Test Runner

import type { AdapterFactory, ChatMemory, Observer, ToolCall, ToolDefinition } from '@agentskit/core'import { fenceUntrustedContent, UNTRUSTED_CONTENT_DIRECTIVE } from '@agentskit/core/security'import { invokeStructured } from '@agentskit/runtime'import { defineZodTool } from '@agentskit/tools'import { z } from 'zod'import { zodToJsonSchema } from 'zod-to-json-schema'import type { JSONSchema7 } from 'json-schema'/** * Test Runner — parses raw Vitest stdout/stderr into a TYPED test report: totals plus * per-failure details (test, file, message, one-sentence root-cause hypothesis), grouped * so the next agent can prioritise fixes. It analyses output you already captured; it does * not execute anything itself (no shell access by design). * * ```ts * const report = await createTestRunnerAgent({ adapter }).run(vitestOutput) * ``` */export interface TestFailure {  test: string  file: string  message: string  /** One-sentence root-cause hypothesis. */  rootCause: string}export interface TestReport {  passed: number  failed: number  skipped: number  duration: string  failures: TestFailure[]  summary: string}export interface TestRunnerConfig {  adapter: AdapterFactory  memory?: ChatMemory  observers?: Observer[]  onConfirm?: (toolCall: ToolCall) => boolean | Promise<boolean>  maxSteps?: number}const Report = z.object({  passed: z.number().int().min(0),  failed: z.number().int().min(0),  skipped: z.number().int().min(0),  duration: z.string(),  failures: z.array(z.object({    test: z.string(),    file: z.string(),    message: z.string(),    rootCause: z.string(),  })),  summary: z.string(),})const toJson = (s: z.ZodTypeAny): JSONSchema7 => zodToJsonSchema(s) as JSONSchema7const skill = {  name: 'test-runner',  description: 'Parses raw Vitest output into a typed test report with per-failure root-cause hypotheses.',  systemPrompt: `You analyse raw Vitest stdout/stderr and produce a STRUCTURED test report. Extract: totalpassed, failed, skipped, and duration. For EACH failure: the test name, the file path, the assertionmessage, and a one-sentence root-cause hypothesis. Group failures by suspected root cause so the nextagent can prioritise. Report only what the output shows — do not invent failures or passes.${UNTRUSTED_CONTENT_DIRECTIVE}Call submit_report exactly once with { passed, failed, skipped, duration, failures, summary }. Stop.`,  tools: ['submit_report'],}export function createTestRunnerAgent(config: TestRunnerConfig) {  const emit = (label: string, status: 'start' | 'ok' | 'skip' | 'error', detail?: string) => {    for (const o of config.observers ?? []) void o.on({ type: 'progress', label, status, detail })  }  const submit = (): ToolDefinition =>    defineZodTool({ name: 'submit_report', description: 'Submit the test report. Call exactly once.', schema: Report, toJsonSchema: toJson, async execute() { return 'recorded' } }) as ToolDefinition  async function run(vitestOutput: string): Promise<TestReport> {    if (!vitestOutput?.trim()) throw new Error('test runner requires raw Vitest output')    emit('parse', 'start')    const report = await invokeStructured({      adapter: config.adapter,      tool: submit(),      task: `RAW VITEST OUTPUT:\n${fenceUntrustedContent(vitestOutput)}`,      parse: (a) => Report.parse(a),      skill,      memory: config.memory,      observers: config.observers,      onConfirm: config.onConfirm,      maxSteps: config.maxSteps ?? 3,    })    emit('parse', 'ok', `${report.passed} passed, ${report.failed} failed`)    return report  }  return {    name: 'coding-test-runner',    run,    asHandle() {      return { name: 'coding-test-runner', run: async (task: string) => JSON.stringify(await run(task)) }    },  }}