blun-king-cli 9.1.610 → 9.1.611

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "blun-king-cli",
-  "version": "9.1.610",
+  "version": "9.1.611",
   "description": "BLUN CLI - your own AI agent with a Telegram channel. Get it done. With BLUN.",
   "license": "MIT",
   "bin": {

package/telegram-plugin/dist/mcp-server.mjs

@@ -73461,6 +73461,10 @@ function assertTelegramLanguageReleased(text, releaseToken, now) {
 function isGroupChat(chatId) {
 	return chatId.startsWith("-");
 }
+const SILENT_TURN_MARKER = "[[BLUN_TELEGRAM_SILENT]]";
+function isSilentTurnMarker(text) {
+	return text.trim() === SILENT_TURN_MARKER;
+}
 function normalize(s) {
 	return s.toLowerCase().replace(/ä/g, "a").replace(/ö/g, "o").replace(/ü/g, "u").replace(/ß/g, "ss").replace(/[‐-―]/g, " ").replace(/[^\p{L}\p{N}\s]/gu, " ").replace(/\s+/g, " ").trim();
 }
@@ -73617,6 +73621,7 @@ async function runTool(api, token, name, args) {
 		case "download_attachment": return runDownload(api, token, args["file_id"]);
 		case "edit_message": {
 			const chatId = resolveAllowedChatId(args["chat_id"]);
+			if (isSilentTurnMarker(args["text"])) throw new Error("silent marker cannot be used to edit a Telegram message");
 			assertTelegramLanguageReleased(args["text"], args["release_token"]);
 			const parseMode = args["format"] === "markdownv2" ? "MarkdownV2" : void 0;
 			const edited = await api.editMessageText(chatId, Number(args["message_id"]), args["text"], parseMode !== void 0 ? { parse_mode: parseMode } : void 0);
@@ -73672,6 +73677,10 @@ async function runReply(api, args) {
 	const replyTo = args["reply_to"] !== null && args["reply_to"] !== void 0 ? Number(args["reply_to"]) : void 0;
 	const files = args["files"] ?? [];
 	const parseMode = args["format"] === "markdownv2" ? "MarkdownV2" : void 0;
+	if (isSilentTurnMarker(text)) {
+		if (files.length > 0) throw new Error("silent marker cannot accompany files");
+		return "suppressed: silent marker - not sent";
+	}
 	if (files.length === 0 && isGroupChat(chatId) && isGroupNoiseReply(text)) {
 		appendJsonl(outboxLog(), {
 			direction: "out",

package/worker-host.mjs

@@ -1,5 +1,5 @@
 #!/usr/bin/env node
-// BLUN_BUILD_INPUT_SHA256:8c6e16ea727bc28d31e8a784ec7ab7cc656e35f70f996bb1e64df97776875360
+// BLUN_BUILD_INPUT_SHA256:0533ca113dc07cc119988be5e76555dec986b6881903f8e2b537f5adf96cc2c5
 import { fileURLToPath as __cjsShimFileURLToPath } from 'node:url';
 import { dirname as __cjsShimDirname } from 'node:path';
 const __filename = __cjsShimFileURLToPath(import.meta.url);
@@ -230115,6 +230115,7 @@ function buildCompactionSummaryText(summary) {
 const MAX_SESSION_COMPACTION_FAILURES = 2;
 const DEFAULT_COMPACTION_MAX_COMPLETION_TOKENS = 128 * 1024;
 const COMPACTION_SUMMARY_RESERVE_RATIO = .1;
+const BLUN_COMPACTION_REQUEST_BUDGET_BYTES = 75e4;
 const MAX_HIERARCHICAL_COMPACTION_PASSES = 64;
 const HIERARCHICAL_COMPACTION_PREFIX = "This is a complete summary of an earlier chronological segment. Preserve it as source material when merging it with the following conversation:";
 var CompactionTruncatedError = class extends Error {
@@ -230604,6 +230605,7 @@ var FullCompaction = class {
 			let charsReceived = 0;
 			let lastProgressEmitAt = 0;
 			let initialCompactionRequestTokens;
+			let initialCompactionRequestBytes;
 			const compactionTools = [];
 			const buildRequestMessages = (history, requestInstruction) => {
 				return stripMediaForCompaction(downgradeUnsupportedMedia([...this.agent.context.project(history, {
@@ -230611,6 +230613,7 @@ var FullCompaction = class {
 					dropOrphanResults: true
 				}), createUserMessage(requestInstruction)], capability));
 			};
+			const estimateRequestBytes = (messages) => Buffer.byteLength(JSON.stringify(messages), "utf8") + Buffer.byteLength(this.agent.config.systemPrompt, "utf8");
 			while (true) {
 				signal.throwIfAborted();
 				if (hierarchicalPassCount >= MAX_HIERARCHICAL_COMPACTION_PASSES) throw new BlunError(ErrorCodes.COMPACTION_FAILED, `Compaction stopped after ${String(MAX_HIERARCHICAL_COMPACTION_PASSES)} staged requests without reaching a final summary. The conversation history was not changed.`, { details: {
