@runtypelabs/cli 2.57.0 → 2.58.0

package/README.md

@@ -1110,7 +1110,7 @@ runs, human takeovers of conversations, and skill proposals.
 | `events watch --once` | Exit after the first poll that returns events                    |
 
 `events watch` takes `--name <type>` (default: every type), the filters `--agent`, `--flow`,
-`--product`, `--surface`, `--schedule`, `--suite`, `--task`, and `--only-regressions`, and
+`--product`, `--surface`, `--schedule`, `--suite`, `--task`, `--request`, and `--only-regressions`, and
 `--since <cursor>` (default `now`). Each event carries its `cursor`; pass the last one to `--since`
 to resume without gaps. Events are kept for 7 days. Diagnostics go to stderr, so
 `runtype events watch --json` can feed a monitor that wakes on each line.

package/dist/index.js

@@ -56,7 +56,7 @@ var __exportAll = (all, no_symbols) => {
 	return target;
 };
 var __require = /* #__PURE__ */ (() => createRequire(import.meta.url))();
-// ../shared/dist/chunk-D9RiVw4h.mjs
+// ../shared/dist/chunk-7fm07YIO.mjs
 /**
 * Every product surface type the platform stores and serves. The single source for the FPO surface
 * enum, the REST create and ensure schemas, and the shared `SurfaceType`; add a new type here only.
@@ -65,6 +65,7 @@ const PRODUCT_SURFACE_TYPES = [
 	"chat",
 	"mcp",
 	"mcp_code",
+	"cli",
 	"api",
 	"webhook",
 	"schedule",
@@ -110,7 +111,7 @@ function mergeResults(...results) {
 	merged.valid = merged.errors.length === 0;
 	return merged;
 }
-// ../shared/dist/chunk-XcxbQPuh.mjs
+// ../shared/dist/chunk-BbTDzQd7.mjs
 /** Names every key of `T` exactly once, so a new credential field fails the build. */
 function credentialKeysOf(presence) {
 	return Object.keys(presence);
@@ -122,7 +123,8 @@ const SURFACE_INBOUND_CREDENTIAL_KEYS = credentialKeysOf({
 	secret: true,
 	publicKey: true,
 	authToken: true,
-	twilioAuthToken: true
+	twilioAuthToken: true,
+	identityClientSecret: true
 });
 const SURFACE_OUTBOUND_CREDENTIAL_KEYS = credentialKeysOf({
 	botToken: true,
@@ -135,6 +137,15 @@ const SURFACE_OUTBOUND_CREDENTIAL_KEYS = credentialKeysOf({
 	smtpPassword: true
 });
 [.../* @__PURE__ */ new Set([...SURFACE_INBOUND_CREDENTIAL_KEYS, ...SURFACE_OUTBOUND_CREDENTIAL_KEYS])];
+const OFFERS_OAUTH = {
+	keys_only: false,
+	consent: true,
+	required: true,
+	none: false
+};
+/** Every `authMode` a surface can store. Any other stored value reads as unset. */
+const SURFACE_AUTH_MODES = Object.keys(OFFERS_OAUTH);
+new Set(SURFACE_AUTH_MODES.filter((mode) => OFFERS_OAUTH[mode]));
 // ../shared/dist/chunk-DRY4gGEy.mjs
 /** Every evaluation model the gateway lists, keyed by the gateway id a call names. */
 const GATEWAY_EVALUATION_MODELS = Object.freeze({
@@ -7135,7 +7146,7 @@ function resolveFallback(parsed, context, tracking, options = {}) {
 	return tracking.match;
 }
 new SimpleTemplateEngine();
-// ../shared/dist/chunk-BfSQ06hZ.mjs
+// ../shared/dist/chunk-CYZSyWdd.mjs
 /**
 * Maps model families to provider-specific model IDs.
 * Use for routing: family + provider → exact model ID
@@ -7157,9 +7168,13 @@ const MODEL_FAMILY_PROVIDER_IDS = {
 		"vercel": "anthropic/claude-fable-5.1"
 	},
 	"claude-fable-5-1-batch": { "vercel": "anthropic/claude-fable-5.1-batch" },
+	"claude-fable-5-2": { "vercel": "anthropic/claude-fable-5.2" },
+	"claude-fable-5-2-batch": { "vercel": "anthropic/claude-fable-5.2-batch" },
 	"claude-fable-5-batch": { "vercel": "anthropic/claude-fable-5-batch" },
 	"claude-fable-5.1": { "vercel": "anthropic/claude-fable-5.1" },
 	"claude-fable-5.1-batch": { "vercel": "anthropic/claude-fable-5.1-batch" },
+	"claude-fable-5.2": { "vercel": "anthropic/claude-fable-5.2" },
+	"claude-fable-5.2-batch": { "vercel": "anthropic/claude-fable-5.2-batch" },
 	"claude-haiku-4-5": {
 		"anthropic": "claude-haiku-4-5-20251001",
 		"vercel": "anthropic/claude-haiku-4.5"
@@ -7839,6 +7854,7 @@ const MODEL_FAMILY_PROVIDER_IDS = {
 	"mistral-small": { "vercel": "mistral/mistral-small" },
 	"morph-v3-fast": { "vercel": "morph/morph-v3-fast" },
 	"morph-v3-large": { "vercel": "morph/morph-v3-large" },
+	"muse-ember": { "vercel": "meta/muse-ember" },
 	"muse-glimmer-30b": { "vercel": "meta/muse-glimmer-30b" },
 	"muse-spark-1-1": { "vercel": "meta/muse-spark-1.1" },
 	"muse-spark-1-2": { "vercel": "meta/muse-spark-1.2" },
@@ -8236,7 +8252,7 @@ const CREATOR_TO_PROVIDER = {
 	"unknown": "openai"
 };
 new RegExp(`^${/Model stream stalled \((awaiting_first_token|idle|empty_completion)\) after \d+ms with no progress/.source}`);
-// ../shared/dist/chunk-BtSflVAD.mjs
+// ../shared/dist/chunk-F1-FFFE7.mjs
 /**
 * Unwraps a single {{variableName}} template to its bare variable name, preserving array/object identity instead of stringifying via the template engine.
