@salesforce/agents/lib/agentTester.js

Client side APIs for working with Salesforce agents

"use strict";
/*
 * Copyright (c) 2024, salesforce.com, inc.
 * All rights reserved.
 * Licensed under the BSD 3-Clause license.
 * For full license text, see LICENSE.txt file in the repo root or https://opensource.org/licenses/BSD-3-Clause
 */
Object.defineProperty(exports, "__esModule", { value: true });
exports.AgentTester = void 0;
exports.normalizeResults = normalizeResults;
const core_1 = require("@salesforce/core");
const kit_1 = require("@salesforce/kit");
const maybe_mock_1 = require("./maybe-mock");
const utils_1 = require("./utils");
/**
 * A service for testing agents using `AiEvaluationDefinition` metadata. Start asynchronous
 * test runs, get or poll for test status, and get detailed test results.
 *
 * **Examples**
 *
 * Create an instance of the service:
 *
 * `const agentTester = new AgentTester(connection);`
 *
 * Start a test run:
 *
 * `const startResponse = await agentTester.start(aiEvalDef);`
 *
 * Get the status for a test run:
 *
 * `const status = await agentTester.status(startResponse.runId);`
 *
 * Get detailed results for a test run:
 *
 * `const results = await agentTester.results(startResponse.runId);`
 */
class AgentTester {
    maybeMock;
    constructor(connection) {
        this.maybeMock = new maybe_mock_1.MaybeMock(connection);
    }
    /**
     * Initiates a test run (i.e., AI evaluation).
     *
     * @param aiEvalDefName - The name of the AI evaluation definition to run.
     * @returns Promise that resolves with the response from starting the test.
     */
    async start(aiEvalDefName) {
        const url = '/einstein/ai-evaluations/runs';
        return this.maybeMock.request('POST', url, {
            aiEvaluationDefinitionName: aiEvalDefName,
        });
    }
    /**
     * Get the status of a test run.
     *
     * @param {string} jobId
     * @returns {Promise<AgentTestStatusResponse>}
     */
    async status(jobId) {
        const url = `/einstein/ai-evaluations/runs/${jobId}`;
        return this.maybeMock.request('GET', url);
    }
    /**
     * Poll the status of a test run until the tests are complete or the timeout is reached.
     *
     * @param {string} jobId
     * @param {Duration} timeout
     * @returns {Promise<AgentTestResultsResponse>}
     */
    async poll(jobId, { timeout = kit_1.Duration.minutes(5), } = {
        timeout: kit_1.Duration.minutes(5),
    }) {
        const frequency = kit_1.env.getNumber('SF_AGENT_TEST_POLLING_FREQUENCY_MS', 1000);
        const lifecycle = core_1.Lifecycle.getInstance();
        const client = await core_1.PollingClient.create({
            poll: async () => {
                const statusResponse = await this.status(jobId);
                if (statusResponse.status.toLowerCase() !== 'new') {
                    const resultsResponse = await this.results(jobId);
                    const totalTestCases = resultsResponse.testCases.length;
                    const passingTestCases = resultsResponse.testCases.filter((tc) => tc.status.toLowerCase() === 'completed' && tc.testResults.every((r) => r.result === 'PASS')).length;
                    const failingTestCases = resultsResponse.testCases.filter((tc) => ['error', 'completed'].includes(tc.status.toLowerCase()) &&
                        tc.testResults.some((r) => r.result === 'FAILURE')).length;
                    if (resultsResponse.status.toLowerCase() === 'completed') {
                        await lifecycle.emit('AGENT_TEST_POLLING_EVENT', {
                            jobId,
                            status: resultsResponse.status,
                            totalTestCases,
                            failingTestCases,
                            passingTestCases,
                        });
                        return { payload: resultsResponse, completed: true };
                    }
                    await lifecycle.emit('AGENT_TEST_POLLING_EVENT', {
                        jobId,
                        status: resultsResponse.status,
                        totalTestCases,
                        failingTestCases,
                        passingTestCases,
                    });
                }
                return { completed: false };
            },
            frequency: kit_1.Duration.milliseconds(frequency),
            timeout,
        });
        return client.subscribe();
    }
    /**
     * Get detailed test run results.
     *
     * @param {string} jobId
     * @returns {Promise<AgentTestResultsResponse>}
     */
    async results(jobId) {
        const url = `/einstein/ai-evaluations/runs/${jobId}/results`;
        const results = await this.maybeMock.request('GET', url);
        return normalizeResults(results);
    }
    /**
     * Cancel an in-progress test run.
     *
     * @param {string} jobId
     * @returns {Promise<{success: boolean}>}
     */
    async cancel(jobId) {
        const url = `/einstein/ai-evaluations/runs/${jobId}/cancel`;
        return this.maybeMock.request('POST', url);
    }
}
exports.AgentTester = AgentTester;
/**
 * Normalizes test results by decoding HTML entities in utterances and test result values.
 *
 * @param results - The agent test results response object to normalize
 * @returns A new AgentTestResultsResponse with decoded HTML entities
 *
 * @example
 * ```
 * const results = {
 *   testCases: [{
 *     inputs: { utterance: "&quot;hello&quot;" },
 *     testResults: [{
 *       actualValue: "&amp;test",
 *       expectedValue: "&lt;value&gt;"
 *     }]
 *   }]
 * };
 * const normalized = normalizeResults(results);
 * ```
 */
function normalizeResults(results) {
    return {
        ...results,
        testCases: results.testCases.map((tc) => ({
            ...tc,
            generatedData: {
                ...tc.generatedData,
                invokedActions: (0, utils_1.decodeHtmlEntities)(tc.generatedData.invokedActions),
            },
            inputs: {
                utterance: (0, utils_1.decodeHtmlEntities)(tc.inputs.utterance),
            },
            testResults: tc.testResults.map((r) => ({
                ...r,
                actualValue: (0, utils_1.decodeHtmlEntities)(r.actualValue),
                expectedValue: (0, utils_1.decodeHtmlEntities)(r.expectedValue),
                metricExplainability: (0, utils_1.decodeHtmlEntities)(r.metricExplainability),
            })),
        })),
    };
}
//# sourceMappingURL=agentTester.js.map