@@ -230620,30 +230623,39 @@ var FullCompaction = class {
 				const compactionRequestLimit = this.getEffectiveMaxContextTokens(model);
 				const safeCompactionRequestLimit = compactionRequestLimit > 0 ? Math.max(1, Math.floor(compactionRequestLimit * (1 - COMPACTION_SUMMARY_RESERVE_RATIO))) : compactionRequestLimit;
 				let messages = buildRequestMessages(historyForModel, instruction);
+				const requestByteBudget = provider.name === "blun" ? BLUN_COMPACTION_REQUEST_BUDGET_BYTES : void 0;
 				let estimatedCompactionRequestTokens = this.estimateRequestTokens(messages, this.agent.effectiveSystemPrompt, compactionTools);
+				let estimatedCompactionRequestBytes = requestByteBudget === void 0 ? void 0 : estimateRequestBytes(messages);
 				initialCompactionRequestTokens ??= estimatedCompactionRequestTokens;
+				initialCompactionRequestBytes ??= estimatedCompactionRequestBytes;
 				let hierarchicalChunkEnd;
-				if (safeCompactionRequestLimit > 0 && estimatedCompactionRequestTokens >= safeCompactionRequestLimit) {
+				const tokenBudgetExceeded = safeCompactionRequestLimit > 0 && estimatedCompactionRequestTokens >= safeCompactionRequestLimit;
+				const byteBudgetExceeded = requestByteBudget !== void 0 && estimatedCompactionRequestBytes !== void 0 && estimatedCompactionRequestBytes >= requestByteBudget;
+				if (tokenBudgetExceeded || byteBudgetExceeded) {
 					if (historyForModel.length > 1) {
 						const chunkInstruction = `${instruction}\n\nThis request contains only the oldest chronological segment. Summarize every fact needed by a later merge pass. Do not assume later messages are visible.`;
-						const chunk = selectHierarchicalCompactionChunk(historyForModel, safeCompactionRequestLimit, safeCompactionRequestLimit, (candidate) => {
+						const chunk = selectHierarchicalCompactionChunk(historyForModel, safeCompactionRequestLimit, safeCompactionRequestLimit, requestByteBudget, (candidate) => {
 							const candidateMessages = buildRequestMessages(candidate, chunkInstruction);
 							return {
 								messages: candidateMessages,
-								estimatedTokens: this.estimateRequestTokens(candidateMessages, this.agent.effectiveSystemPrompt, compactionTools)
+								estimatedTokens: this.estimateRequestTokens(candidateMessages, this.agent.effectiveSystemPrompt, compactionTools),
+								estimatedBytes: requestByteBudget === void 0 ? void 0 : estimateRequestBytes(candidateMessages)
 							};
 						});
 						if (chunk !== void 0) {
 							hierarchicalChunkEnd = chunk.end;
 							messages = chunk.messages;
 							estimatedCompactionRequestTokens = chunk.estimatedTokens;
+							estimatedCompactionRequestBytes = chunk.estimatedBytes;
 						}
 					}
-					if (hierarchicalChunkEnd === void 0) throw new BlunError(ErrorCodes.CONTEXT_OVERFLOW, `Compaction stopped before upload: the conversation (${String(tokensBefore)} tokens) could not be divided into a request below the active model window (${String(compactionRequestLimit)} tokens).`, { details: {
+					if (hierarchicalChunkEnd === void 0) throw new BlunError(ErrorCodes.CONTEXT_OVERFLOW, byteBudgetExceeded ? "Compaction stopped before upload: the conversation could not be divided into a request below the active byte limit." : `Compaction stopped before upload: the conversation (${String(tokensBefore)} tokens) could not be divided into a request below the active model window (${String(compactionRequestLimit)} tokens).`, { details: {
 						contextBrakeBlocked: true,
 						estimatedRequestTokens: estimatedCompactionRequestTokens,
 						maxContextTokens: compactionRequestLimit,
 						requestBudgetTokens: safeCompactionRequestLimit,
+						estimatedRequestBytes: estimatedCompactionRequestBytes,
+						requestBudgetBytes: requestByteBudget,
 						summaryReserveTokens: compactionRequestLimit - safeCompactionRequestLimit,
 						contextUnchanged: true
 					} });
@@ -230652,7 +230664,7 @@ var FullCompaction = class {
 				attemptCount += 1;