 * Returns source unchanged when it is not a lone {{...}} wrapper.
@@ -12659,7 +12675,7 @@ object$4({
 	externalId: string$2().min(1).max(500).optional().describe("The conversation id in the source system. Re-creating with the same system and externalId returns the existing conversation with imported: false instead of a duplicate."),
 	importedAt: string$2().datetime({ offset: true }).optional()
 });
-// ../shared/dist/chunk-CoA2IB3c.mjs
+// ../shared/dist/chunk-Dkn7TqhO.mjs
 const voiceInterruptionModeSchema = _enum([
 	"none",
 	"cancel",
@@ -26276,12 +26292,13 @@ function loadBrandConfig(env) {
 }
 loadBrandConfig();
 BRAND_CONFIGS[DEFAULT_BRAND_NAME];
-// ../shared/dist/chunk-2W0oIwXh.mjs
+// ../shared/dist/chunk-DV7xJaYR.mjs
 const ARCHITECTURE_PREVIEW_SURFACE_TYPES = [
 	"chat",
 	"api",
 	"mcp",
 	"mcp_code",
+	"cli",
 	"webhook",
 	"schedule",
 	"a2a",
@@ -27747,7 +27764,14 @@ object$4({
 	simulationsAvailable: boolean().optional(),
 	enableDelegateToRuntype: boolean().optional(),
 	enableDelegateToOtto: boolean().optional(),
-	signalsConfigured: boolean().optional()
+	signalsConfigured: boolean().optional(),
+	enableCliSurface: boolean().optional(),
+	enableCliSurfacePublish: boolean().optional(),
+	enableCliSurfaceGraduation: boolean().optional(),
+	enableMcpSurfaceSignIn: boolean().optional(),
+	enableMcpSurfaceDeviceGrant: boolean().optional(),
+	enableMcpSurfaceOobApprovals: boolean().optional(),
+	enableMcpSurfaceCustomerIdp: boolean().optional()
 });
 const SLACK_REQUIRED_BOT_SCOPES = [
 	"channels:history",
@@ -28079,7 +28103,8 @@ const PRODUCT_EVENT_NAMES = [
 	"runtype.task.failed",
 	"runtype.task.cancelled",
 	"runtype.signal.fired",
-	"runtype.signal.cluster_appeared"
+	"runtype.signal.cluster_appeared",
+	"runtype.api_key_request.decided"
 ];
 const PRODUCT_EVENT_DEFAULT_NEXT_POLL_MS = 5e3;
 const PRODUCT_EVENT_MAX_WAIT_MS = 5e4;
@@ -28099,6 +28124,7 @@ const ProductEventFilterFieldSchemas = {
 	suiteId: resourceId("Only occurrences for this eval suite"),
 	simulationId: resourceId("Only occurrences for this simulation"),
 	taskId: resourceId("Only occurrences for this delegated task"),
+	requestId: resourceId("Only occurrences for this API key request"),
 	key: string$2().min(1).max(64).describe("Only signal occurrences for this signal key, such as tool_error"),
 	onlyRegressions: boolean().describe("Only eval runs whose pass rate dropped below the previous run"),
 	minIntervalMs: number$2().int().min(0).max(PRODUCT_EVENT_MAX_MIN_INTERVAL_MS).describe("Coalesce repeats: after one execution failure per agent or flow and error class, or one fired signal per agent, key and value, drop later ones within this window")
@@ -28142,6 +28168,7 @@ const SignalClusterAppearedArgumentsSchema = object$4({
 	agentId: filter.agentId.optional(),
 	key: filter.key.optional()
 }).strict();
+const ApiKeyRequestDecidedArgumentsSchema = object$4({ requestId: filter.requestId.optional() }).strict();
 object$4({
 	agentId: filter.agentId.optional(),
 	flowId: filter.flowId.optional(),
@@ -28151,6 +28178,7 @@ object$4({
 	suiteId: filter.suiteId.optional(),
 	simulationId: filter.simulationId.optional(),
 	taskId: filter.taskId.optional(),
+	requestId: filter.requestId.optional(),
 	key: filter.key.optional(),
 	onlyRegressions: filter.onlyRegressions.optional(),
 	minIntervalMs: filter.minIntervalMs.optional()
@@ -28270,6 +28298,17 @@ const SignalClusterAppearedPayloadSchema = object$4({
 	memberCount: number$2().int().min(0).describe("Answers in the cluster"),
 	firstSeenAt: isoTimestamp("When the earliest member was seen")
 });
+const ApiKeyRequestDecidedPayloadSchema = object$4({
+	requestId: resourceId("The API key request"),
+	status: _enum([
+		"approved",
+		"declined",
+		"expired",
+		"cancelled"
+	]).describe("Where the request went from pending; on approved, redeem it with claim_api_key"),
+	decidedAt: isoTimestamp("When the request left pending"),
+	claimExpiresAt: isoTimestamp("Approved only: redeem before this time or the approval expires").optional()
+});
 const PRODUCT_EVENT_ARGUMENT_SCHEMAS = {
 	"runtype.execution.failed": ExecutionFailedArgumentsSchema,
 	"runtype.approval.requested": ApprovalRequestedArgumentsSchema,
@@ -28283,7 +28322,8 @@ const PRODUCT_EVENT_ARGUMENT_SCHEMAS = {
 	"runtype.task.failed": DelegateTaskArgumentsSchema,
 	"runtype.task.cancelled": DelegateTaskArgumentsSchema,
 	"runtype.signal.fired": SignalFiredArgumentsSchema,
-	"runtype.signal.cluster_appeared": SignalClusterAppearedArgumentsSchema
+	"runtype.signal.cluster_appeared": SignalClusterAppearedArgumentsSchema,
+	"runtype.api_key_request.decided": ApiKeyRequestDecidedArgumentsSchema
 };
 const PRODUCT_EVENT_PAYLOAD_SCHEMAS = {
 	"runtype.execution.failed": ExecutionFailedPayloadSchema,
@@ -28298,7 +28338,8 @@ const PRODUCT_EVENT_PAYLOAD_SCHEMAS = {
 	"runtype.task.failed": TaskFailedPayloadSchema,
 	"runtype.task.cancelled": TaskCancelledPayloadSchema,
 	"runtype.signal.fired": SignalFiredPayloadSchema,
-	"runtype.signal.cluster_appeared": SignalClusterAppearedPayloadSchema
+	"runtype.signal.cluster_appeared": SignalClusterAppearedPayloadSchema,
+	"runtype.api_key_request.decided": ApiKeyRequestDecidedPayloadSchema
 };
 const PRODUCT_EVENT_DESCRIPTIONS = {
 	"runtype.execution.failed": "An agent or flow execution ended in failure. An API key sees agent runs with AGENTS:READ and flow runs with FLOWS:READ.",
@@ -28313,7 +28354,8 @@ const PRODUCT_EVENT_DESCRIPTIONS = {
 	"runtype.task.failed": "A delegated task ended in failure.",
 	"runtype.task.cancelled": "A delegated task was cancelled before it finished.",
 	"runtype.signal.fired": "A trace signal fired on an agent turn, including the built-in failure facts every agent gets.",
-	"runtype.signal.cluster_appeared": "A new cluster of similar signal answers appeared for an agent. Declared ahead of clustering; nothing emits it yet."
