@cursor/july 0.1.16 → 0.1.18

package/dist/internal/cli-ax.d.ts

@@ -6,6 +6,7 @@
  * filesystem-first evals under evals/ (*.eval.ts). Live A/B metrics
  * (`defineAB` under `agent/ab`) plug into serve — there is no `ab` CLI.
  */
+import type { EvalAssertionResult } from "../evals.js";
 export interface AxCliOptions {
     cwd?: string;
     dir: string;
@@ -36,6 +37,19 @@ export interface AxCliOptions {
     tags?: string[];
     /** Eval ids / path prefixes. */
     evalIds?: string[];
+    /** `eval`: a soft threshold miss (`scored`) also fails the exit code. */
+    strict?: boolean;
+    /** `eval`: write JUnit XML to this path. */
+    junit?: string;
+    /**
+     * `eval`: run-artifact directory. Defaults to a timestamped directory under
+     * `<dir>/.agent-serve/evals/`; `false` disables artifacts entirely.
+     */
+    artifacts?: string | false;
+    /** `eval`: skip reporters declared in `evals.config.ts` and eval files. */
+    skipReport?: boolean;
+    /** `eval`: override `evals.config.ts` `maxConcurrency`. */
+    maxConcurrency?: number;
     /** Scaffold target directory for `init`. */
     initTarget?: string;
     /** Tool name for `call`. */
@@ -145,6 +159,19 @@ export declare function cmdEvalCancel(runId: string | undefined, options: AxCliO
  * the batch is still running (for poll loops after `--no-wait`).
  */
 export declare function cmdEvalStatus(runId: string | undefined, options: AxCliOptions): Promise<number>;
+/**
+ * Where a local run writes its artifacts.
+ *
+ * Defaults to a timestamped directory under the project's gitignored
+ * `.agent-serve/evals/`, so the full story of a failure is always on disk
+ * without the eval author opting in. `--no-artifacts` turns it off.
+ */
+export declare function resolveEvalArtifactsDir(projectRoot: string, options: Pick<AxCliOptions, "artifacts">): string | undefined;
+/**
+ * One assertion line. Soft assertions read `~` rather than `✗` when they miss,
+ * so a tracked score is visibly different from a failed gate.
+ */
+export declare function formatEvalAssertionLine(assertion: EvalAssertionResult): string;
 /**
  * `logs` — follow deploy progress (`--prod`) then runtime agent logs.
  * Local / `--url` tails `/v1/logs` only.

package/dist/internal/cli-ax.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"cli-ax.d.ts","sourceRoot":"","sources":["../../src/internal/cli-ax.ts"],"names":[],"mappings":"AAAA;;;;;;;GAOG;AA6DH,MAAM,WAAW,YAAY;IAC3B,GAAG,CAAC,EAAE,MAAM,CAAC;IACb,GAAG,EAAE,MAAM,CAAC;IACZ;;;;OAIG;IACH,IAAI,CAAC,EAAE,MAAM,CAAC;IACd,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,8BAA8B;IAC9B,QAAQ,CAAC,EAAE,MAAM,EAAE,CAAC;IACpB,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,iBAAiB,CAAC,EAAE,MAAM,CAAC;IAC3B,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB;;;;;OAKG;IACH,MAAM,CAAC,EAAE,MAAM,GAAG,IAAI,GAAG,KAAK,CAAC;IAC/B,IAAI,CAAC,EAAE,OAAO,CAAC;IACf,IAAI,CAAC,EAAE,OAAO,CAAC;IACf,IAAI,CAAC,EAAE,OAAO,CAAC;IACf,OAAO,CAAC,EAAE,OAAO,CAAC;IAClB,IAAI,CAAC,EAAE,MAAM,EAAE,CAAC;IAChB,gCAAgC;IAChC,OAAO,CAAC,EAAE,MAAM,EAAE,CAAC;IACnB,4CAA4C;IAC5C,UAAU,CAAC,EAAE,MAAM,CAAC;IACpB,4BAA4B;IAC5B,QAAQ,CAAC,EAAE,MAAM,CAAC;IAClB,6BAA6B;IAC7B,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,iEAAiE;IACjE,GAAG,CAAC,EAAE,MAAM,CAAC;IACb;;;;;;OAMG;IACH,IAAI,CAAC,EAAE,OAAO,CAAC;IACf;;;;OAIG;IACH,MAAM,CAAC,EAAE,OAAO,CAAC;IACjB,oEAAoE;IACpE,IAAI,CAAC,EAAE,MAAM,CAAC;IACd,6DAA6D;IAC7D,MAAM,CAAC,EAAE,MAAM,CAAC;IAChB,yCAAyC;IACzC,SAAS,CAAC,EAAE,OAAO,KAAK,CAAC;IACzB;;;OAGG;IACH,MAAM,CAAC,EAAE,OAAO,CAAC;IACjB,6CAA6C;IAC7C,OAAO,CAAC,EAAE,OAAO,CAAC;IAClB,4DAA4D;IAC5D,IAAI,CAAC,EAAE,OAAO,CAAC;IACf;;;OAGG;IACH,SAAS,CAAC,EAAE,OAAO,CAAC;IACpB;;;OAGG;IACH,KAAK,CAAC,EAAE,OAAO,CAAC;IAChB;;;OAGG;IACH,MAAM,CAAC,EAAE,OAAO,CAAC;IACjB,sEAAsE;IACtE,MAAM,CAAC,EAAE,WAAW,CAAC;CACtB;AAED,MAAM,WAAW,mBAAmB;IAClC,QAAQ,EAAE,MAAM,CAAC;IACjB,OAAO,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC;IACjC,KAAK,CAAC,EAAE,MAAM,OAAO,CAAC,IAAI,CAAC,CAAC;CAC7B;AAsED,wBAAsB,aAAa,CACjC,OAAO,EAAE,YAAY,GACpB,OAAO,CAAC,mBAAmB,CAAC,CAmB9B;AAED,wBAAsB,OAAO,CAAC,OAAO,EAAE,YAAY,GAAG,OAAO,CAAC,MAAM,CAAC,CAoFpE;AAED;;;GAGG;AACH;;;;;;;;GAQG;AACH,wBAAsB,SAAS,CAC7B,SAAS,EAAE,MAAM,GAAG,SAAS,EAC7B,OAAO,EAAE,YAAY,GACpB,OAAO,CAAC,MAAM,CAAC,CAcjB;AAUD,wBAAsB,OAAO,CAAC,OAAO,EAAE,YAAY,GAAG,OAAO,CAAC,MAAM,CAAC,CA8GpE;AAED,wBAAsB,MAAM,CAAC,OAAO,EAAE,YAAY,GAAG,OAAO,CAAC,MAAM,CAAC,CA4GnE;AAED;;;GAGG;AACH,wBAAsB,OAAO,CAAC,OAAO,EAAE,YAAY,GAAG,OAAO,CAAC,MAAM,CAAC,CA0DpE;AAED,wBAAsB,aAAa,CAAC,OAAO,EAAE,YAAY,GAAG,OAAO,CAAC,MAAM,CAAC,CAe1E;AAED;;;;GAIG;AACH,wBAAgB,qBAAqB,CACnC,OAAO,EAAE,IAAI,CAAC,YAAY,EAAE,MAAM,GAAG,KAAK,CAAC,GAC1C,OAAO,GAAG,KAAK,GAAG,YAAY,GAAG;IAAE,KAAK,EAAE,MAAM,CAAA;CAAE,CAepD;AAiDD,wBAAsB,OAAO,CAAC,OAAO,EAAE,YAAY,GAAG,OAAO,CAAC,MAAM,CAAC,CA4LpE;AAED;;;GAGG;AACH,wBAAgB,qBAAqB,CACnC,OAAO,EAAE,IAAI,CAAC,YAAY,EAAE,MAAM,GAAG,KAAK,GAAG,MAAM,GAAG,MAAM,CAAC,GAC5D,MAAM,CAcR;AAsID;;GAEG;AACH,wBAAsB,aAAa,CACjC,KAAK,EAAE,MAAM,GAAG,SAAS,EACzB,OAAO,EAAE,YAAY,GACpB,OAAO,CAAC,MAAM,CAAC,CAwDjB;AAED;;;;;GAKG;AACH,wBAAsB,aAAa,CACjC,KAAK,EAAE,MAAM,GAAG,SAAS,EACzB,OAAO,EAAE,YAAY,GACpB,OAAO,CAAC,MAAM,CAAC,CA8FjB;AA8KD;;;GAGG;AACH,wBAAsB,OAAO,CAAC,OAAO,EAAE,YAAY,GAAG,OAAO,CAAC,MAAM,CAAC,CAgFpE;AAmCD;;GAEG;AACH,wBAAsB,WAAW,CAAC,OAAO,EAAE,YAAY,GAAG,OAAO,CAAC,MAAM,CAAC,CAkCxE;AAED;;;;GAIG;AACH,wBAAsB,UAAU,CAC9B,SAAS,EAAE,MAAM,GAAG,SAAS,EAC7B,OAAO,EAAE,YAAY,GACpB,OAAO,CAAC,MAAM,CAAC,CAkDjB;AAED;;;;;;GAMG;AACH,wBAAsB,aAAa,CAAC,OAAO,EAAE,YAAY,GAAG,OAAO,CAAC,MAAM,CAAC,CAoF1E"}
+{"version":3,"file":"cli-ax.d.ts","sourceRoot":"","sources":["../../src/internal/cli-ax.ts"],"names":[],"mappings":"AAAA;;;;;;;GAOG;AAMH,OAAO,KAAK,EACV,mBAAmB,EAIpB,MAAM,aAAa,CAAC;AAwDrB,MAAM,WAAW,YAAY;IAC3B,GAAG,CAAC,EAAE,MAAM,CAAC;IACb,GAAG,EAAE,MAAM,CAAC;IACZ;;;;OAIG;IACH,IAAI,CAAC,EAAE,MAAM,CAAC;IACd,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,8BAA8B;IAC9B,QAAQ,CAAC,EAAE,MAAM,EAAE,CAAC;IACpB,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,iBAAiB,CAAC,EAAE,MAAM,CAAC;IAC3B,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB;;;;;OAKG;IACH,MAAM,CAAC,EAAE,MAAM,GAAG,IAAI,GAAG,KAAK,CAAC;IAC/B,IAAI,CAAC,EAAE,OAAO,CAAC;IACf,IAAI,CAAC,EAAE,OAAO,CAAC;IACf,IAAI,CAAC,EAAE,OAAO,CAAC;IACf,OAAO,CAAC,EAAE,OAAO,CAAC;IAClB,IAAI,CAAC,EAAE,MAAM,EAAE,CAAC;IAChB,gCAAgC;IAChC,OAAO,CAAC,EAAE,MAAM,EAAE,CAAC;IACnB,yEAAyE;IACzE,MAAM,CAAC,EAAE,OAAO,CAAC;IACjB,4CAA4C;IAC5C,KAAK,CAAC,EAAE,MAAM,CAAC;IACf;;;OAGG;IACH,SAAS,CAAC,EAAE,MAAM,GAAG,KAAK,CAAC;IAC3B,2EAA2E;IAC3E,UAAU,CAAC,EAAE,OAAO,CAAC;IACrB,2DAA2D;IAC3D,cAAc,CAAC,