 				charsReceived = 0;
 				const stage = hierarchicalPassCount + 1;
-				const estimatedStageCount = estimateCompactionStageCount(initialCompactionRequestTokens, safeCompactionRequestLimit, stage);
+				const estimatedStageCount = estimateCompactionStageCount(initialCompactionRequestTokens, safeCompactionRequestLimit, stage, initialCompactionRequestBytes, requestByteBudget);
 				const estimatedProgressPercent = estimateCompactionProgressPercent(stage, estimatedStageCount);
 				const windowUsagePercent = estimateCompactionWindowUsagePercent(estimatedCompactionRequestTokens, compactionRequestLimit);
 				this.agent.log.info("compaction stage request", {
@@ -230661,6 +230673,8 @@ var FullCompaction = class {
 					estimatedStageCount,
 					activeContextTokens: this.estimateProjectedRequestTokens(),
 					estimatedInputTokens: estimatedCompactionRequestTokens,
+					estimatedInputBytes: estimatedCompactionRequestBytes,
+					requestBudgetBytes: requestByteBudget,
 					windowUsagePercent,
 					safeInputLimitTokens: safeCompactionRequestLimit,
 					maxContextTokens: compactionRequestLimit
@@ -230732,13 +230746,17 @@ var FullCompaction = class {
 					if (hierarchicalChunkEnd !== void 0) {
 						hierarchicalPassCount += 1;
 						const previousTokens = estimateTokensForMessages(historyForModel);
+						const previousBytes = requestByteBudget === void 0 ? void 0 : Buffer.byteLength(JSON.stringify(historyForModel), "utf8");
 						const nextHistory = [createUserMessage(`${HIERARCHICAL_COMPACTION_PREFIX}\n${extractedSummary}`), ...historyForModel.slice(hierarchicalChunkEnd)];
 						const nextTokens = estimateTokensForMessages(nextHistory);
-						if (nextTokens >= previousTokens) throw new BlunError(ErrorCodes.COMPACTION_FAILED, "A staged compaction response did not reduce the pending history. The conversation history was not changed.", { details: {
+						const nextBytes = requestByteBudget === void 0 ? void 0 : Buffer.byteLength(JSON.stringify(nextHistory), "utf8");
+						if (nextTokens >= previousTokens || previousBytes !== void 0 && nextBytes !== void 0 && nextBytes >= previousBytes) throw new BlunError(ErrorCodes.COMPACTION_FAILED, "A staged compaction response did not reduce the pending history. The conversation history was not changed.", { details: {
 							contextUnchanged: true,
 							hierarchicalPassCount,
 							previousTokens,
-							nextTokens
+							nextTokens,
+							previousBytes,
+							nextBytes
 						} });
 						historyForModel = nextHistory;
 						retryCount = 0;
@@ -230934,9 +230952,10 @@ function isModelFallbackStatus(error) {
 function compactionTimingKey(provider) {
 	return `${provider.name}\u0000${provider.modelName}`;
 }
-function estimateCompactionStageCount(initialInputTokens, safeRequestLimitTokens, currentStage) {
-	if (safeRequestLimitTokens <= 0) return currentStage;
-	return Math.max(currentStage, Math.ceil(initialInputTokens / safeRequestLimitTokens));
+function estimateCompactionStageCount(initialInputTokens, safeRequestLimitTokens, currentStage, initialInputBytes, safeRequestLimitBytes) {
+	const tokenStages = safeRequestLimitTokens > 0 ? Math.ceil(initialInputTokens / safeRequestLimitTokens) : 0;
+	const byteStages = initialInputBytes !== void 0 && safeRequestLimitBytes !== void 0 && safeRequestLimitBytes > 0 ? Math.ceil(initialInputBytes / safeRequestLimitBytes) : 0;
+	return Math.max(currentStage, tokenStages, byteStages);
 }
 function estimateCompactionProgressPercent(stage, estimatedStageCount) {
 	if (estimatedStageCount <= 0) return 0;
@@ -230952,7 +230971,7 @@ const COMPACTION_OVERFLOW_SHRINK_RATIOS = [
 	.5,
 	.35
 ];
-function selectHierarchicalCompactionChunk(history, targetRequestTokens, hardRequestLimit, build) {
+function selectHierarchicalCompactionChunk(history, targetRequestTokens, hardRequestLimit, requestByteBudget, build) {
 	const safeEnds = [];
 	for (let end = 1; end < history.length; end++) if (history[end]?.role !== "tool") safeEnds.push(end);
 	if (safeEnds.length === 0) return void 0;
