@post-print/agent-test 1.1.9 → 2.0.0

package/dist/cli.js

@@ -1,702 +1,37 @@
-#!/usr/bin/env -S node --disable-warning=ExperimentalWarning
-import { join, resolve } from "node:path";
+#!/usr/bin/env node
 import { fileURLToPath } from "node:url";
-import { cleanupStaleScenarioWorktrees, isPathUnderRoot, knownAgentHosts, missingClassifierAuth, parseHostAuthModeFlag, setProcessAuthMode, } from "@post-print/agent-harness";
-import { collectSuiteHosts, formatCheckReport, formatCheckSummary, missingHostsAuth, runCheck, } from "./check.js";
+import { parseArgs } from "node:util";
 import { isCliMain } from "./cli-entry.js";
-import { parseCompareArmId } from "./compare-scenario.js";
-import { missingAgentAuth } from "./doctor.js";
-import { installHostLogFilter } from "./host-log.js";
-import { ensureCursorSdkLoginForLive, offerCursorSdkReloginIfNeeded, runHostLogin, } from "./host-login.js";
-import { parseHostList, uniqueHosts } from "./hosts.js";
-import { writeHtmlReport } from "./html-report.js";
-import { loadHostAdapters } from "./load-adapters.js";
-import { assertDirectAgentPreflight } from "./preflight.js";
-import { logProgress } from "./progress.js";
-import { cleanupStagingSession, createLiveStagingSessionId, getLiveStagingSessionRoot, setLiveStagingRootOverride, } from "./record-trace.js";
-import { shouldStartReportPreview, startDetachedViewer } from "./report-preview.js";
-import { registerLiveRunHandlers, runAllSuites, selectedRunNeedsJudge } from "./run-suite.js";
-import { formatRunSummary, shouldFailScenario, summarizeReports, } from "./suite-summary.js";
-import { configureCliColor, theme } from "./theme.js";
-import { loadViewerCatalog } from "./viewer/catalog.js";
-import { startViewerCli } from "./viewer/cli.js";
-import { suppressNoisyRuntimeWarnings } from "./warnings.js";
-import { parseWorkerCount } from "./worker-pool.js";
-suppressNoisyRuntimeWarnings();
-configureCliColor();
-function splitArgvFlag(token) {
-    if (!token.startsWith("-") || token === "-") {
-        return undefined;
-    }
-    if (token.startsWith("--")) {
-        const eq = token.indexOf("=");
-        if (eq === -1) {
-            return { flag: token };
-        }
-        return { flag: token.slice(0, eq), inline: token.slice(eq + 1) };
-    }
-    return { flag: token };
-}
-function readFlagValue(flag, inline, argv, index) {
-    if (inline !== undefined) {
-        if (inline.length === 0) {
-            throw new Error(`${flag} requires a value`);
-        }
-        return { value: inline, nextIndex: index };
-    }
-    const next = argv[index + 1];
-    if (!next || next.startsWith("--") || /^-[A-Za-z]/.test(next)) {
-        throw new Error(`${flag} requires a value`);
-    }
-    return { value: next, nextIndex: index + 1 };
-}
-function enableCheckMode(target) {
-    target.check = true;
-    target.validateOnly = true;
-    target.validatePaths = true;
-    target.validateSeeds = true;
-}
-/** Usage text for `--help`. */
-export function formatHelp() {
-    return [
-        "agent-test [options] [suite]",
-        "agent-test login [--host cursor|claude|openai]",
-        "agent-test viewer [--suites-dir agent-suites]",
-        "",
-        "Launch a host agent and score the transcript.",
-        "A run checks the suite and host first.",
-        "",
-        "  login                           Store a Cursor SDK login. Codex uses `codex login`.",
-        "  viewer                          Open the unified localhost catalog and session run history.",
-        "  --check                         Check the suite and host. Do not launch an agent.",
-        "  --help, -h                      Print this help.",
-        "",
-        "  --host cursor|claude|openai|all Host adapter (default: suite or cursor).",
-        "                                  Repeat or comma-separate for a matrix.",
-        "  --auth-mode subscription|api-key",
-        "                                  Host auth (default: subscription).",
-        "                                  --auth-method is the same flag.",
-        "  --adapter <module>              Load a consumer host adapter module.",
-        "  --suites-dir <path>             Suite root (default: agent-suites)",