EAAE,MAAM,CAAC;IACxB,4CAA4C;IAC5C,UAAU,CAAC,EAAE,MAAM,CAAC;IACpB,4BAA4B;IAC5B,QAAQ,CAAC,EAAE,MAAM,CAAC;IAClB,6BAA6B;IAC7B,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,iEAAiE;IACjE,GAAG,CAAC,EAAE,MAAM,CAAC;IACb;;;;;;OAMG;IACH,IAAI,CAAC,EAAE,OAAO,CAAC;IACf;;;;OAIG;IACH,MAAM,CAAC,EAAE,OAAO,CAAC;IACjB,oEAAoE;IACpE,IAAI,CAAC,EAAE,MAAM,CAAC;IACd,6DAA6D;IAC7D,MAAM,CAAC,EAAE,MAAM,CAAC;IAChB,yCAAyC;IACzC,SAAS,CAAC,EAAE,OAAO,KAAK,CAAC;IACzB;;;OAGG;IACH,MAAM,CAAC,EAAE,OAAO,CAAC;IACjB,6CAA6C;IAC7C,OAAO,CAAC,EAAE,OAAO,CAAC;IAClB,4DAA4D;IAC5D,IAAI,CAAC,EAAE,OAAO,CAAC;IACf;;;OAGG;IACH,SAAS,CAAC,EAAE,OAAO,CAAC;IACpB;;;OAGG;IACH,KAAK,CAAC,EAAE,OAAO,CAAC;IAChB;;;OAGG;IACH,MAAM,CAAC,EAAE,OAAO,CAAC;IACjB,sEAAsE;IACtE,MAAM,CAAC,EAAE,WAAW,CAAC;CACtB;AAED,MAAM,WAAW,mBAAmB;IAClC,QAAQ,EAAE,MAAM,CAAC;IACjB,OAAO,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC;IACjC,KAAK,CAAC,EAAE,MAAM,OAAO,CAAC,IAAI,CAAC,CAAC;CAC7B;AAsED,wBAAsB,aAAa,CACjC,OAAO,EAAE,YAAY,GACpB,OAAO,CAAC,mBAAmB,CAAC,CAmB9B;AAED,wBAAsB,OAAO,CAAC,OAAO,EAAE,YAAY,GAAG,OAAO,CAAC,MAAM,CAAC,CAoFpE;AAED;;;GAGG;AACH;;;;;;;;GAQG;AACH,wBAAsB,SAAS,CAC7B,SAAS,EAAE,MAAM,GAAG,SAAS,EAC7B,OAAO,EAAE,YAAY,GACpB,OAAO,CAAC,MAAM,CAAC,CAcjB;AAUD,wBAAsB,OAAO,CAAC,OAAO,EAAE,YAAY,GAAG,OAAO,CAAC,MAAM,CAAC,CA8GpE;AAED,wBAAsB,MAAM,CAAC,OAAO,EAAE,YAAY,GAAG,OAAO,CAAC,MAAM,CAAC,CA4GnE;AAED;;;GAGG;AACH,wBAAsB,OAAO,CAAC,OAAO,EAAE,YAAY,GAAG,OAAO,CAAC,MAAM,CAAC,CA0DpE;AAED,wBAAsB,aAAa,CAAC,OAAO,EAAE,YAAY,GAAG,OAAO,CAAC,MAAM,CAAC,CAe1E;AAED;;;;GAIG;AACH,wBAAgB,qBAAqB,CACnC,OAAO,EAAE,IAAI,CAAC,YAAY,EAAE,MAAM,GAAG,KAAK,CAAC,GAC1C,OAAO,GAAG,KAAK,GAAG,YAAY,GAAG;IAAE,KAAK,EAAE,MAAM,CAAA;CAAE,CAepD;AAiDD,wBAAsB,OAAO,CAAC,OAAO,EAAE,YAAY,GAAG,OAAO,CAAC,MAAM,CAAC,CAkMpE;AAED;;;GAGG;AACH,wBAAgB,qBAAqB,CACnC,OAAO,EAAE,IAAI,CAAC,YAAY,EAAE,MAAM,GAAG,KAAK,GAAG,MAAM,GAAG,MAAM,CAAC,GAC5D,MAAM,CAcR;AAsID;;GAEG;AACH,wBAAsB,aAAa,CACjC,KAAK,EAAE,MAAM,GAAG,SAAS,EACzB,OAAO,EAAE,YAAY,GACpB,OAAO,CAAC,MAAM,CAAC,CAwDjB;AAED;;;;;GAKG;AACH,wBAAsB,aAAa,CACjC,KAAK,EAAE,MAAM,GAAG,SAAS,EACzB,OAAO,EAAE,YAAY,GACpB,OAAO,CAAC,MAAM,CAAC,CA8FjB;AAmGD;;;;;;GAMG;AACH,wBAAgB,uBAAuB,CACrC,WAAW,EAAE,MAAM,EACnB,OAAO,EAAE,IAAI,CAAC,YAAY,EAAE,WAAW,CAAC,GACvC,MAAM,GAAG,SAAS,CASpB;AA8HD;;;GAGG;AACH,wBAAgB,uBAAuB,CACrC,SAAS,EAAE,mBAAmB,GAC7B,MAAM,CAgBR;AAED;;;GAGG;AACH,wBAAsB,OAAO,CAAC,OAAO,EAAE,YAAY,GAAG,OAAO,CAAC,MAAM,CAAC,CAgFpE;AAmCD;;GAEG;AACH,wBAAsB,WAAW,CAAC,OAAO,EAAE,YAAY,GAAG,OAAO,CAAC,MAAM,CAAC,CAkCxE;AAED;;;;GAIG;AACH,wBAAsB,UAAU,CAC9B,SAAS,EAAE,MAAM,GAAG,SAAS,EAC7B,OAAO,EAAE,YAAY,GACpB,OAAO,CAAC,MAAM,CAAC,CAkDjB;AAED;;;;;;GAMG;AACH,wBAAsB,aAAa,CAAC,OAAO,EAAE,YAAY,GAAG,OAAO,CAAC,MAAM,CAAC,CAoF1E"}

package/dist/internal/cli-ax.js

@@ -18,6 +18,7 @@ var __awaiter = (this && this.__awaiter) || function (thisArg, _arguments, P, ge
 import { mkdir, mkdtemp, writeFile } from "node:fs/promises";
 import { tmpdir } from "node:os";
 import { join, resolve } from "node:path";
+import { Artifacts, JUnit } from "../evals/reporters.js";
 import { serve } from "../index.js";
 import { lookupSessionContinuation, runChat } from "./chat-client.js";
 import { cmdLogin } from "./cli-cursor.js";
@@ -616,6 +617,7 @@ export function cmdEval(options) {
             // batch lands in playground history and defineStorage evals (when set).
