eve/dist/src/evals/assertions/run.d.ts

Filesystem-first framework for durable backend AI agents that run anywhere.

import type { StandardSchemaV1 } from "#compiled/@standard-schema/spec/index.js";
import type { HandleMessageStreamEvent } from "#protocol/message.js";
import { type EveEvalSkillLoadMatchOptions, type EveEvalSubagentCallMatchOptions, type EveEvalToolCallMatchOptions } from "#evals/match.js";
import type { RunAssertion } from "#evals/assertions/collector.js";
/**
 * Asserts the run ran to completion: it did not fail and did not park on an
 * unanswered HITL input request.
 */
export declare function completed(): RunAssertion;
/**
 * Asserts the run ended parked on HITL input. For approval gates and
 * ask-question flows.
 */
export declare function waiting(): RunAssertion;
/**
 * Asserts the run did not fail: terminal status is not `"failed"` and no
 * `turn.failed` / `step.failed` events were emitted. Parked runs pass; use
 * {@link completed} to also reject parking.
 */
export declare function didNotFail(): RunAssertion;
/**
 * Asserts the joined assistant message text contains `token` (substring for
 * strings, `test` for RegExps).
 */
export declare function messageIncludes(token: string | RegExp): RunAssertion;
/**
 * Asserts a tool call with `name` happened. Options constrain the call
 * further: `input` partial-deep-matches, `output` matches the result,
 * `isError` constrains error state, and `times` requires an exact count.
 */
export declare function calledTool(name: string, options?: EveEvalToolCallMatchOptions): RunAssertion;
/**
 * Sugar over {@link calledTool} for the framework `load_skill` tool: asserts a
 * skill with id `skill` was loaded. `output`/`isError`/`times` constrain the
 * matching call exactly as for `calledTool`.
 */
export declare function loadedSkill(skill: string, options?: EveEvalSkillLoadMatchOptions): RunAssertion;
/**
 * Asserts no tool call with `name` happened.
 */
export declare function notCalledTool(name: string): RunAssertion;
/**
 * Asserts the named tools were called in the given order (subsequence match:
 * other calls may interleave).
 */
export declare function toolOrder(names: readonly string[]): RunAssertion;
/**
 * Asserts the run made no tool calls at all.
 */
export declare function usedNoTools(): RunAssertion;
/**
 * Asserts the run made at most `max` tool calls.
 */
export declare function maxToolCalls(max: number): RunAssertion;
/**
 * Asserts no action result (tool, subagent, or skill) reported a failure.
 */
export declare function noFailedActions(): RunAssertion;
/**
 * Asserts a subagent delegation to `name` occurred. `remoteUrl` matches the
 * `subagent.called` remote metadata, `output` matches the `subagent.completed`
 * output.
 */
export declare function calledSubagent(name: string, options?: EveEvalSubagentCallMatchOptions): RunAssertion;
/**
 * Escape hatch: asserts an arbitrary predicate over the full typed event
 * stream. `label` names the assertion in reports.
 */
export declare function event(predicate: (events: readonly HandleMessageStreamEvent[]) => boolean, label: string): RunAssertion;
/**
 * Asserts `result.output` (the final assistant message) deep-equals `value`.
 */
export declare function outputEquals(value: unknown): RunAssertion;
/**
 * Asserts `result.output` validates against a Standard Schema (e.g. a Zod
 * schema).
 */
export declare function outputMatches(schema: StandardSchemaV1): RunAssertion;