llmz 0.7.1 → 0.9.0

package/dist/{execute-msui51b2.cjs → execute-DM2-7y86.cjs}

@@ -1,18 +1,19 @@
-const require__baseForOwn = require("./_baseForOwn-buOZDFbK.cjs");
-const require_dual_modes = require("./dual-modes-CfqTk9M8.cjs");
-const require_custom_client = require("./custom-client-Bp1TgTKk.cjs");
-const require_utils = require("./utils-ZpJm00ed.cjs");
-const require_jsx = require("./jsx-BWvT6ukg.cjs");
+const require_compiler = require("./compiler-C-OI9sim.cjs");
+const require_dual_modes = require("./dual-modes-DPXiuvNe.cjs");
+const require_common = require("./common-BnvrDnIT.cjs");
+const require_custom_client = require("./custom-client-DAsGRnYc.cjs");
+const require_utils = require("./utils-3ESwP3yg.cjs");
+const require_jsx = require("./jsx-owqxrquU.cjs");
 const require_stack_traces = require("./stack-traces-ntw_qhoy.cjs");
 const require_errors = require("./errors-C962uJdS.cjs");
-const require_truncator = require("./truncator-B6aaJaN1.cjs");
-const require_vm = require("./vm-BjkJqSQH.cjs");
+const require_truncator = require("./truncator-CLkouTCO.cjs");
+const require_vm = require("./vm-D6s1zF6M.cjs");
 let _bpinternal_zui = require("@bpinternal/zui");
 let ulid = require("ulid");
 let _botpress_client = require("@botpress/client");
 let _botpress_cognitive = require("@botpress/cognitive");
 let ms = require("ms");
-ms = require__baseForOwn.__toESM(ms, 1);
+ms = require_compiler.__toESM(ms, 1);
 //#region ../../node_modules/.pnpm/lodash-es@4.17.21/node_modules/lodash-es/_baseClamp.js
 /**
 * The base implementation of `_.clamp` which doesn't coerce arguments.
@@ -128,35 +129,7 @@ const assertSuccessfulGeneration = (metadata) => {
 	if (metadata.stopReason === "max_tokens" || metadata.stopReason === "content_filter") throw new require_errors.CognitiveError(`LLM generation did not complete: stopReason=${metadata.stopReason}`);
 };
 const getModelOutputLimit = (inputLength) => clamp(RESPONSE_LENGTH_BUFFER.PERCENTAGE * inputLength, RESPONSE_LENGTH_BUFFER.MIN_TOKENS, RESPONSE_LENGTH_BUFFER.MAX_TOKENS);
-/**
-* Models sometimes wrap their whole response in a code fence. The fence has to
-* be removed before it reaches the incremental parser to avoid a spurious
-* unexpected-text diagnostic. Works on arbitrary chunk boundaries
-* by holding content back until the first newline.
-*/
-var LeadingFenceFilter = class {
-	_buffer = "";
-	_done = false;
-	push(text) {
-		if (this._done) return text;
-		this._buffer += text;
-		const newline = this._buffer.indexOf("\n");
-		if (newline === -1) return "";
-		this._done = true;
-		const firstLine = this._buffer.slice(0, newline);
-		const rest = this._buffer.slice(newline + 1);
-		this._buffer = "";
-		return firstLine.trim().startsWith("```") ? rest : `${firstLine}\n${rest}`;
-	}
-	flush() {
-		if (this._done) return "";
-		this._done = true;
-		const buffered = this._buffer;
-		this._buffer = "";
-		return buffered.trim().startsWith("```") ? "" : buffered;
-	}
-};
-const generateCode = async ({ iteration, ctx, cognitive, controller, metadata, onSend, onSendDelta, onRunStart, onRunComplete }) => {
+const generateCode = async ({ iteration, ctx, cognitive, controller, metadata, onSend, onSendDelta, onRunStart }) => {
 	const startedAt = Date.now();
 	const traces = iteration.traces;
 	const modelRef = Array.isArray(iteration.model) ? iteration.model[0] : iteration.model;
@@ -219,6 +192,7 @@ const generateCode = async ({ iteration, ctx, cognitive, controller, metadata, o
 	const liveContent = /* @__PURE__ */ new Map();
 	let codeGenerationTraced = false;
 	let runCompleted = false;
+	let completions = [];
 	const dispatchSends = async (events) => {
 		for (const event of events) if (event.type === "item-start") {
 			liveItems.set(event.item.id, event.item);
@@ -232,7 +206,7 @@ const generateCode = async ({ iteration, ctx, cognitive, controller, metadata, o
 			}
 		} else if (event.type === "body-delta" && onSendDelta) {
 			const item = liveItems.get(event.itemId);
-			if ((item === null || item === void 0 ? void 0 : item.kind) !== "send") continue;
+			if ((item === null || item === void 0 ? void 0 : item.kind) !== "send" || runCompleted) continue;
 			const content = (liveContent.get(item.id) ?? "") + event.delta;