-        "  --suite <name>                  Run one suite",
-        "  --scenario <name>               Run one scenario",
-        "  --compare-arm <id>              Isolated child: run one compare arm",
-        "  --port <n>                      Viewer listen port. Default is a free port.",
-        "  --no-judge                      Skip the judge",
-        "  --fail-on all|behavior|infra-only",
-        "  --allow-user-input              Let a user agent answer AskQuestion tools",
-        "  --timeout-ms <n>                Agent deadline",
-        "  --no-timeout                    Disable the deadline",
-        "  --scenario-retries <n>          Announce-stop retries",
-        "  --workers <n>                   Parallel live agents (1-32). Viewer default 4. CLI default 1.",
-        "  --debug                         Keep recordings and write a debug bundle",
-        "  --debug-dir <path>              Debug session parent directory",
-        "  --report-out <path>             Self-contained read-only HTML report file or directory",
-        "  --no-html-report                Skip the HTML report",
-        "  --no-worktree                   Run in the caller checkout (needs AGENT_TEST_ALLOW_IN_PLACE=1)",
-        "",
-        "--doctor, --validate-only, --validate-paths, and --validate-seeds run --check.",
-    ].join("\n");
-}
-/** Parse agent-test CLI argv (exported for unit tests). */
-export function parseCliArgs(argv) {
-    const cwd = process.cwd();
-    let suitesDir = "agent-suites";
-    let rubricsDir;
-    const parsedHosts = [];
-    let hostAll = false;
-    const adapterModules = [];
-    let filter;
-    let scenarioFilter;
-    let stagingSessionId;
-    let judge;
-    let worktree;
-    let keepRecordings = false;
-    let timeoutMs;
-    let noTimeout = false;
-    let allowUserInput = false;
-    let check = false;
-    let help = false;
-    let doctor = false;
-    let htmlReport = true;
-    let reportOut;
-    let debug = process.env.AGENT_TEST_DEBUG === "1" || process.env.AGENT_TEST_DEBUG === "true";
-    let debugDir;
-    let validateOnly = false;
-    let validateSeeds = false;
-    let validatePaths = false;
-    let failOn = "all";
-    let scenarioRetries;
-    let authMode;
-    let compareArm;
-    let viewerPort;
-    let workers;
-    if (argv[2] === "compare") {
-        throw new Error("The compare subcommand is removed. Add compare.a and compare.b, or compare.arms, on one scenario. Add compare.gates for required outcomes or metrics.");
-    }
-    const login = argv[2] === "login";
-    const viewer = argv[2] === "viewer";
-    const startIndex = login || viewer ? 3 : 2;
-    const checkMode = {
-        check: false,
-        validateOnly: false,
-        validatePaths: false,
-        validateSeeds: false,
-    };
-    for (let i = startIndex; i < argv.length; i++) {
-        const token = argv[i];
-        if (!token) {
-            continue;
-        }
-        const parsed = splitArgvFlag(token);
-        if (!parsed) {
-            if (login) {
-                throw new Error("login does not take a suite name");
-            }
-            filter = token;
-            continue;
-        }
-        const { flag, inline } = parsed;
-        const read = () => {
-            const result = readFlagValue(flag, inline, argv, i);
-            i = result.nextIndex;
-            return result.value;
-        };
-        switch (flag) {
-            case "--help":
-            case "-h":
-                help = true;
-                break;
-            case "--check":
-                enableCheckMode(checkMode);
-                break;
-            case "--doctor":
-                doctor = true;
-                enableCheckMode(checkMode);
-                break;
-            case "--validate-only":
-                enableCheckMode(checkMode);
-                break;
-            case "--validate-seeds":
-                enableCheckMode(checkMode);
-                break;
-            case "--validate-paths":
-                enableCheckMode(checkMode);
-                break;
-            case "--host": {
-                const value = read();
-                if (value
-                    .split(",")