             if (mode === "local") {
                 const progress = stream ? createStreamProgress() : undefined;
+                const artifactsDir = resolveEvalArtifactsDir(projectRoot, options);
                 try {
                     const results = yield runDiscoveredEvals({
                         projectRoot,
@@ -623,6 +625,11 @@ export function cmdEval(options) {
                         filterIds,
                         tags,
                         timeoutMs: options.timeoutMs,
+                        maxConcurrency: options.maxConcurrency,
+                        strict: options.strict === true,
+                        skipReport: options.skipReport === true,
+                        reporters: buildCliEvalReporters(options, artifactsDir),
+                        target: { baseUrl: target.agentUrl, mode: "local" },
                         bearerToken: options.bearerToken,
                         headers: target.headers,
                         verbose: Boolean(options.verbose) || stream,
@@ -632,7 +639,7 @@ export function cmdEval(options) {
                         process.stderr.write("No matching evals found.\n");
                         return 2;
                     }
-                    return printEvalResults(results, options);
+                    return printEvalResults(results, options, undefined, artifactsDir);
                 }
                 finally {
                     progress === null || progress === void 0 ? void 0 : progress.close();
@@ -1031,33 +1038,76 @@ function createRemoteEvalProgress(enabled) {
         },
     };
 }
-function printEvalResults(results, options, batch) {
+/**
+ * Where a local run writes its artifacts.
+ *
+ * Defaults to a timestamped directory under the project's gitignored
+ * `.agent-serve/evals/`, so the full story of a failure is always on disk
+ * without the eval author opting in. `--no-artifacts` turns it off.
+ */
+export function resolveEvalArtifactsDir(projectRoot, options) {
+    if (options.artifacts === false) {
+        return undefined;
+    }
+    if (typeof options.artifacts === "string" && options.artifacts !== "") {
+        return resolve(options.artifacts);
+    }
+    const stamp = new Date().toISOString().replace(/[:.]/g, "-");
+    return join(projectRoot, ".agent-serve", "evals", stamp);
+}
+/** JUnit / artifacts reporters requested via CLI flags. */
+function buildCliEvalReporters(options, artifactsDir) {
+    const reporters = [];
+    if (artifactsDir !== undefined) {
+        reporters.push(Artifacts({ dir: artifactsDir }));
+    }
+    if (options.junit !== undefined && options.junit !== "") {
+        reporters.push(JUnit({ filePath: options.junit }));
+    }
+    return reporters;
+}
+function evalVerdictOf(result) {
+    var _a;
+    return (_a = result.verdict) !== null && _a !== void 0 ? _a : (result.ok ? "passed" : "failed");
+}
+const EVAL_VERDICT_LABEL = {
+    passed: "PASS",
+    failed: "FAIL",
+    scored: "SCORED",
+    skipped: "SKIP",
+};
+function printEvalResults(results, options, batch, artifactsDir) {
     var _a, _b;
-    const passed = results.filter((r) => r.ok).length;
-    const failed = results.length - passed;
+    const strict = options.strict === true;
+    const counts = { passed: 0, failed: 0, scored: 0, skipped: 0 };
+    for (const result of results) {
+        counts[evalVerdictOf(result)]++;
+    }
     const batchFailed = (batch === null || batch === void 0 ? void 0 : batch.status) === "failed";
-    const ok = failed === 0 && !batchFailed;
+    const ok = counts.failed === 0 && !(strict && counts.scored > 0) && !batchFailed;
     if ((batch === null || batch === void 0 ? void 0 : batch.error) !== undefined && batch.error !== "" && !options.json) {
         // Always surface harness-level failures (not only --verbose).