 			liveContent.set(item.id, content);
 			const delta = {
@@ -245,21 +219,19 @@ const generateCode = async ({ iteration, ctx, cognitive, controller, metadata, o
 			};
 			await preview(delta);
 		} else if (event.type === "item-complete") {
-			if (event.item.kind === "send" && onSend) await onSend({
-				name: event.item.name,
-				props: event.item.props,
-				body: event.item.body
-			}, messageMetadata(event.item.id));
-			else if (event.item.kind === "run" && event.item.status === "complete" && !runCompleted) {
-				runCompleted = true;
-				const code = (event.item.body ?? "").trim();
-				onRunComplete === null || onRunComplete === void 0 || onRunComplete(code);
-			}
+			if (event.item.kind === "send" && onSend && !runCompleted) {
+				const send = {
+					name: event.item.name,
+					props: event.item.props,
+					body: event.item.body
+				};
+				const metadata = messageMetadata(event.item.id);
+				completions.push(() => onSend(send, metadata));
+			} else if (event.item.kind === "run" && event.item.status === "complete" && !runCompleted) runCompleted = true;
 		}
 	};
 	if (typeof cognitive.generateTextStream === "function") {
-		let parser = new require_dual_modes.StreamingMessageParser();
-		let fence = new LeadingFenceFilter();
+		let parser = new require_common.ResponseParser();
 		const streamController = createJoinedAbortController([controller.signal]);
 		const requestedAt = Date.now();
 		const stream = cognitive.generateTextStream({
@@ -287,6 +259,7 @@ const generateCode = async ({ iteration, ctx, cognitive, controller, metadata, o
 		};
 		raw = "";
 		let streamCompleted = false;
+		let accepted = false;
 		try {
 			while (true) {
 				var _chunk$value, _chunk$value2, _chunk$value3, _chunk$value4;
@@ -311,8 +284,9 @@ const generateCode = async ({ iteration, ctx, cognitive, controller, metadata, o
 						...chunk.value.restart
 					});
 					raw = "";
-					parser = new require_dual_modes.StreamingMessageParser();
-					fence = new LeadingFenceFilter();
+					completions = [];
+					accepted = false;
+					parser = new require_common.ResponseParser();
 					liveItems.clear();
 					liveContent.clear();
 					codeGenerationTraced = false;
@@ -334,19 +308,52 @@ const generateCode = async ({ iteration, ctx, cognitive, controller, metadata, o
 				timeToLastToken = Date.now() - requestedAt;
 				timeToFirstToken ??= timeToLastToken;
 				raw += delta;
-				await dispatchSends(parser.push(fence.push(delta)));
+				await dispatchSends(parser.push(delta));
 			}
 			if (!responseMetadata) throw new require_errors.CognitiveError("LLM streaming completed without metadata");
 			assertSuccessfulGeneration(responseMetadata);
-			const remaining = fence.flush();
-			if (remaining) await dispatchSends(parser.push(remaining));
-			await dispatchSends(parser.finish());
-			assistantResponse = require_dual_modes.toParsedAssistantResponse(parser.items, raw, parser.diagnostics);
+			await dispatchSends(parser.finish(responseMetadata.stopReason));
+			if (parser.valid) {
+				for (const complete of completions) await complete();
+				accepted = true;
+			}
+			assistantResponse = require_common.toParsedAssistantResponse(parser.items, raw, parser.diagnostics);
+		} catch (error) {
+			parser.finish();
+			const usage = (responseMetadata === null || responseMetadata === void 0 ? void 0 : responseMetadata.usage) ?? {
+				inputTokens: 0,
+				outputTokens: 0,
+				inputCost: 0,
+				outputCost: 0
+			};
+			iteration.llm = {
+				started_at: startedAt,
+				ended_at: Date.now(),
+				status: "error",
+				cached: (responseMetadata === null || responseMetadata === void 0 ? void 0 : responseMetadata.cached) ?? false,
+				tokens: usage.inputTokens + usage.outputTokens,
+				spend: (responseMetadata === null || responseMetadata === void 0 ? void 0 : responseMetadata.cost) ?? usage.inputCost + usage.outputCost,
+				output: raw,
+				diagnostics: parser.diagnostics,
+				model: (responseMetadata === null || responseMetadata === void 0 ? void 0 : responseMetadata.model) ?? model.id,
+				time_to_first_token: timeToFirstToken,
+				time_to_last_token: timeToLastToken,
+				usage
+			};
+			throw error;
 		} finally {
 			streamController.abort("LLM stream closed");