-                    .map((token) => token.trim().toLowerCase())
-                    .includes("all")) {
-                    hostAll = true;
-                }
-                parsedHosts.push(...parseHostList(value));
-                break;
-            }
-            case "--auth-mode":
-            case "--auth-method":
-                authMode = parseHostAuthModeFlag(read(), flag);
-                break;
-            case "--adapter":
-                adapterModules.push(resolve(cwd, read()));
-                break;
-            case "--suites-dir":
-                suitesDir = read();
-                break;
-            case "--rubrics-dir":
-                rubricsDir = read();
-                break;
-            case "--suite":
-                filter = read();
-                break;
-            case "--scenario":
-                scenarioFilter = read();
-                break;
-            case "--compare-arm": {
-                const value = parseCompareArmId(read());
-                if (!value) {
-                    throw new Error("--compare-arm must be a lowercase slug");
-                }
-                compareArm = value;
-                break;
-            }
-            case "--port": {
-                const parsedPort = Number(read());
-                if (!Number.isInteger(parsedPort) || parsedPort < 0 || parsedPort > 65535) {
-                    throw new Error("--port must be an integer 0-65535");
-                }
-                viewerPort = parsedPort;
-                break;
-            }
-            case "--staging-session-id":
-                stagingSessionId = read();
-                break;
-            case "--live":
-                throw new Error("--live was removed because agent-test now always runs a real agent");
-            case "--record":
-                throw new Error("--record was removed; direct runs capture transient traces automatically (use --keep-recordings to retain them)");
-            case "--record-fixtures":
-                throw new Error("--record-fixtures was removed because replay-based testing is deprecated and no longer supported");
-            case "--keep-recordings":
-                keepRecordings = true;
-                break;
-            case "--judge":
-                judge = true;
-                break;
-            case "--no-judge":
-                judge = false;
-                break;
-            case "--no-worktree":
-                worktree = false;
-                break;
-            case "--timeout-ms": {
-                const parsedTimeout = Number(read());
-                if (Number.isFinite(parsedTimeout) && parsedTimeout > 0) {
-                    timeoutMs = parsedTimeout;
-                }
-                break;
-            }
-            case "--no-timeout":
-                noTimeout = true;
-                break;
-            case "--allow-user-input":
-                allowUserInput = true;
-                break;
-            case "--fail-on": {
-                const mode = read();
-                if (mode !== "all" && mode !== "behavior" && mode !== "infra-only") {
-                    throw new Error("--fail-on must be all|behavior|infra-only");
-                }
-                failOn = mode;
-                break;
-            }
-            case "--scenario-retries": {
-                const parsedRetries = Number(read());
-                if (!Number.isInteger(parsedRetries) || parsedRetries < 0) {
-                    throw new Error("--scenario-retries must be an integer >= 0");
-                }
-                scenarioRetries = parsedRetries;
-                break;
-            }
-            case "--workers":
-                workers = parseWorkerCount(read());
-                break;
-            case "--compare-pairs":
-            case "--a":
-            case "--b":
-            case "--compare-a":
-            case "--compare-b":
-            case "--out-dir":
-            case "--compare-out":
-                throw new Error(`${flag} is removed. Add compare.a and compare.b, or compare.arms, on one scenario. Add compare.gates for required outcomes or metrics.`);
-            case "--report-out":
-                reportOut = read();