+	"runtype.signal.cluster_appeared": "A new cluster of similar signal answers appeared for an agent. Declared ahead of clustering; nothing emits it yet.",
+	"runtype.api_key_request.decided": "An API key request left pending: approved, declined or cancelled, or expired, which is recorded only when the request is read after its expiresAt. On approved, call claim_api_key before claimExpiresAt. An API key without API_KEYS:READ sees only the requests it filed."
 };
 /** API-key scopes that may read each event; holding any one of them is enough. */
 const PRODUCT_EVENT_REQUIRED_SCOPES = {
@@ -28333,7 +28375,8 @@ const PRODUCT_EVENT_REQUIRED_SCOPES = {
 	"runtype.task.failed": ["DELEGATE_TASKS:READ", "DELEGATE_TASKS:WRITE"],
 	"runtype.task.cancelled": ["DELEGATE_TASKS:READ", "DELEGATE_TASKS:WRITE"],
 	"runtype.signal.fired": ["SIGNALS:READ", "SIGNALS:WRITE"],
-	"runtype.signal.cluster_appeared": ["SIGNALS:READ", "SIGNALS:WRITE"]
+	"runtype.signal.cluster_appeared": ["SIGNALS:READ", "SIGNALS:WRITE"],
+	"runtype.api_key_request.decided": ["API_KEYS:READ"]
 };
 /** Derive a draft-07 JSON Schema object from a Zod object, without the `$schema` marker. */
 function productEventJsonSchema(schema) {
@@ -54137,6 +54180,20 @@ modelsCommand.command("usage").description("Show model usage statistics").option
 		process.exit(1);
 	}
 });
+/** Reads a schedule's whole run history, newest first, one page at a time. */
+async function fetchAllScheduleRuns(client, scheduleId) {
+	const runs = [];
+	let cursor;
+	do {
+		const page = await client.get(`/schedules/${scheduleId}/runs`, {
+			limit: 500,
+			...cursor ? { cursor } : {}
+		});
+		runs.push(...page.data ?? []);
+		cursor = page.hasMore && page.nextCursor ? page.nextCursor : void 0;
+	} while (cursor);
+	return runs;
+}
 // src/commands/schedules.ts
 const schedulesCommand = new Command("schedules").description("Manage schedules");
 schedulesCommand.command("list").description("List all schedules").option("--limit <n>", "Limit results (defaults to the API page size)").option("--cursor <cursor>", "Pagination cursor for the next page").option("--json", "Output as JSON").option("--tty", "Force TTY mode").option("--no-tty", "Force non-TTY mode").action(async (options) => {
@@ -54334,12 +54391,15 @@ schedulesCommand.command("runs <id>").description("List run history for a schedu
 	const client = createCliClient(apiKey);
 	if (!isTTY(options) || options.json) {
 		try {
-			const data = await client.get(`/schedules/${id}/runs`);
+			const runs = await fetchAllScheduleRuns(client, id);
 			if (options.json) {
-				printJson(data);
+				printJson({
+					data: runs,
+					hasMore: false,
+					nextCursor: null
+				});
 				return;
 			}
-			const runs = data.data ?? [];
 			if (runs.length === 0) {
 				console.log("No runs found");
 				return;
@@ -54366,7 +54426,7 @@ schedulesCommand.command("runs <id>").description("List run history for a schedu
 		const [items, setItems] = useState(null);
 		const [error, setError] = useState(null);
 		useEffect(() => {
-			client.get(`/schedules/${id}/runs`).then((res) => setItems(res.data ?? [])).catch((err) => setError(err instanceof Error ? err : new Error(String(err))));
+			fetchAllScheduleRuns(client, id).then(setItems).catch((err) => setError(err instanceof Error ? err : new Error(String(err))));
 		}, []);
 		return React.createElement(DataList, {
 			title: "Schedule Runs",
@@ -60022,11 +60082,13 @@ function parseEventName(value) {
 	if (value === void 0 || isProductEventName(value)) return value;
 	throw new Error(`Unknown event type "${value}". Known types: ${PRODUCT_EVENT_NAMES.join(", ")}`);
 }
+/** `--request` filters only this event, and a key with no event scope may read it only by name. */
+const API_KEY_REQUEST_DECIDED_EVENT = "runtype.api_key_request.decided";
 /** Query string for one GET /events call; `after` omitted asks the API for the head cursor. */
 function buildEventsQuery(options, after) {
 	const params = new URLSearchParams();
 	const entries = [
-		["name", options.name],
+		["name", options.name ?? (options.request ? API_KEY_REQUEST_DECIDED_EVENT : void 0)],
 		["agentId", options.agent],
 		["flowId", options.flow],
 		["productId", options.product],
@@ -60035,6 +60097,7 @@ function buildEventsQuery(options, after) {
 		["suiteId", options.suite],
 		["simulationId", options.simulation],
 		["taskId", options.task],
+		["requestId", options.request],
 		["key", options.key],
 		["onlyRegressions", options.onlyRegressions ? "true" : void 0],
 		["after", after],
@@ -60202,7 +60265,7 @@ const eventsCommand = new Command("events").description("Watch product events (f
 eventsCommand.command("types").description("List the product event types you can watch").option("--json", "Output as JSON").option("--no-color", "Disable color output").action(async (options) => {
 	await runTypes(options);