         process.stderr.write(`eval run error: ${batch.error}\n`);
     }
     if (options.json) {
-        console.log(JSON.stringify(Object.assign(Object.assign(Object.assign({ ok,
-            passed,
-            failed,
-            results }, ((batch === null || batch === void 0 ? void 0 : batch.runId) === undefined ? {} : { runId: batch.runId })), ((batch === null || batch === void 0 ? void 0 : batch.status) === undefined ? {} : { status: batch.status })), ((batch === null || batch === void 0 ? void 0 : batch.error) === undefined || batch.error === ""
+        console.log(JSON.stringify(Object.assign(Object.assign(Object.assign(Object.assign({ ok, passed: counts.passed, failed: counts.failed, scored: counts.scored, skipped: counts.skipped, strict,
+            results }, (artifactsDir === undefined ? {} : { artifactsDir })), ((batch === null || batch === void 0 ? void 0 : batch.runId) === undefined ? {} : { runId: batch.runId })), ((batch === null || batch === void 0 ? void 0 : batch.status) === undefined ? {} : { status: batch.status })), ((batch === null || batch === void 0 ? void 0 : batch.error) === undefined || batch.error === ""
             ? {}
             : { error: batch.error })), null, 2));
     }
     else {
         for (const r of results) {
-            process.stdout.write(`${r.ok ? "PASS" : "FAIL"}  ${r.id} (${(r.durationMs / 1000).toFixed(1)}s)`);
+            const verdict = evalVerdictOf(r);
+            process.stdout.write(`${EVAL_VERDICT_LABEL[verdict]}  ${r.id} (${(r.durationMs / 1000).toFixed(1)}s)`);
             if (r.description !== undefined) {
                 process.stdout.write(` — ${r.description}`);
             }
             process.stdout.write("\n");
+            if (r.skipReason !== undefined) {
+                process.stdout.write(`  skipped: ${r.skipReason}\n`);
+            }
             for (const a of r.assertions) {
-                process.stdout.write(`  ${a.passed ? "✓" : "✗"} ${a.name}${a.detail !== undefined ? ` (${a.detail})` : ""}\n`);
+                process.stdout.write(`  ${formatEvalAssertionLine(a)}\n`);
             }
             for (const [name, value] of Object.entries((_a = r.metrics) !== null && _a !== void 0 ? _a : {})) {
                 process.stdout.write(`  metric ${name}=${String(value)}\n`);
@@ -1072,10 +1122,38 @@ function printEvalResults(results, options, batch) {
                 process.stdout.write(`  reply: ${r.finalText.slice(0, 200).replace(/\s+/g, " ")}\n`);
             }
         }
-        process.stdout.write(`\n${passed} passed, ${failed} failed, ${results.length} total\n`);
+        const extra = [];
+        if (counts.scored > 0) {
+            extra.push(`${counts.scored} scored${strict ? " (fatal: --strict)" : ""}`);
+        }
+        if (counts.skipped > 0) {
+            extra.push(`${counts.skipped} skipped`);
+        }
+        process.stdout.write(`\n${counts.passed} passed, ${counts.failed} failed${extra.length === 0 ? "" : `, ${extra.join(", ")}`}, ${results.length} total\n`);
+        if (artifactsDir !== undefined) {
+            process.stdout.write(`artifacts: ${artifactsDir}\n`);
+        }
     }
     return ok ? 0 : 1;
 }
+/**
+ * One assertion line. Soft assertions read `~` rather than `✗` when they miss,
+ * so a tracked score is visibly different from a failed gate.
+ */
+export function formatEvalAssertionLine(assertion) {
+    var _a;
+    const soft = ((_a = assertion.severity) !== null && _a !== void 0 ? _a : "gate") === "soft";
+    const mark = assertion.passed ? "✓" : soft ? "~" : "✗";
+    const parts = [];
+    if (assertion.score !== undefined) {
+        parts.push(`score=${assertion.score.toFixed(3)}${assertion.threshold === undefined ? "" : ` >= ${assertion.threshold}`}`);
+    }
+    if (assertion.detail !== undefined) {
+        parts.push(assertion.detail);
+    }
+    const suffix = parts.length === 0 ? "" : ` (${parts.join(", ")})`;
+    return `${mark} ${assertion.name}${soft ? " [soft]" : ""}${suffix}`;
+}
 /**
  * `logs` — follow deploy progress (`--prod`) then runtime agent logs.
  * Local / `--url` tails `/v1/logs` only.

package/dist/internal/cli-mcp.d.ts

@@ -25,6 +25,7 @@
  * stdout is reserved for the MCP wire in serve mode; all logging goes to
  * stderr.
  */
+import { resolveDeployAuth } from "./cli-deploy.js";
 import { resolveProdTarget } from "./resolve-prod-target.js";
 export interface McpCliOptions {
     dir: string;
@@ -43,9 +44,16 @@ export interface McpCliOptions {
     print?: boolean;
     /** `install`: machine-readable output. */
     json?: boolean;
+    /**
+     * `install --prod`: write a REMOTE HTTP entry pointing at the stable Cursor
+     * MCP gateway (`<api>/…/mcp/<team>/<slug>`) instead of the local stdio
+     * proxy. For remote-only MCP hosts (Glass) that cannot spawn stdio.
+     */
+    remote?: boolean;
     /** Injected seams for tests. */
     fetchImpl?: typeof fetch;