-                break;
-            case "--no-html-report":
-                htmlReport = false;
-                break;
-            case "--debug":
-                debug = true;
-                break;
-            case "--debug-dir": {
-                const value = read();
-                debugDir = value;
-                debug = true;
-                break;
-            }
-            default:
-                throw new Error(`Unknown flag: ${flag}`);
-        }
-    }
-    check = checkMode.check;
-    validateOnly = checkMode.validateOnly;
-    validatePaths = checkMode.validatePaths;
-    validateSeeds = checkMode.validateSeeds;
-    const hosts = uniqueHosts(parsedHosts);
-    const host = hosts?.[0];
-    judge = judge ?? true;
-    worktree = worktree ?? true;
-    if (workers === undefined) {
-        const fromEnv = process.env.AGENT_TEST_WORKERS?.trim();
-        if (fromEnv) {
-            workers = parseWorkerCount(fromEnv, "AGENT_TEST_WORKERS");
-        }
-    }
-    if (debug) {
-        keepRecordings = true;
-        process.env.AGENT_TEST_DEBUG = "1";
-        process.env.AGENT_TEST_VERBOSE = process.env.AGENT_TEST_VERBOSE ?? "1";
-        process.env.AGENT_TEST_VERBOSE_PATHS = process.env.AGENT_TEST_VERBOSE_PATHS ?? "1";
-    }
-    return {
-        cwd,
-        suitesDir,
-        rubricsDir,
-        host,
-        hosts,
-        hostAll,
-        adapterModules: adapterModules.length > 0 ? adapterModules : undefined,
-        filter,
-        scenarioFilter,
-        stagingSessionId,
-        judge,
-        worktree,
-        keepRecordings,
-        timeoutMs: noTimeout ? 0 : timeoutMs,
-        noTimeout,
-        allowUserInput,
-        check,
-        help,
-        doctor,
-        htmlReport,
-        reportOut: reportOut ? resolve(cwd, reportOut) : undefined,
-        debug,
-        debugDir: debugDir ? resolve(cwd, debugDir) : undefined,
-        validateOnly,
-        validateSeeds,
-        validatePaths,
-        failOn,
-        scenarioRetries,
-        authMode,
-        login,
-        viewer,
-        compareArm,
-        viewerPort,
-        workers,
-    };
-}
-/** Resolve an explicit report target into the HTML path and optional artifact directory. */
-export function resolveReportOutput(reportOut) {
-    if (!reportOut) {
-        return {};
-    }
-    if (reportOut.toLowerCase().endsWith(".html")) {
-        return { htmlPath: reportOut };
-    }
-    return { htmlPath: join(reportOut, "report.html"), outDir: reportOut };
-}
-/** Path passed to `writeHtmlReport`. `undefined` keeps the temp default. */
-export function resolveHtmlReportWritePath(htmlReport, reportOut) {
-    if (!htmlReport) {
-        return undefined;
-    }
-    return resolveReportOutput(reportOut).htmlPath;
-}
-async function htmlReportTip(label, reportPath, reports, args) {
-    if (shouldStartReportPreview()) {
-        const url = await startDetachedViewer({
-            cliPath: process.argv[1],
-            cwd: args.cwd,
-            suitesDir: args.suitesDir,
-            rubricsDir: args.rubricsDir,
-            judge: args.judge,
-            timeoutMs: args.timeoutMs,
-            authMode: args.authMode,
-            adapterModules: args.adapterModules,
-            workers: args.workers,
-            worktree: args.worktree,
-            keepRecordings: args.keepRecordings,
-            allowUserInput: args.allowUserInput,
-            debug: args.debug,
-            debugDir: args.debugDir,
-            request: {
-                ...(args.filter ? { suite: args.filter } : {}),
-                ...(args.scenarioFilter ? { scenario: args.scenarioFilter } : {}),
-                ...(args.hosts?.length ? { hosts: args.hosts } : args.host ? { hosts: [args.host] } : {}),
-            },
-            reports,
+import { runHostLogin } from "./host-login.js";
+import { runSdkCli } from "./sdk/cli.js";
+import { runSdkViewerCli } from "./sdk/viewer.js";
+export async function runCli(args) {
+    const [command, ...rest] = args;
+    if (command === "test")
+        return runSdkCli(rest);
+    if (command === "viewer")
+        return runSdkViewerCli(rest);
+    if (command === "login") {
+        const { values } = parseArgs({