 });
-eventsCommand.command("watch").description("Long-poll product events and print one line per event").option("--name <name>", "Event type to watch (default: every type; see `runtype events types`)").option("--agent <id>", "Filter by agent ID").option("--flow <id>", "Filter by flow ID").option("--product <id>", "Filter by product ID").option("--surface <id>", "Filter by surface ID").option("--schedule <id>", "Filter by schedule ID").option("--suite <id>", "Filter by eval suite ID").option("--simulation <id>", "Filter by simulation ID").option("--task <id>", "Filter by delegated task ID").option("--key <key>", "Signal events: only this signal key, such as tool_error").option("--only-regressions", "Eval runs: only those whose pass rate dropped").option("--since <cursor>", "Start after this cursor, or \"now\" (default)").option("--json", "Output NDJSON: one event object per line on stdout").option("--once", "Exit after the first poll that returns events").option("--no-color", "Disable color output").action(async (options) => {
+eventsCommand.command("watch").description("Long-poll product events and print one line per event").option("--name <name>", "Event type to watch (default: every type; see `runtype events types`)").option("--agent <id>", "Filter by agent ID").option("--flow <id>", "Filter by flow ID").option("--product <id>", "Filter by product ID").option("--surface <id>", "Filter by surface ID").option("--schedule <id>", "Filter by schedule ID").option("--suite <id>", "Filter by eval suite ID").option("--simulation <id>", "Filter by simulation ID").option("--task <id>", "Filter by delegated task ID").option("--request <id>", "Filter by API key request ID").option("--key <key>", "Signal events: only this signal key, such as tool_error").option("--only-regressions", "Eval runs: only those whose pass rate dropped").option("--since <cursor>", "Start after this cursor, or \"now\" (default)").option("--json", "Output NDJSON: one event object per line on stdout").option("--once", "Exit after the first poll that returns events").option("--no-color", "Disable color output").action(async (options) => {
 	await runWatch(options);
 });
 // src/commands/validate-product.ts
@@ -62752,7 +62815,7 @@ function h$2(e, t) {
 		}
 	}
 }
-// ../mcp-cli/dist/chunk-CL8BNzsR.mjs
+// ../mcp-cli/dist/chunk-Ce22LqPV.mjs
 const i = /* @__PURE__ */ new Set([
 	`approvals`,
 	`auth`,
@@ -62799,7 +62862,10 @@ function l(e, t = {}) {
 	for (let i of e) {
 		if (a.has(i.name)) continue;
 		let e = Object.prototype.hasOwnProperty.call(t, i.name) ? t[i.name] : void 0, c = s(i, e), l = {
-			tool: i,
+			tool: e?.description ? {
+				...i,
+				description: e.description
+			} : i,
 			path: c,
 			aliases: [],
 			hidden: e?.hidden === !0,
@@ -63090,16 +63156,16 @@ function te(e) {
 		required: Array.isArray(t.required) ? t.required.filter((e) => typeof e == `string` && Object.hasOwn(i, e)) : []
 	};
 }
-function C$1(e, t) {
+function ne(e, t) {
 	let n = (Array.isArray(t.anyOf) ? t.anyOf : Array.isArray(t.oneOf) ? t.oneOf : []).filter((e) => u(e) && !v$1(e));
 	return n.length === 1 ? g$1(e, n[0]).schema : t;
 }
-function w$1(e, t) {
+function re(e, t) {
 	let n = e;
 	for (let r of t) {
 		let { schema: t, recursive: i } = g$1(e, n);
 		if (i) return;
-		let a = C$1(e, t);
+		let a = ne(e, t);
 		if (r === `[]`) {
 			if (!u(a.items)) return;
 			n = a.items;
@@ -63109,11 +63175,11 @@ function w$1(e, t) {
 	}
 	return S$1(e, n);
 }
-const T$1 = /* @__PURE__ */ new Set(/* @__PURE__ */ `answers.approvals.approve.approve-all.color.deny.detach.dry-run.example.field.help.help-all.input.json.json-body.jsonl.max-output.no-color.no-input.output.profile.raw-field.raw-strings.save.stream.timeout.token.verbose.wait.yes`.split(`.`));
-function E$1(e) {
+const C$1 = /* @__PURE__ */ new Set(/* @__PURE__ */ `answers.approvals.approve.approve-all.color.deny.detach.dry-run.example.field.help.help-all.input.json.json-body.jsonl.max-output.no-color.no-input.output.profile.raw-field.raw-strings.save.stream.timeout.token.verbose.wait.yes`.split(`.`));
+function w$1(e) {
 	return e.kind === `string` || e.kind === `enum` || e.kind === `binary`;
 }
-function D$1(e, t) {
+function T$1(e, t) {
 	let r = e.inputSchema ?? {}, i = te(r);
 	if (!i) return {
 		tool: e,
@@ -63132,21 +63198,23 @@ function D$1(e, t) {
 		let u = S$1(r, t);
 		if (u.readOnly) continue;
 		if (u.kind === `const`) {
-			O$1(a, e, u.constValue);
+			E$1(a, e, u.constValue);
 			continue;
 		}
 		let d = e$1(e) || e, f = d;
-		if (T$1.has(f) || (l.get(d) ?? 0) > 1 || c.has(f)) for (f = `arg-${d}`, (c.has(f) || (l.get(d) ?? 0) > 1) && (f = `arg-${e}`); c.has(f);) f = `arg-${f}`;
-		c.add(f), s.push({
+		if (C$1.has(f) || (l.get(d) ?? 0) > 1 || c.has(f)) for (f = `arg-${d}`, (c.has(f) || (l.get(d) ?? 0) > 1) && (f = `arg-${e}`); c.has(f);) f = `arg-${f}`;