@@ -230964,7 +230983,7 @@ function selectHierarchicalCompactionChunk(history, targetRequestTokens, hardReq
 			const middle = Math.floor((low + high) / 2);
 			const end = safeEnds[middle];
 			const request = build(history.slice(0, end));
-			if (request.estimatedTokens < limit) {
+			if (request.estimatedTokens < limit && (requestByteBudget === void 0 || request.estimatedBytes !== void 0 && request.estimatedBytes < requestByteBudget)) {
 				best = {
 					end,
 					...request
@@ -236771,8 +236790,14 @@ var PersonalMemoryRecallInjector = class {
 	* to read, write, list, or change any personal-memory state.
 	*/
 	filterToolsForTurn(turnId, input, origin, tools) {
-		if (this.toolsAllowedForTurn(turnId, input, origin)) return tools.filter((tool) => !isPersonalMemoryRecallToolName(tool.name) && (!isPersonalMemoryToolName(tool.name) || this.isTrustedHostTool(tool.name)));
-		return tools.filter((tool) => !isPersonalMemoryToolName(tool.name));
+		const brokerAllowed = this.toolsAllowedForTurn(turnId, input, origin);
+		const localSearchAllowed = origin.kind === "user" && this.agent.type === "main" && !isTelegramGroupInput(input) && this.agent.localWorkHistorySearchEligibility?.() === true;
+		return tools.filter((tool) => {
+			if (!isPersonalMemoryToolName(tool.name)) return true;
+			if (isPersonalMemoryRecallToolName(tool.name)) return false;
+			if (tool.name === "mcp__personal-memory__memory_search" && localSearchAllowed) return true;
+			return brokerAllowed && this.isTrustedHostTool(tool.name);
+		});
 	}
 	isTrustedHostTool(name) {
 		const isHostTool = this.agent.isPersonalMemoryHostTool;
@@ -239295,9 +239320,12 @@ var import_turn_tool_performance_policy = (/* @__PURE__ */ __commonJSMin(((expor
 			const name = String(tool?.name || "").trim();
 			if (name && !available.has(name)) available.set(name, tool);
 		}
+		const usedNames = new Set([...selectedTools, ...available.values()].map((tool) => String(tool?.name ?? "")));
+		let loaderName = DEFERRED_TOOL_LOADER_NAME;
+		for (let suffix = 1; usedNames.has(loaderName); suffix += 1) loaderName = `${DEFERRED_TOOL_LOADER_NAME}_${suffix}`;
 		const deferredCount = available.size;
 		const loader = {
-			name: DEFERRED_TOOL_LOADER_NAME,
+			name: loaderName,
 			description: [
 				`Search ${deferredCount} deferred ${deferredCount === 1 ? "tool schema" : "tool schemas"} without loading all definitions.`,
 				"Use select:leaf_name for one known tool, plain keywords to search, or +word to require that word in the tool name.",
@@ -239317,7 +239345,7 @@ var import_turn_tool_performance_policy = (/* @__PURE__ */ __commonJSMin(((expor
 				const query = normalizeDeferredToolQuery(args?.query);
 				return {
 					description: query ? `Searching tool schemas for ${query}` : "Searching tool schemas",
-					approvalRule: DEFERRED_TOOL_LOADER_NAME,
+					approvalRule: loaderName,
 					execute: async () => {
 						let matches = searchDeferredTools([...available.values()], query);
 						let relatedFallback = false;
@@ -239330,7 +239358,7 @@ var import_turn_tool_performance_policy = (/* @__PURE__ */ __commonJSMin(((expor
 							relatedFallback = matches.length > 0;
 						}
 						if (matches.length === 0) {
-							const alreadyLoaded = query.toLowerCase().startsWith("select:") ? searchDeferredTools(selectedTools.filter((tool) => tool?.name !== DEFERRED_TOOL_LOADER_NAME), query) : [];
+							const alreadyLoaded = query.toLowerCase().startsWith("select:") ? searchDeferredTools(selectedTools.filter((tool) => tool?.name !== loaderName), query) : [];
 							if (alreadyLoaded.length === 1) return {
 								isError: false,
 								output: `Tool schema already loaded: ${alreadyLoaded[0].name}. Invoke it now.`
@@ -239340,7 +239368,7 @@ var import_turn_tool_performance_policy = (/* @__PURE__ */ __commonJSMin(((expor
 								output: [
 									`No eligible deferred tool is available for: ${query || "[empty query]"}.`,
 									"Use a loaded alternative or report this integration as unavailable.",
-									"ToolSearch remains available for a different tool needed later."