+            args: rest,
+            options: { host: { type: "string", default: "cursor" } },
         });
-        if (url) {
-            return theme.fileTip(label, url);
-        }
+        return runHostLogin(values.host);
     }
-    return theme.fileTip(label, reportPath);
-}
-async function cleanupRunArtifacts(stagingSessionRoot, keepRecordings) {
-    if (keepRecordings) {
-        return;
-    }
-    if (stagingSessionRoot) {
-        try {
-            await cleanupStagingSession(stagingSessionRoot);
-        }
-        catch {
-            // best-effort
-        }
-    }
-}
-async function main() {
-    let args;
-    try {
-        args = parseCliArgs(process.argv);
-        if (args.authMode) {
-            setProcessAuthMode(args.authMode);
-        }
-        installHostLogFilter();
-    }
-    catch (error) {
-        console.error(error instanceof Error ? error.message : error);
-        return 1;
-    }
-    if (args.help) {
-        console.log(formatHelp());
+    if (!command || command === "--help" || command === "-h") {
+        console.log("agent-test test [Playwright options]\nagent-test viewer [--config path] [--port number]\nagent-test login [--host cursor|openai|claude]\n\nConfigure agents in agent-test.config.ts. JSON suites and legacy scoring flags were removed.");
         return 0;
     }
-    if (args.login) {
-        try {
-            return await runHostLogin(args.host ?? "cursor");
-        }
-        catch (error) {
-            console.error(error instanceof Error ? error.message : error);
-            return 1;
-        }
-    }
-    try {
-        await loadHostAdapters({ cwd: args.cwd, adapterModules: args.adapterModules });
-        if (args.hostAll) {
-            args.hosts = uniqueHosts(knownAgentHosts());
-            args.host = args.hosts?.[0];
-        }
-    }
-    catch (error) {
-        console.error(error instanceof Error ? error.message : error);
-        return 1;
-    }
-    if (args.viewer) {
-        try {
-            return await startViewerCli(args);
-        }
-        catch (error) {
-            console.error(error instanceof Error ? error.message : error);
-            return 1;
-        }
-    }
-    if (args.check || args.doctor || args.validateOnly || args.validateSeeds) {
-        const report = await runCheck({
-            cwd: args.cwd,
-            suitesDir: args.suitesDir,
-            filter: args.filter,
-            rubricsDir: args.rubricsDir,
-            host: args.host,
-            hosts: args.hosts,
-        });
-        console.log(formatCheckReport(report));
-        return report.ok ? 0 : 1;
-    }
-    if (args.debugDir) {
-        setLiveStagingRootOverride(args.debugDir);
-    }
-    const isChild = process.env.AGENT_TEST_CHILD === "1";
-    if (args.debugDir && isPathUnderRoot(args.debugDir, args.cwd) && !isChild) {
-        console.warn(theme.warn(`--debug-dir is inside the repo (${args.debugDir}). Default is $TMPDIR/agent-spec — prefer that for direct runs so debug output stays out of git status.`));
-    }
-    const verbose = args.debug || process.env.AGENT_TEST_VERBOSE === "1" || process.env.AGENT_TEST_DEBUG === "1";
-    const stagingSessionId = args.stagingSessionId?.trim() ||
-        process.env.AGENT_TEST_STAGING_SESSION_ID?.trim() ||
-        createLiveStagingSessionId();
-    const stagingSessionRoot = stagingSessionId
-        ? getLiveStagingSessionRoot(stagingSessionId)
-        : undefined;
-    try {
-        let checkHosts = args.hosts ?? (args.host ? [args.host] : ["cursor"]);
-        if (!isChild) {
-            checkHosts = await collectSuiteHosts({
-                cwd: args.cwd,
-                suitesDir: args.suitesDir,
-                filter: args.filter,
-                rubricsDir: args.rubricsDir,
-                host: args.host,
-                hosts: args.hosts,
-            });
-            await ensureCursorSdkLoginForLive({
-                hosts: checkHosts,
-                authMode: args.authMode,
-            });
-            const report = await runCheck({
-                cwd: args.cwd,
-                suitesDir: args.suitesDir,
-                filter: args.filter,
-                rubricsDir: args.rubricsDir,