 			if (!streamCompleted) stream.return(void 0).catch((err) => {});
+			if (!accepted && liveContent.size) await preview({
+				restart: true,
+				iterationId: iteration.id,
+				attempt: attempt + 1,
+				fromModel: model.id,
+				toModel: model.id,
+				reason: "invalid or incomplete response envelope"
+			});
 		}
-		if (midStreamFallback) controller.signal.throwIfAborted();
+		controller.signal.throwIfAborted();
 	} else {
 		const response = await cognitive.generateText(input, { signal: controller.signal }).catch((thrown) => {
 			throw new require_errors.CognitiveError(`LLM generation failed: ${require_utils.getErrorMessage(thrown)}`);
@@ -356,7 +363,7 @@ const generateCode = async ({ iteration, ctx, cognitive, controller, metadata, o
 		responseMetadata = response.metadata;
 		assertSuccessfulGeneration(responseMetadata);
 		raw = response.output;
-		assistantResponse = ctx.version.parseAssistantResponse(raw);
+		assistantResponse = ctx.version.parseAssistantResponse(raw, responseMetadata.stopReason);
 		for (const [index, send] of assistantResponse.sends.entries()) await (onSend === null || onSend === void 0 ? void 0 : onSend(send, messageMetadata(`send-${index}`)));
 	}
 	iteration.code = assistantResponse.code;
@@ -398,6 +405,7 @@ const generateCode = async ({ iteration, ctx, cognitive, controller, metadata, o
 //#endregion
 //#region src/runtime/interpret-result.ts
 const interpretVMResult = async ({ iteration, result, controller, startedAt, onExit }) => {
+	Object.assign(iteration.variables, Object.fromEntries(Object.entries(result.variables ?? {}).filter(([, value]) => value !== "[[non-primitive]]")));
 	if (result.error && result.error instanceof require_errors.InvalidCodeError) {
 		iteration.end({
 			type: "invalid_code_error",
@@ -445,7 +453,8 @@ const interpretVMResult = async ({ iteration, result, controller, startedAt, onE
 			thinking_requested: {
 				variables: result.signal.context,
 				reason: result.signal.reason,
-				metadata: result.signal.metadata
+				metadata: result.signal.metadata,
+				interrupted: true
 			}
 		});
 		return;
@@ -458,7 +467,7 @@ const interpretVMResult = async ({ iteration, result, controller, startedAt, onE
 		return;
 	}
 	const returnValue = result.return_value;
-	const codeRequestsInspection = require_vm.hasTopLevelReturn(iteration.code ?? "");
+	const codeRequestsInspection = require_compiler.hasTopLevelReturn(iteration.code ?? "");
 	if (iteration.next && !codeRequestsInspection) {
 		await applyNextExit({
 			iteration,
@@ -506,11 +515,14 @@ const applyNextExit = async ({ iteration, controller, onExit }) => {
 		value: parsedExit.success ? parsedExit.value : next.props
 	};
 	if (!parsedExit.success) {
+		var _parsedExit$validatio;
+		const details = (_parsedExit$validatio = parsedExit.validationErrors) === null || _parsedExit$validatio === void 0 ? void 0 : _parsedExit$validatio.join("; ");
+		const error = [parsedExit.error.split("\n")[0], details].filter(Boolean).join(". ");
 		iteration.end({
 			type: "exit_error",
 			exit_error: {
 				exit: next.name,
-				message: parsedExit.error,
+				message: `${error}. Received props: ${JSON.stringify(next.props)}. Choose an exit from <exits> and supply its required props as ONE JSON object on the SAME LINE as ■next=${next.name}. Do not use a JavaScript return statement or wrap the fields in "value" or "props".`,
 				return_value: returnValue
 			}
 		});
@@ -750,6 +762,7 @@ const executeContextInternal = async (props) => {
 	const ctx = new require_custom_client.Context({
 		chat: props.chat,
 		instructions: props.instructions,
+		examples: props.examples,
 		objects: props.objects,
 		tools: props.tools,
 		loop: (_props$options = props.options) === null || _props$options === void 0 ? void 0 : _props$options.loop,
@@ -950,8 +963,6 @@ const executeIteration = async ({ iteration, ctx, cognitive, controller, onExit,
 			};
 		});
 	};
-	const canExecuteEarly = typeof onBeforeExecution !== "function" && !ctx.midStreamFallback;
-	let earlyExecution;
 	await generateCode({
 		iteration,
 		ctx,
@@ -959,22 +970,15 @@ const executeIteration = async ({ iteration, ctx, cognitive, controller, onExit,
 		controller,
 		metadata,
 		onRunStart: () => require_vm.warmupVM(),
