Skip to main content

Interface: ForgeTestCase

Defined in: packages/agentos/src/cognition/emergent/types.ts:775

A single test case used by the LLM judge to evaluate a newly forged tool.

The judge invokes the tool with input and compares the result against expectedOutput using semantic equivalence (not strict equality).

Properties​

expectedOutput​

expectedOutput: unknown

Defined in: packages/agentos/src/cognition/emergent/types.ts:786

Expected output value used for correctness scoring. The judge uses this as a reference — partial matches may still score well.


input​

input: Record<string, unknown>

Defined in: packages/agentos/src/cognition/emergent/types.ts:780

Input arguments object passed to the tool's run / execution entry point. Must conform to the tool's declared inputSchema.


responses?​

optional responses: ForgeTestResponse[]

Defined in: packages/agentos/src/cognition/emergent/types.ts:803

Answers for the state-changing requests a code-forged tool makes while it is forged, which are never sent: a request takes the first unused answer whose method and URL match it, otherwise a 204 with the header x-agentos-forge-test: not-sent. At most 16, each body within the ceiling's maxResponseBytes.


stepOutputs?​

optional stepOutputs: Record<string, unknown>

Defined in: packages/agentos/src/cognition/emergent/types.ts:794

Outputs for the steps of a composition that have side effects (or are compositions themselves). While a tool is forged those steps are not executed; each takes its output from here, keyed by step name, so later steps receive real data.