-                host: args.host,
-                hosts: args.hosts,
-            });
-            console.log(theme.tip(formatCheckSummary(report)));
-            if (!report.ok) {
-                console.error(formatCheckReport(report));
-                return 1;
-            }
-            checkHosts = report.hosts;
-            const missingHost = missingHostsAuth(report.hosts);
-            if (missingHost) {
-                console.error(missingHost);
-                return 1;
-            }
-        }
-        else {
-            const missingHost = missingAgentAuth(args.host ?? "cursor");
-            if (missingHost) {
-                console.error(missingHost);
-                return 1;
-            }
-        }
-        if (args.judge !== false) {
-            let needsJudge = false;
-            try {
-                needsJudge = await selectedRunNeedsJudge({
-                    cwd: args.cwd,
-                    suitesDir: args.suitesDir,
-                    filter: args.filter,
-                    scenarioFilter: args.scenarioFilter,
-                    rubricsDir: args.rubricsDir,
-                    judge: args.judge,
-                });
-            }
-            catch {
-                needsJudge = false;
-            }
-            if (needsJudge) {
-                for (const host of checkHosts) {
-                    const missingJudge = missingClassifierAuth(host);
-                    if (missingJudge) {
-                        console.error(`${missingJudge} (use --no-judge to skip)`);
-                        return 1;
-                    }
-                }
-            }
-        }
-        {
-            try {
-                await assertDirectAgentPreflight(args.cwd, args.suitesDir);
-            }
-            catch (error) {
-                console.error(error instanceof Error ? error.message : error);
-                return 1;
-            }
-            const inPlaceAllowed = process.env.AGENT_TEST_ALLOW_IN_PLACE === "1";
-            const worktreeDisabled = args.worktree === false ||
-                process.env.AGENT_TEST_NO_WORKTREE === "1" ||
-                process.env.AGENT_TEST_NO_WORKTREE === "true";
-            if (worktreeDisabled && !inPlaceAllowed) {
-                console.error("Direct agent tests require a sealed temp workspace. Set AGENT_TEST_ALLOW_IN_PLACE=1 to run in repo cwd (--no-worktree leaks agent edits into your working tree).");
-                return 1;
-            }
-            registerLiveRunHandlers();
-            if (!isChild) {
-                const removed = await cleanupStaleScenarioWorktrees(args.cwd);
-                if (removed.length > 0) {
-                    console.log(theme.warn(`Cleaned ${removed.length} stale agent-test worktree(s) from a prior crash`));
-                }
-                console.log(theme.banner(args.debug ? "direct debug" : "direct"));
-                if (stagingSessionRoot) {
-                    console.log(theme.bannerSession(stagingSessionRoot));
-                }
-                const hints = ["Ctrl+C cancels"];
-                if (args.debug || process.env.AGENT_TEST_VERBOSE === "1") {
-                    hints.push(args.keepRecordings ? "recordings kept" : "traces removed on exit");
-                    hints.push("exit 137 = macOS OOM");
-                }
-                else if (!args.keepRecordings) {
-                    hints.push("traces removed on exit");
-                }
-                console.log(theme.bannerHints(hints));
-                if (worktreeDisabled) {
-                    console.warn(theme.warn("running in repo cwd — agent file edits will persist in your working tree"));
-                }
-            }
-        }
-        const reports = await runAllSuites({
-            ...args,
-            stagingSessionId,
-            timeoutMs: args.timeoutMs,
-            allowUserInput: args.allowUserInput,
-            debug: args.debug,
-            debugDir: args.debugDir,
-        });
-        let exitCode = 0;
-        if (!isChild) {
-            for (const report of reports) {
-                const failed = report.results.filter((result) => !result.passed && !result.skipped);
-                if (failed.length > 0) {