-		onRunComplete: canExecuteEarly ? (code) => {
-			if (!code.length || controller.signal.aborted || earlyExecution) return;
-			iteration.code = code;
-			earlyExecution = {
-				code,
-				started_at: Date.now(),
-				promise: runCode(code)
-			};
-		} : void 0,
 		onSend: async (send, messageMetadata) => {
 			if (!ctx.chat) return;
 			const sendStartedAt = Date.now();
-			const component = require_jsx.createJsxComponent({
+			const registered = iteration.components.find((component) => component.definition.name.toLowerCase() === send.name);
+			const children = send.body ? [send.body] : [];
+			const component = registered ? registered.render(send.props, children) : require_jsx.createJsxComponent({
 				type: send.name,
 				props: send.props,
-				children: send.body ? [send.body] : []
+				children
 			});
 			try {
 				await ctx.chat.handler(component, messageMetadata);
@@ -990,7 +994,6 @@ const executeIteration = async ({ iteration, ctx, cognitive, controller, onExit,
 		},
 		onSendDelta: ((_ctx$chat = ctx.chat) === null || _ctx$chat === void 0 ? void 0 : _ctx$chat.onMessageDelta) ? (delta) => ctx.chat.onMessageDelta(delta) : void 0
 	});
-	if (earlyExecution) iteration.code = earlyExecution.code;
 	if (typeof onBeforeExecution === "function") try {
 		const hookRes = await onBeforeExecution(iteration, controller);
 		if (typeof (hookRes === null || hookRes === void 0 ? void 0 : hookRes.code) === "string" && hookRes.code.trim().length > 0) iteration.code = hookRes.code.trim();
@@ -1028,7 +1031,7 @@ const executeIteration = async ({ iteration, ctx, cognitive, controller, onExit,
 		return;
 	}
 	if (!iteration.code) {
-		var _iteration$sends;
+		var _iteration$llm;
 		if (iteration.next) {
 			await applyNextExit({
 				iteration,
@@ -1037,26 +1040,14 @@ const executeIteration = async ({ iteration, ctx, cognitive, controller, onExit,
 			});
 			return;
 		}
-		if (ctx.chat && ((_iteration$sends = iteration.sends) === null || _iteration$sends === void 0 ? void 0 : _iteration$sends.length)) {
-			iteration.next = {
-				name: "listen",
-				props: {}
-			};
-			await applyNextExit({
-				iteration,
-				controller,
-				onExit
-			});
-			return;
-		}
 		iteration.end({
 			type: "invalid_code_error",
-			invalid_code_error: { message: "The response did not include a ■run block or a ■next exit. Reply using ■ blocks and end your response with ■run or ■next=<exit>." }
+			invalid_code_error: { message: ((_iteration$llm = iteration.llm) === null || _iteration$llm === void 0 || (_iteration$llm = _iteration$llm.diagnostics) === null || _iteration$llm === void 0 || (_iteration$llm = _iteration$llm.find((diagnostic) => diagnostic.code === "invalid-envelope")) === null || _iteration$llm === void 0 ? void 0 : _iteration$llm.message) ?? "The response did not include a ■run block or a ■next exit. Reply with ■start, the required ■ blocks, and ■end." }
 		});
 		return;
 	}
-	const startedAt = (earlyExecution === null || earlyExecution === void 0 ? void 0 : earlyExecution.started_at) ?? Date.now();
-	const result = earlyExecution ? await earlyExecution.promise : await runCode(iteration.code ?? "");
+	const startedAt = Date.now();
+	const result = await runCode(iteration.code);
 	await interpretVMResult({
 		iteration,
 		result,

package/dist/{exit-BGi0Kk0y.cjs → exit-DUM96DN1.cjs}

@@ -1,9 +1,9 @@
-const require__baseForOwn = require("./_baseForOwn-buOZDFbK.cjs");
+const require_compiler = require("./compiler-C-OI9sim.cjs");
 const require_uniq = require("./uniq-G872RWmE.cjs");
-const require_utils = require("./utils-ZpJm00ed.cjs");
+const require_utils = require("./utils-3ESwP3yg.cjs");
 let _bpinternal_zui = require("@bpinternal/zui");
 //#region src/exit.ts
-var exit_exports = /* @__PURE__ */ require__baseForOwn.__exportAll({ Exit: () => Exit });
+var exit_exports = /* @__PURE__ */ require_compiler.__exportAll({ Exit: () => Exit });
 /**
 * Defines how LLMz agent execution can terminate.
 *

package/dist/exit-parser.d.ts

@@ -9,6 +9,8 @@ export type ParsedExit = {
 } | {
     success: false;
     error: string;
+    /** Field-level schema errors, separate from the legacy return-statement guidance. */
+    validationErrors?: string[];
     returnValue: unknown;
 };
 /**