+		c.add(f);
+		let p = i.required.includes(e);
+		p && u.hasDefault && !Object.hasOwn(o, e) && E$1(o, e, u.defaultValue), s.push({
 			wire: e,
 			flag: f,
 			field: u,
-			required: i.required.includes(e) && !u.hasDefault && !Object.hasOwn(o, e),
+			required: p && !Object.hasOwn(o, e),
 			hidden: u.deprecated
 		});
 	}
 	let u = new Map(s.map((e) => [e.wire, e])), d = [];
-	if (t?.positional && t.positional.length > 0) d = t.positional.map((e) => u.get(e)).filter((e) => e !== void 0 && E$1(e.field));
+	if (t?.positional && t.positional.length > 0) d = t.positional.map((e) => u.get(e)).filter((e) => e !== void 0 && w$1(e.field));
 	else {
 		let e = s.filter((e) => e.required);
 		e.length === 1 && e[0].field.kind === `string` && (d = e);
@@ -63165,7 +63233,7 @@ function D$1(e, t) {
 		required: s.filter((e) => e.required).map((e) => e.wire)
 	};
 }
-function O$1(e, t, n) {
+function E$1(e, t, n) {
 	Object.defineProperty(e, t, {
 		value: n,
 		enumerable: !0,
@@ -63173,14 +63241,14 @@ function O$1(e, t, n) {
 		writable: !0
 	});
 }
-function k$1(e) {
+function D$1(e) {
 	return e?.startsWith(`\0`) === !0;
 }
-function A$1(t, n) {
+function O$1(t, n) {
 	let r = n.slice(1), i = t.startsWith(`--`) ? `${t}=${r}` : `${t}${r}`;
 	return new n$2(`usage`, `${t} needs a value, but the word after it, ${r}, is a global flag`, `Write ${i} to pass it as the value, or put ${r} after the value`);
 }
-const j$1 = /* @__PURE__ */ new Set([
+const k$1 = /* @__PURE__ */ new Set([
 	`auto`,
 	`json`,
 	`text`,
@@ -63188,7 +63256,7 @@ const j$1 = /* @__PURE__ */ new Set([
 	`ndjson`,
 	`table`
 ]);
-const M$1 = /* @__PURE__ */ new Set([
+const A$1 = /* @__PURE__ */ new Set([
 	`output`,
 	`timeout`,
 	`save`,
@@ -63204,34 +63272,34 @@ const M$1 = /* @__PURE__ */ new Set([
 	`wait`,
 	`stream`
 ]);
-const N$1 = /* @__PURE__ */ new Set([
+const j$1 = /* @__PURE__ */ new Set([
 	`auto`,
 	`raw`,
 	`off`
 ]);
-const re = /* @__PURE__ */ new Set([
+const M$1 = /* @__PURE__ */ new Set([
 	`prompt`,
 	`fail`,
 	`deny`
 ]);
-const P$1 = {
+const N$1 = {
 	h: `help`,
 	o: `output`,
 	y: `yes`,
 	f: `raw-field`,
 	F: `field`
 };
-function F(e) {
+function P$1(e) {
 	return e.field.kind === `scalar-array` || e.field.kind === `object-array` || e.field.kind === `map`;
 }
-function I(t, n) {
+function F(t, n) {
 	let r = /^(\d+(?:\.\d+)?)(ms|s|m|h)?$/.exec(t.trim());
 	if (!r) throw new n$2(`invalid_value`, `${n} expects a duration such as 30000, 30s or 5m`);
 	let i = Number(r[1]), a = r[2] ?? `ms`, o = Math.round(i * (a === `h` ? 36e5 : a === `m` ? 6e4 : a === `s` ? 1e3 : 1));
 	if (o <= 0) throw new n$2(`invalid_value`, `${n} must be greater than zero`);
 	return o;
 }
-function ie(t) {
+function I(t) {
 	if (t === `0` || t === `none` || t === `unlimited`) return null;
 	if (!/^\d+$/.test(t)) throw new n$2(`invalid_value`, `--max-output expects a character count, or 0 for unlimited`);
 	return Number(t);
@@ -63260,11 +63328,11 @@ function ae(t, n) {
 			if (s.has(t)) throw new n$2(`usage`, `--${t} was given more than once`);
 			switch (s.add(t), t) {
 				case `output`:
-					if (!j$1.has(n)) throw new n$2(`invalid_value`, `--output must be one of ${[...j$1].join(`, `)}`);
+					if (!k$1.has(n)) throw new n$2(`invalid_value`, `--output must be one of ${[...k$1].join(`, `)}`);
 					r.output = L(r.output, n);
 					return;
 				case `timeout`:
-					r.timeoutMs = I(n, `--timeout`);
+					r.timeoutMs = F(n, `--timeout`);
 					return;
 				case `save`:
 					r.save = n;
@@ -63276,10 +63344,10 @@ function ae(t, n) {
 					r.jsonBody = n;
 					return;
 				case `max-output`:
-					r.maxOutput = ie(n);
+					r.maxOutput = I(n);
 					return;
 				case `approvals`:
-					if (!re.has(n)) throw new n$2(`invalid_value`, `--approvals must be prompt, fail or deny`);
+					if (!M$1.has(n)) throw new n$2(`invalid_value`, `--approvals must be prompt, fail or deny`);
 					r.approvals = n;
 					return;
 				case `answers`:
@@ -63289,7 +63357,7 @@ function ae(t, n) {
 					r.wait = n;
 					return;
 				case `stream`:
-					if (!N$1.has(n)) throw new n$2(`invalid_value`, `--stream must be auto, raw or off`);
+					if (!j$1.has(n)) throw new n$2(`invalid_value`, `--stream must be auto, raw or off`);
 					r.stream = n;
 					return;
 			}
@@ -63310,7 +63378,7 @@ function ae(t, n) {
 		}
 	}, u = (t, n) => {
 		let r = i.get(t) ?? [];
-		if (r.length > 0 && !F(t)) throw new n$2(`usage`, `--${t.flag} was given more than once`, oe(t));
+		if (r.length > 0 && !P$1(t)) throw new n$2(`usage`, `--${t.flag} was given more than once`, oe(t));
 		r.push(n), i.set(t, r);
 	}, d = (t, n) => {
 		if (a.has(t)) throw new n$2(`usage`, `--${t.flag} was given more than once`);
@@ -63318,7 +63386,7 @@ function ae(t, n) {
 	};
 	for (let r = 0; r < t.length; r += 1) {
 		let i = t[r];
-		if (!k$1(i)) {
+		if (!D$1(i)) {