     resolveProdTargetImpl?: typeof resolveProdTarget;
+    resolveDeployAuthImpl?: typeof resolveDeployAuth;
     stdin?: NodeJS.ReadableStream;
     stdout?: {
         write(chunk: string): unknown;

package/dist/internal/cli-mcp.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"cli-mcp.d.ts","sourceRoot":"","sources":["../../src/internal/cli-mcp.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;;;;;;;GA0BG;AAMH,OAAO,EAGL,iBAAiB,EAClB,MAAM,0BAA0B,CAAC;AA0BlC,MAAM,WAAW,aAAa;IAC5B,GAAG,EAAE,MAAM,CAAC;IACZ,GAAG,CAAC,EAAE,MAAM,CAAC;IACb,gFAAgF;IAChF,IAAI,CAAC,EAAE,MAAM,CAAC;IACd,IAAI,CAAC,EAAE,MAAM,CAAC;IACd,MAAM,CAAC,EAAE,MAAM,CAAC;IAChB,sEAAsE;IACtE,GAAG,CAAC,EAAE,MAAM,CAAC;IACb,IAAI,CAAC,EAAE,OAAO,CAAC;IACf,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,4EAA4E;IAC5E,IAAI,CAAC,EAAE,MAAM,CAAC;IACd,wEAAwE;IACxE,KAAK,CAAC,EAAE,OAAO,CAAC;IAChB,0CAA0C;IAC1C,IAAI,CAAC,EAAE,OAAO,CAAC;IACf,gCAAgC;IAChC,SAAS,CAAC,EAAE,OAAO,KAAK,CAAC;IACzB,qBAAqB,CAAC,EAAE,OAAO,iBAAiB,CAAC;IACjD,KAAK,CAAC,EAAE,MAAM,CAAC,cAAc,CAAC;IAC9B,MAAM,CAAC,EAAE;QAAE,KAAK,CAAC,KAAK,EAAE,MAAM,GAAG,OAAO,CAAA;KAAE,CAAC;IAC3C,GAAG,CAAC,EAAE,CAAC,IAAI,EAAE,MAAM,KAAK,IAAI,CAAC;IAC7B,GAAG,CAAC,EAAE,MAAM,MAAM,CAAC;IACnB,KAAK,CAAC,EAAE,CAAC,EAAE,EAAE,MAAM,KAAK,OAAO,CAAC,IAAI,CAAC,CAAC;CACvC;AAED,wBAAsB,MAAM,CAC1B,UAAU,EAAE,MAAM,GAAG,SAAS,EAC9B,OAAO,EAAE,aAAa,GAAG;IACvB,eAAe,CAAC,EAAE,MAAM,CAAC;IACzB,KAAK,CAAC,EAAE,OAAO,CAAC;CACjB,GACA,OAAO,CAAC,MAAM,CAAC,CAiCjB;AAMD,wBAAsB,WAAW,CAAC,OAAO,EAAE,aAAa,GAAG,OAAO,CAAC,MAAM,CAAC,CAkFzE;AAyRD,wBAAsB,aAAa,CAAC,OAAO,EAAE,aAAa,GAAG,OAAO,CAAC,MAAM,CAAC,CAqG3E;AAED,+DAA+D;AAC/D,wBAAgB,iBAAiB,CAC/B,IAAI,GAAE,MAAsC,GAC3C,MAAM,CAGR;AAqED,iFAAiF;AACjF,wBAAgB,WAAW,CAAC,OAAO,EAAE,MAAM,GAAG,MAAM,CAQnD"}
+{"version":3,"file":"cli-mcp.d.ts","sourceRoot":"","sources":["../../src/internal/cli-mcp.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;;;;;;;GA0BG;AAKH,OAAO,EAAE,iBAAiB,EAAE,MAAM,iBAAiB,CAAC;AAGpD,OAAO,EAGL,iBAAiB,EAClB,MAAM,0BAA0B,CAAC;AA0BlC,MAAM,WAAW,aAAa;IAC5B,GAAG,EAAE,MAAM,CAAC;IACZ,GAAG,CAAC,EAAE,MAAM,CAAC;IACb,gFAAgF;IAChF,IAAI,CAAC,EAAE,MAAM,CAAC;IACd,IAAI,CAAC,EAAE,MAAM,CAAC;IACd,MAAM,CAAC,EAAE,MAAM,CAAC;IAChB,sEAAsE;IACtE,GAAG,CAAC,EAAE,MAAM,CAAC;IACb,IAAI,CAAC,EAAE,OAAO,CAAC;IACf,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,4EAA4E;IAC5E,IAAI,CAAC,EAAE,MAAM,CAAC;IACd,wEAAwE;IACxE,KAAK,CAAC,EAAE,OAAO,CAAC;IAChB,0CAA0C;IAC1C,IAAI,CAAC,EAAE,OAAO,CAAC;IACf;;;;OAIG;IACH,MAAM,CAAC,EAAE,OAAO,CAAC;IACjB,gCAAgC;IAChC,SAAS,CAAC,EAAE,OAAO,KAAK,CAAC;IACzB,qBAAqB,CAAC,EAAE,OAAO,iBAAiB,CAAC;IACjD,qBAAqB,CAAC,EAAE,OAAO,iBAAiB,CAAC;IACjD,KAAK,CAAC,EAAE,MAAM,CAAC,cAAc,CAAC;IAC9B,MAAM,CAAC,EAAE;QAAE,KAAK,CAAC,KAAK,EAAE,MAAM,GAAG,OAAO,CAAA;KAAE,CAAC;IAC3C,GAAG,CAAC,EAAE,CAAC,IAAI,EAAE,MAAM,KAAK,IAAI,CAAC;IAC7B,GAAG,CAAC,EAAE,MAAM,MAAM,CAAC;IACnB,KAAK,CAAC,EAAE,CAAC,EAAE,EAAE,MAAM,KAAK,OAAO,CAAC,IAAI,CAAC,CAAC;CACvC;AAED,wBAAsB,MAAM,CAC1B,UAAU,EAAE,MAAM,GAAG,SAAS,EAC9B,OAAO,EAAE,aAAa,GAAG;IACvB,eAAe,CAAC,EAAE,MAAM,CAAC;IACzB,KAAK,CAAC,EAAE,OAAO,CAAC;CACjB,GACA,OAAO,CAAC,MAAM,CAAC,CAiCjB;AAMD,wBAAsB,WAAW,CAAC,OAAO,EAAE,aAAa,GAAG,OAAO,CAAC,MAAM,CAAC,CAkFzE;AAwRD,wBAAsB,aAAa,CAAC,OAAO,EAAE,aAAa,GAAG,OAAO,CAAC,MAAM,CAAC,CA4G3E;AAoGD,+DAA+D;AAC/D,wBAAgB,iBAAiB,CAC/B,IAAI,GAAE,MAAsC,GAC3C,MAAM,CAGR;AAqED,iFAAiF;AACjF,wBAAgB,WAAW,CAAC,OAAO,EAAE,MAAM,GAAG,MAAM,CAQnD"}

package/dist/internal/cli-mcp.js

@@ -37,6 +37,8 @@ var __awaiter = (this && this.__awaiter) || function (thisArg, _arguments, P, ge
 import { mkdir, readFile, writeFile } from "node:fs/promises";
 import { homedir } from "node:os";
 import { dirname, join } from "node:path";
+import { resolveDeployAuth } from "./cli-deploy.js";
+import { DeployApiError } from "./deploy-client.js";
 import { CLI_COMMAND_NAME as CLI, LOG_PREFIX } from "./distribution.js";
 import { ProdTargetError, resolveProdSlug, resolveProdTarget, } from "./resolve-prod-target.js";
 /** Mount path of the MCP surface relative to an agent base URL. */
@@ -388,6 +390,13 @@ export function cmdMcpInstall(options) {
             err("--prod and --url are mutually exclusive\n");
             return 1;
         }
+        if (options.remote === true) {
+            if (options.prod !== true) {
+                err("--remote requires --prod (the hosted deployment MCP gateway)\n");
+                return 1;
+            }
+            return cmdMcpInstallRemote(options);
+        }
         let name;
         let entry;
         if (options.prod === true) {
@@ -475,6 +484,95 @@ export function cmdMcpInstall(options) {
         return 0;
     });
 }
+/**
+ * `mcp install --prod --remote` — write a REMOTE HTTP entry pointing at the
+ * stable Cursor MCP gateway for the hosted deployment, for MCP hosts that
+ * cannot spawn the stdio proxy (Glass / Sand). The gateway mints/refreshes
+ * engineAccess server-side, so the entry never carries a pod hostname or a
+ * short-lived port token — only the durable, stable gateway URL.