-                    const behaviorFailures = failed.filter((result) => shouldFailScenario(result.failures, args.failOn));
-                    if (behaviorFailures.length > 0) {
-                        exitCode = 1;
-                    }
-                    console.log(`\n${theme.failedScenariosHeader()}`);
-                    for (const result of failed) {
-                        console.log(theme.failedScenarioName(result.scenario));
-                        if (verbose) {
-                            for (const failure of result.failures) {
-                                console.log(theme.verboseFailure(failure.matcher, failure.message, failure.evidence, failure.category));
-                            }
-                        }
-                        if (args.debug && result.debugBundleDir) {
-                            console.log(`      ${theme.tip(result.debugBundleDir)}`);
-                        }
-                    }
-                }
-                console.log(theme.summary(report.suite, report.passed, report.failed, report.skipped));
-            }
-            const runSummary = summarizeReports(reports);
-            const runSummaryText = formatRunSummary(runSummary);
-            if (runSummaryText) {
-                console.log(`\n${theme.runSummary(runSummaryText)}`);
-            }
-            if (args.failOn === "behavior" && runSummary.infraFailures > 0 && exitCode === 0) {
-                console.log(theme.tip(`${runSummary.infraFailures} infra failure(s) ignored (--fail-on=behavior)`));
-            }
-            if (args.debug && stagingSessionRoot) {
-                console.log(`\n${theme.tip(`debug session: ${stagingSessionRoot}`)}`);
-            }
-            if (args.htmlReport && reports.length > 0) {
-                try {
-                    const catalog = await loadViewerCatalog({
-                        cwd: args.cwd,
-                        suitesDir: args.suitesDir,
-                        rubricsDir: args.rubricsDir,
-                    }).catch(() => undefined);
-                    const reportPath = await writeHtmlReport(reports, {
-                        host: args.host,
-                        suitesDir: args.suitesDir,
-                        ...(catalog ? { catalog } : {}),
-                    }, resolveHtmlReportWritePath(args.htmlReport, args.reportOut));
-                    console.log(`\n${await htmlReportTip("View report", reportPath, reports, args)}`);
-                }
-                catch (error) {
-                    console.warn(theme.warn(`HTML report failed: ${error instanceof Error ? error.message : String(error)}`));
-                }
-            }
-        }
-        else {
-            for (const report of reports) {
-                const failed = report.results.filter((result) => !result.passed && !result.skipped);
-                if (failed.some((result) => shouldFailScenario(result.failures, args.failOn))) {
-                    exitCode = 1;
-                }
-            }
-        }
-        if (reports.length === 0) {
-            logProgress(`No suites found under ${args.suitesDir}`);
-            return 1;
-        }
-        if (!isChild) {
-            await offerCursorSdkReloginIfNeeded({ reports });
-        }
-        return exitCode;
-    }
-    finally {
-        if (stagingSessionId && !isChild) {
-            await cleanupRunArtifacts(stagingSessionRoot, args.keepRecordings);
-        }
-    }
+    throw new Error(`Unsupported command: ${command}. Use agent-test test with a TypeScript suite. Run agent-test --help for commands.`);
 }
-const entry = fileURLToPath(import.meta.url);
-if (isCliMain(process.argv[1], entry)) {
-    void main()
-        .then((code) => process.exit(code))
+if (isCliMain(process.argv[1], fileURLToPath(import.meta.url))) {
+    runCli(process.argv.slice(2))
+        .then((code) => {
+        process.exitCode = code;
+    })
         .catch((error) => {
-        console.error(error instanceof Error ? error.message : error);
-        process.exit(1);
+        console.error(error instanceof Error ? error.message : String(error));
+        process.exitCode = 1;
     });
 }
 //# sourceMappingURL=cli.js.map