 			if (i === `--`) {
 				o.push(...t.slice(r + 1));
 				break;
@@ -63326,7 +63394,7 @@ function ae(t, n) {
 			if (i.startsWith(`--`) && i.length > 2) {
 				let a = i.indexOf(`=`), o = a === -1 ? i.slice(2) : i.slice(2, a), s = a === -1 ? void 0 : i.slice(a + 1), f = () => {
 					if (s !== void 0) return s;
-					if (k$1(t[r + 1])) throw A$1(i, t[r + 1]);
+					if (D$1(t[r + 1])) throw O$1(i, t[r + 1]);
 					if (r + 1 >= t.length) throw new n$2(`missing_argument`, `--${o} needs a value`);
 					return r += 1, t[r];
 				}, p = n.byFlag.get(o);
@@ -63342,7 +63410,7 @@ function ae(t, n) {
 						continue;
 					}
 				}
-				if (M$1.has(o)) {
+				if (A$1.has(o)) {
 					c(o, f());
 					continue;
 				}
@@ -63350,13 +63418,13 @@ function ae(t, n) {
 				throw R(o, n);
 			}
 			if (i.startsWith(`-`) && i.length > 1 && !/^-\d/.test(i)) {
-				let a = i.slice(1, 2), o = P$1[a];
+				let a = i.slice(1, 2), o = N$1[a];
 				if (!o) throw R(i.slice(1), n);
 				let s = i.slice(2);
-				if (M$1.has(o)) {
+				if (A$1.has(o)) {
 					let n = s;
 					if (n === ``) {
-						if (k$1(t[r + 1])) throw A$1(i, t[r + 1]);
+						if (D$1(t[r + 1])) throw O$1(i, t[r + 1]);
 						if (r + 1 >= t.length) throw new n$2(`missing_argument`, `-${a} needs a value`);
 						r += 1, n = t[r];
 					}
@@ -63389,7 +63457,7 @@ function se(t, n) {
 	throw new n$2(`invalid_value`, `${n} expects true or false, got ${JSON.stringify(t)}`);
 }
 function R(n, r) {
-	let a = n$1(n, [...r.flags.filter((e) => !e.hidden).map((e) => e.flag), ...T$1]);
+	let a = n$1(n, [...r.flags.filter((e) => !e.hidden).map((e) => e.flag), ...C$1]);
 	return new n$2(`unknown_flag`, `Unknown flag --${n} for ${r.tool.name}`, a.length > 0 ? `Did you mean ${a.map((e) => `--${e}`).join(`, `)}?` : `Run with --help to list flags`);
 }
 function z(e) {
@@ -63433,7 +63501,7 @@ function U(e, t, n = 0) {
 	return a.length > 0 ? `{ ${a.join(`; `)} }` : `object`;
 }
 function W(e) {
-	let t = e.inputSchema ?? {}, n = D$1(e);
+	let t = e.inputSchema ?? {}, n = T$1(e);
 	if (n.bodyOnly) return r(`${e.name}(body: unknown)`);
 	let i = n.flags.filter((e) => !e.hidden).map((e) => `${e.wire}${e.required ? `` : `?`}: ${V(e.field, t)}${e.field.nullable ? ` | null` : ``}`);
 	return r(`${e.name}(${i.length > 0 ? `{ ${i.join(`; `)} }` : ``})`);
@@ -63456,7 +63524,7 @@ function de(e) {
 	return r(`${e.map((e) => `  ${e.command.padEnd(t)}  ${e.summary}${e.badges.length > 0 ? ` [${e.badges.join(`, `)}]` : ``}`.trimEnd()).join(`
 `)}\n`);
 }
-// ../mcp-cli/dist/chunk-DqN1GrzM.mjs
+// ../mcp-cli/dist/chunk-BzzREe8T.mjs
 var p = class extends n$2 {};
 function m(e) {
 	return typeof e == `object` && !!e && !Array.isArray(e);
@@ -63472,17 +63540,17 @@ function g(e) {
 	if (Array.isArray(e)) return e.map(g);
 	if (m(e)) {
 		let t = {};
-		for (let [r, i] of Object.entries(e)) O$1(t, r, g(i));
+		for (let [r, i] of Object.entries(e)) E$1(t, r, g(i));
 		return t;
 	}
 	return e;
 }
 function _(e, t) {
 	let r = {};
-	for (let [t, i] of Object.entries(e)) O$1(r, t, i);
+	for (let [t, i] of Object.entries(e)) E$1(r, t, i);
 	for (let [e, i] of Object.entries(t)) {
 		let t = Object.prototype.hasOwnProperty.call(r, e) ? r[e] : void 0;
-		O$1(r, e, m(t) && m(i) ? _(t, i) : i);
+		E$1(r, e, m(t) && m(i) ? _(t, i) : i);
 	}
 	return r;
 }
@@ -63594,7 +63662,7 @@ async function w(t, r, i) {
 			if (r <= 0) throw new p(`invalid_value`, `${a} expects key=value`);
 			let c = s.slice(0, r);
 			if (Object.prototype.hasOwnProperty.call(t, c)) throw new n$2(`usage`, `${a} sets ${c} more than once`);
-			O$1(t, c, await C(o.item ?? o, s.slice(r + 1), `${a} ${c}`, i));
+			E$1(t, c, await C(o.item ?? o, s.slice(r + 1), `${a} ${c}`, i));
 		}
 		return t;
 	}
@@ -63624,7 +63692,7 @@ function T(t, n) {
 async function E(e, n, r, i) {
 	let a = `${e.typed ? `-F` : `-f`} ${e.path.join(`.`)}`;
 	if (!e.typed || r) return e.raw;
-	let s = w$1(n, e.append ? [...e.path, `[]`] : e.path), c = e.raw;
+	let s = re(n, e.append ? [...e.path, `[]`] : e.path), c = e.raw;
 	if (c.startsWith(`@@`)) c = c.slice(1);
 	else if (c.startsWith(`@`)) {
 		let e = await v(c, i, a);
@@ -63651,19 +63719,19 @@ function O(t, r, i) {
 		let i = Object.prototype.hasOwnProperty.call(a, t) ? a[t] : void 0;
 		if (i === void 0) {
 			let e = {};
-			O$1(a, t, e), a = e;
+			E$1(a, t, e), a = e;
 		} else if (m(i)) a = i;
 		else throw new n$2(`usage`, `-F ${r.path.join(`.`)} conflicts with a non-object value`);
 	}
 	let o = r.path[r.path.length - 1];
 	if (r.append) {
 		let t = Object.prototype.hasOwnProperty.call(a, o) ? a[o] : void 0, s = r.empty ? [] : [i];
-		if (t === void 0) O$1(a, o, s);
+		if (t === void 0) E$1(a, o, s);
 		else if (Array.isArray(t)) t.push(...s);
 		else throw new n$2(`usage`, `-F ${o}[] conflicts with a non-array value`);
 	} else {
