
  input: input,
) -> Result(model.ToolCall, codec.EncodeError)

A call to definition with input encoded by its input codec, as a model would request it: for a scripted model (model.new) that calls the application’s tools. The arguments always decode under the tool bound from the same definition.

trusting_approvers

</>

pub fn trusting_approvers() -> @internal Approvers(
  reviewer.Reviewer,
  approvers.Denial,
)

Approvers that verify nothing: the credential is the reviewer itself, and every requirement is granted. For tests that are not about who answers; never give them to a production agent, since any code that can name a reviewer can then answer. They are named "fabric/testing.trusting_approvers", the verifier their answers record. Unlike approvers from approvers.new, every call returns the same approvers (in one VM), so a test can give them to its agents in one place and check proofs with them in another.

let approvers = testing.trusting_approvers()
// agent.new(..) |> agent.with_approvers(approvers) |> agent.build
let assert Ok(proof) =
  approvers.check(approvers, alice, pending.reference.requirement)
fabric.approve(handle, pending.reference, proof:, context:)