+ *
+ * Auth: this headless/CLI entry sends the `agentkit login` API key as a
+ * bearer against the user-API-key gateway surface. That key is written to
+ * `mcp.json` in plain text (flagged below); a signed-in Cursor client
+ * (Glass) can instead use the session gateway surface with no header at all
+ * once first-party session attachment ships (see PR follow-ups).
+ */
+function cmdMcpInstallRemote(options) {
+    return __awaiter(this, void 0, void 0, function* () {
+        var _a, _b, _c;
+        const err = (_a = options.err) !== null && _a !== void 0 ? _a : defaultErr;
+        const stdout = (_b = options.stdout) !== null && _b !== void 0 ? _b : process.stdout;
+        let slug;
+        try {
+            slug = resolveProdSlug({
+                dir: options.dir,
+                cwd: options.cwd,
+                slug: options.slug,
+            });
+        }
+        catch (error) {
+            err(`${describeError(error)}\n`);
+            return 1;
+        }
+        const resolveAuthImpl = (_c = options.resolveDeployAuthImpl) !== null && _c !== void 0 ? _c : resolveDeployAuth;
+        const auth = yield resolveAuthImpl({
+            team: options.team,
+            apiKey: options.apiKey,
+            fetchImpl: options.fetchImpl,
+            err,
+        });
+        if (typeof auth === "number") {
+            return auth;
+        }
+        // Verify the deployment resolves before writing config that cannot work.
+        // This read is admin-tier, but calling the gateway only needs team
+        // membership — so a 403 means "cannot verify", not "cannot use": warn and
+        // still write the entry rather than locking members out of their own agent.
+        try {
+            yield auth.client.getDeployment({ teamId: auth.teamId, slug });
+        }
+        catch (error) {
+            if (error instanceof DeployApiError && error.status === 403) {
+                err(`Could not verify deployment "${slug}" (reading deployments needs team-admin); writing the entry anyway.\n`);
+            }
+            else {
+                err(`${describeError(error)}\n`);
+                if (error instanceof DeployApiError && error.status === 404) {
+                    err(`Deploy it first: ${CLI} deploy --slug ${slug}\n`);
+                }
+                return 1;
+            }
+        }
+        const name = resolveServerName(options.name, slug);
+        const url = auth.client.mcpGatewayUrl(auth.teamId, slug);
+        const entry = {
+            url,
+            headers: { Authorization: `Bearer ${auth.apiKey}` },
+        };
+        if (options.print === true) {
+            stdout.write(`${JSON.stringify({ mcpServers: { [name]: entry } }, null, 2)}\n`);
+            return 0;
+        }
+        const path = cursorMcpJsonPath();
+        try {
+            yield upsertMcpJsonEntry(path, name, entry);
+        }
+        catch (error) {
+            err(`${describeError(error)}\n`);
+            return 1;
+        }
+        err("note: the API key is written to mcp.json in plain text; it is a durable " +
+            "credential for a remote MCP host that cannot spawn the stdio proxy.\n");
+        if (options.json === true) {
+            stdout.write(`${JSON.stringify({ ok: true, path, name, entry })}\n`);
+            return 0;
+        }
+        stdout.write(`Installed remote MCP server "${name}" in ${path}\n` +
+            `  ${url}\n` +
+            "Cursor watches this file; the server appears without a restart.\n");
+        return 0;
+    });
+}
 /** `~/.cursor/mcp.json` — the user-level Cursor MCP config. */
 export function cursorMcpJsonPath(home) {
     var _a;

package/dist/internal/cli-slack.d.ts

@@ -1,6 +1,9 @@
 /**
- * `agent-serve slack …` CLI — manifests, setup, doctor (Socket Mode).
+ * `agent-serve slack …` CLI — manifests, setup, doctor (Socket Mode), and
+ * managed provisioning (`create` / `destroy`) through Cursor's
+ * "Add to Slack" API (`slack-provision-client.ts`).