 		if (Object.prototype.hasOwnProperty.call(a, o)) throw new n$2(`usage`, `${r.path.join(`.`)} is set more than once by -f/-F`);
-		O$1(a, o, i);
+		E$1(a, o, i);
 	}
 }
 async function k(e, t) {
@@ -63745,12 +63813,12 @@ async function P(t, r, i) {
 	for (let [a, o] of r.positionals.entries()) {
 		let s = t.positionals[a], c = r.values.get(s);
 		if (c && c.some((e) => e !== o)) throw new n$2(`usage`, `${s.wire} was given both positionally and as --${s.flag}, with different values`);
-		O$1(d, s.wire, await C(s.field, o, s.wire, i));
+		E$1(d, s.wire, await C(s.field, o, s.wire, i));
 	}
-	for (let [e, t] of r.values) O$1(d, e.wire, await w(e, t, i));
-	for (let [e, t] of r.booleans) O$1(d, e.wire, t);
+	for (let [e, t] of r.values) E$1(d, e.wire, await w(e, t, i));
+	for (let [e, t] of r.booleans) E$1(d, e.wire, t);
 	u = _(u, d);
-	for (let [e, r] of [...Object.entries(t.constFields), ...Object.entries(t.defaults)]) Object.prototype.hasOwnProperty.call(u, e) || O$1(u, e, structuredClone(r));
+	for (let [e, r] of [...Object.entries(t.constFields), ...Object.entries(t.defaults)]) Object.prototype.hasOwnProperty.call(u, e) || E$1(u, e, structuredClone(r));
 	let f = t.required.filter((e) => !Object.prototype.hasOwnProperty.call(u, e));
 	if (f.length > 0) {
 		let e = f.map((e) => {
@@ -63779,7 +63847,7 @@ function toMcpCliArgv(options) {
 }
 /** Whether resolving these arguments will consume the process stdin stream; the parser that reads it decides. */
 function usesMcpStdin(options) {
-	const spec = D$1(SCHEMALESS);
+	const spec = T$1(SCHEMALESS);
 	try {
 		return N(ae(toMcpCliArgv(options), spec), spec);
 	} catch {
@@ -63809,8 +63877,8 @@ async function resolveMcpArguments(options, dependencies = {}) {
 		},
 		...dependencies
 	};
-	const toolSpec = D$1(deps.tool ?? SCHEMALESS);
-	const spec = toolSpec.bodyOnly ? D$1(SCHEMALESS) : {
+	const toolSpec = T$1(deps.tool ?? SCHEMALESS);
+	const spec = toolSpec.bodyOnly ? T$1(SCHEMALESS) : {
 		...toolSpec,
 		required: [],
 		positionals: []
@@ -70302,6 +70370,43 @@ function createHarnessCaptureFactory(options) {
 	});
 	return factory;
 }
+// ../shared/dist/simulation/reported-usage.mjs
+const amount = number$2().finite().nonnegative();
+const tokens = amount.int();
+/** Producer estimates for one turn, independent of Runtype's measured usage and billing. */
+const reportedUsageSchema = object$4({
+	costUsd: amount.nullable(),
+	completeness: _enum(["complete", "partial"]),
+	scope: string$2().min(1).max(120),
+	models: array(object$4({
+		model: string$2().min(1).max(200),
+		canonicalModel: string$2().min(1).max(200).optional(),
+		provider: string$2().max(100).optional(),
+		pricingBasis: _enum([
+			"list",
+			"managed",
+			"unknown"
+		]).optional(),
+		costUsd: amount.nullable().optional(),
+		inputTokens: tokens.optional(),
+		outputTokens: tokens.optional(),
+		cacheReadInputTokens: tokens.optional(),
+		cacheCreationInputTokens: tokens.optional()
+	})).max(100).optional()
+});
+/** Known estimates and reporting coverage; a partial sum is never a whole-episode cost. */
+const selfReportedCostSchema = object$4({
+	knownCostUsd: amount.nullable(),
+	reportedTurns: tokens,
+	totalTurns: tokens,
+	completeTurns: tokens,
+	unknownEpisodes: tokens.optional(),
+	completeness: _enum([
+		"complete",
+		"partial",
+		"unknown"
+	])
+});
 const simTerminationSchema = _enum([
 	"goal_met",
 	"gave_up",
@@ -70432,43 +70537,7 @@ const productKindSchema = _enum([
 	"external",
 	"spec_only"
 ]);
-// ../shared/dist/chunk-CLmUyAjf.mjs
-const amount = number$2().finite().nonnegative();
-const tokens = amount.int();
-/** Producer estimates for one turn, independent of Runtype's measured usage and billing. */
-const reportedUsageSchema = object$4({
-	costUsd: amount.nullable(),
-	completeness: _enum(["complete", "partial"]),
-	scope: string$2().min(1).max(120),
-	models: array(object$4({
-		model: string$2().min(1).max(200),
-		canonicalModel: string$2().min(1).max(200).optional(),
-		provider: string$2().max(100).optional(),
-		pricingBasis: _enum([
-			"list",
-			"managed",
-			"unknown"
-		]).optional(),
-		costUsd: amount.nullable().optional(),
-		inputTokens: tokens.optional(),
-		outputTokens: tokens.optional(),
-		cacheReadInputTokens: tokens.optional(),
-		cacheCreationInputTokens: tokens.optional()
-	})).max(100).optional()
-});
-/** Known estimates and reporting coverage; a partial sum is never a whole-episode cost. */
-const selfReportedCostSchema = object$4({
-	knownCostUsd: amount.nullable(),
-	reportedTurns: tokens,
-	totalTurns: tokens,
-	completeTurns: tokens,
-	unknownEpisodes: tokens.optional(),
-	completeness: _enum([
-		"complete",
-		"partial",
-		"unknown"
-	])
-});
+// ../shared/dist/chunk-BobaAxaK.mjs
 /** The manifest file at the root of a candidate's source bundle. */
 const CANDIDATE_MANIFEST_FILENAME = "runtype-candidate.json";