  */
+import { runSlackDoctor } from "../channels/slack/index.js";
 export interface SlackCliOptions {
     dir: string;
     name?: string;
@@ -13,6 +16,31 @@ export interface SlackCliOptions {
     noPrefix?: boolean;
     /** Subscribe manifests to channel-post events (opt-in channel watch). */
     channelPosts?: boolean;
+    /** create/destroy: target Slack workspace id (`T…`). */
+    slackTeam?: string;
+    /** create/destroy: target the prod app (default dev). */
+    prod?: boolean;
+    /** create/destroy: Cursor team id; resolved from the account when omitted. */
+    team?: string;
+    /** create/destroy: Cursor API key (defaults to the stored login / env). */
+    apiKey?: string;
+    /** create: app icon — an https image URL or a local png/jpg/gif ≤512KB. */
+    icon?: string;
+    /** create: overwrite manifest edits made in Slack (skip the drift check). */
+    force?: boolean;
+    /** `slack icon <source>` positional (same URL-or-file forms as --icon). */
+    iconSource?: string;
+    /** Test seams. */
+    fetchImpl?: typeof fetch;
+    openBrowserImpl?: (url: string) => void;
+    connectPollIntervalMs?: number;
+    connectPollTimeoutMs?: number;
+    doctorImpl?: typeof runSlackDoctor;
+    out?: (text: string) => void;
+    err?: (text: string) => void;
 }
 export declare function cmdSlack(subcommand: string | undefined, options: SlackCliOptions): Promise<number>;
+export declare function cmdSlackCreate(options: SlackCliOptions): Promise<number>;
+export declare function cmdSlackDestroy(options: SlackCliOptions): Promise<number>;
+export declare function cmdSlackIcon(options: SlackCliOptions): Promise<number>;
 //# sourceMappingURL=cli-slack.d.ts.map

package/dist/internal/cli-slack.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"cli-slack.d.ts","sourceRoot":"","sources":["../../src/internal/cli-slack.ts"],"names":[],"mappings":"AAAA;;GAEG;AAWH,MAAM,WAAW,eAAe;IAC9B,GAAG,EAAE,MAAM,CAAC;IACZ,IAAI,CAAC,EAAE,MAAM,CAAC;IACd,GAAG,CAAC,EAAE,KAAK,GAAG,MAAM,GAAG,MAAM,CAAC;IAC9B,IAAI,CAAC,EAAE,OAAO,CAAC;IACf,KAAK,CAAC,EAAE,OAAO,CAAC;IAChB,oDAAoD;IACpD,MAAM,CAAC,EAAE,MAAM,CAAC;IAChB,2DAA2D;IAC3D,QAAQ,CAAC,EAAE,OAAO,CAAC;IACnB,yEAAyE;IACzE,YAAY,CAAC,EAAE,OAAO,CAAC;CACxB;AAoBD,wBAAsB,QAAQ,CAC5B,UAAU,EAAE,MAAM,GAAG,SAAS,EAC9B,OAAO,EAAE,eAAe,GACvB,OAAO,CAAC,MAAM,CAAC,CAsBjB"}
+{"version":3,"file":"cli-slack.d.ts","sourceRoot":"","sources":["../../src/internal/cli-slack.ts"],"names":[],"mappings":"AAAA;;;;GAIG;AAIH,OAAO,EAML,cAAc,EAGf,MAAM,4BAA4B,CAAC;AAoBpC,MAAM,WAAW,eAAe;IAC9B,GAAG,EAAE,MAAM,CAAC;IACZ,IAAI,CAAC,EAAE,MAAM,CAAC;IACd,GAAG,CAAC,EAAE,KAAK,GAAG,MAAM,GAAG,MAAM,CAAC;IAC9B,IAAI,CAAC,EAAE,OAAO,CAAC;IACf,KAAK,CAAC,EAAE,OAAO,CAAC;IAChB,oDAAoD;IACpD,MAAM,CAAC,EAAE,MAAM,CAAC;IAChB,2DAA2D;IAC3D,QAAQ,CAAC,EAAE,OAAO,CAAC;IACnB,yEAAyE;IACzE,YAAY,CAAC,EAAE,OAAO,CAAC;IACvB,wDAAwD;IACxD,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,yDAAyD;IACzD,IAAI,CAAC,EAAE,OAAO,CAAC;IACf,8EAA8E;IAC9E,IAAI,CAAC,EAAE,MAAM,CAAC;IACd,2EAA2E;IAC3E,MAAM,CAAC,EAAE,MAAM,CAAC;IAChB,2EAA2E;IAC3E,IAAI,CAAC,EAAE,MAAM,CAAC;IACd,6EAA6E;IAC7E,KAAK,CAAC,EAAE,OAAO,CAAC;IAChB,2EAA2E;IAC3E,UAAU,CAAC,EAAE,MAAM,CAAC;IACpB,kBAAkB;IAClB,SAAS,CAAC,EAAE,OAAO,KAAK,CAAC;IACzB,eAAe,CAAC,EAAE,CAAC,GAAG,EAAE,MAAM,KAAK,IAAI,CAAC;IACxC,qBAAqB,CAAC,EAAE,MAAM,CAAC;IAC/B,oBAAoB,CAAC,EAAE,MAAM,CAAC;IAC9B,UAAU,CAAC,EAAE,OAAO,cAAc,CAAC;IACnC,GAAG,CAAC,EAAE,CAAC,IAAI,EAAE,MAAM,KAAK,IAAI,CAAC;IAC7B,GAAG,CAAC,EAAE,CAAC,IAAI,EAAE,MAAM,KAAK,IAAI,CAAC;CAC9B;AA6CD,wBAAsB,QAAQ,CAC5B,UAAU,EAAE,MAAM,GAAG,SAAS,EAC9B,OAAO,EAAE,eAAe,GACvB,OAAO,CAAC,MAAM,CAAC,CA4BjB;AAmbD,wBAAsB,cAAc,CAClC,OAAO,EAAE,eAAe,GACvB,OAAO,CAAC,MAAM,CAAC,CAmOjB;AAiDD,wBAAsB,eAAe,CACnC,OAAO,EAAE,eAAe,GACvB,OAAO,CAAC,MAAM,CAAC,CA4CjB;AAED,wBAAsB,YAAY,CAAC,OAAO,EAAE,eAAe,GAAG,OAAO,CAAC,MAAM,CAAC,CAsD5E"}