 const candidateRuntimeSchema = _enum([
@@ -71044,7 +71113,8 @@ const targetRecordingSchema = object$4({
 	enabled: boolean(),
 	retentionDays: number$2().int().min(1).max(365)
 });
-object$4({
+/** One target's recording policy as admission froze it into the run plan. */
+const frozenTargetRecordingSchema = object$4({
 	mode: _enum([
 		"off",
 		"shape-only",
@@ -73164,10 +73234,18 @@ const draftSimulationToolsSchema = array(object$4({
 	/** The tool's input JSON Schema as the target listed it; the drafting agent reads its parameters. */
 	inputSchema: record$2(string$2(), unknown()).refine((schema) => jsonChars(schema) <= MAX_DRAFT_TOOL_INPUT_SCHEMA_CHARS, { message: `An inputSchema may take at most ${MAX_DRAFT_TOOL_INPUT_SCHEMA_CHARS} characters as JSON.` }).optional()
 })).min(1).max(1e3).refine((tools) => tools.reduce((sum, tool) => sum + (tool.inputSchema ? jsonChars(tool.inputSchema) : 0), 0) <= MAX_DRAFT_TOOL_INPUT_SCHEMAS_TOTAL_CHARS, { message: `The inputSchemas of one request may take at most ${MAX_DRAFT_TOOL_INPUT_SCHEMAS_TOTAL_CHARS} characters together.` });
+/** Harder tasks than a simulation already runs: each chains two or more tool calls, around the tools agents used least. */
+const draftHarderTasksSchema = object$4({
+	/** Tools agents called least in a report; when any is listed, every task uses at least one of them. */
+	focusTools: array(string$2().min(1).max(128)).max(12),
+	/** The titles of the tasks the simulation already has; a task repeating one is turned back. */
+	existingTitles: array(string$2().min(1).max(255)).max(200)
+});
 object$4({
 	targetName: string$2().min(1).max(255),
 	tools: draftSimulationToolsSchema,
-	count: number$2().int().min(1).max(12).optional()
+	count: number$2().int().min(1).max(12).optional(),
+	harder: draftHarderTasksSchema.optional()
 });
 /** One drafted task; every `expectedTools` name is one the request listed. */
 const draftedSimulationTaskSchema = object$4({
@@ -73282,11 +73360,18 @@ object$4({
 	trials: number$2().int().min(1).max(10).optional(),
 	trigger: simulationRunTriggerSchema.optional(),
 	/** The variant the others are compared against: one of the run's variants, over the simulation's own baseline. */
-	baselineVariantId: entityId.optional()
+	baselineVariantId: entityId.optional(),
+	/**
+	* A recording run: these server tools answer live on every variant, so the recorder keeps their real answers for
+	* later replay. Every variant must be today's version (no overlay), and every target must record in `full`.
+	*/
+	recordTools: array(string$2().min(1).max(255)).min(1).max(64).optional()
 });
 discriminatedUnion("dryRun", [object$4({
 	dryRun: literal(true),
-	estimate: runEstimateSchema
+	estimate: runEstimateSchema,
+	/** How each variant target would record this run, keyed by target id: the kill switch, then the target (D1). */
+	recording: record$2(string$2(), frozenTargetRecordingSchema).optional()
 }), object$4({
 	dryRun: literal(false),
 	runId: entityId,
@@ -73298,6 +73383,12 @@ const simulationRunResponseSchema = object$4({
 	run: simulationRunSchema,
 	reportId: entityId.nullable()
 });
+object$4({ 
+/** Comma-separated server tools; the response then counts this run's live calls to each that the recorder kept. */
+recordedCallsFor: string$2().max(4096).optional() });
+simulationRunResponseSchema.extend({ 
+/** Per tool named in `recordedCallsFor`: this run's live calls the recorder kept, 0 for a tool with none. */
+recordedCalls: record$2(string$2(), number$2().int()).optional() });
 object$4({ run: simulationRunSchema });
 simulationRunResponseSchema.extend({ repair: object$4({
 	/** Episodes whose AI judge failed, graded again from their stored transcripts. */

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@runtypelabs/cli",
-  "version": "2.57.0",
+  "version": "2.58.0",
   "description": "Command-line interface for Runtype AI platform",
   "type": "module",
   "main": "dist/index.js",
@@ -12,8 +12,8 @@
     "@modelcontextprotocol/client": "2.3.1",
     "@modelcontextprotocol/core": "2.3.1",
     "@napi-rs/keyring": "2.1.0",
-    "@runtypelabs/ink-components": "0.3.13",
-    "@runtypelabs/sdk": "11.17.0",
+    "@runtypelabs/ink-components": "0.3.14",
+    "@runtypelabs/sdk": "11.18.2",
     "@runtypelabs/terminal-animations": "0.2.4",
     "add-mcp": "2.6.0",
     "chalk": "^5.3.0",
@@ -37,10 +37,10 @@
   "devDependencies": {
     "@modelcontextprotocol/server": "2.3.1",
     "@runtypelabs/device-bridge": "0.2.5",
-    "@runtypelabs/mcp-cli": "0.1.0",
-    "@runtypelabs/model-execution": "0.41.6",
+    "@runtypelabs/mcp-cli": "0.2.1",
+    "@runtypelabs/model-execution": "0.41.8",
     "@runtypelabs/persona": "4.34.0",
-    "@runtypelabs/shared": "3.110.0",
+    "@runtypelabs/shared": "3.113.0",
     "@types/express": "^5.0.6",
     "@types/micromatch": "^4.0.9",
     "@types/node": "^25.3.3",