+									`${loaderName} remains available for a different tool needed later.`
 								].join(" ")
 							};
 						}
@@ -250032,6 +250060,9 @@ function logStepTiming(log, turnId, step, response) {
 		streamDurationMs: timing.streamDurationMs,
 		...timing.serverDecodeMs !== void 0 ? { serverDecodeMs: timing.serverDecodeMs } : {},
 		...timing.clientConsumeMs !== void 0 ? { clientConsumeMs: timing.clientConsumeMs } : {},
+		inputOtherTokens: response.usage.inputOther,
+		cacheReadTokens: response.usage.inputCacheRead,
+		cacheCreationTokens: response.usage.inputCacheCreation,
 		outputTokens: response.usage.output
 	});
 }
@@ -266492,7 +266523,7 @@ var HistoricalToolProjection = class {
 		for (const { index, message, outputPath, text, reference, shellOutput } of candidates) try {
 			context.signal.throwIfAborted();
 			if (!await this.canRead(read, outputPath, context)) continue;
-			const save = this.pending.then(() => this.persist(outputPath, text, context.signal));
+			const save = this.pending.then(() => this.persist(outputPath, text, context.signal, true));
 			this.pending = save.catch(() => void 0);
 			await save;
 			const stat = await this.agent.kaos.stat(outputPath);
@@ -266584,7 +266615,7 @@ var HistoricalToolProjection = class {
 			revision: revision(after)
 		});
 	}
-	async persist(path, text, signal) {
+	async persist(path, text, signal, finalVerificationFollows = false) {
 		signal.throwIfAborted();
 		await checkedAncestors(dirname(dirname(dirname(path))));
 		signal.throwIfAborted();
@@ -266593,7 +266624,7 @@ var HistoricalToolProjection = class {
 		await this.ensureDirectory(dirname(path));
 		signal.throwIfAborted();
 		if (this.files.has(path)) {
-			await this.verifyFile(path, text);
+			if (!finalVerificationFollows) await this.verifyFile(path, text);
 			return;
 		}
 		try {
@@ -267062,7 +267093,7 @@ function blunToolsForOrigin(tools, origin) {
 * the ordinary permission pipeline.
 */
 function blunSelectTurnTools(tools, maxContextTokens, loadedToolNames = /* @__PURE__ */ new Set(), requiredToolNames = /* @__PURE__ */ new Set(), selected = []) {
-	if (estimateTokensForTools(tools) <= (0, import_turn_tool_performance_policy.toolSchemaBudgetTokens)(maxContextTokens) || tools.some((tool) => tool.name === "ToolSearch")) {
+	if (estimateTokensForTools(tools) <= (0, import_turn_tool_performance_policy.toolSchemaBudgetTokens)(maxContextTokens) || tools.some(import_turn_tool_performance_policy.isDeferredToolLoader)) {
 		selected.splice(0, selected.length, ...tools);
 		return selected;
 	}
@@ -271831,6 +271862,7 @@ var Agent$1 = class {
 	}
 	personalMemoryRecallEligibility = () => this.type === "main" && this.personalMemoryHostCapabilityEnabled;
 	isPersonalMemoryHostTool = (name) => this.personalMemoryHostCapabilityEnabled && this.tools.hasUserTool(name);
+	localWorkHistorySearchEligibility = () => this.type === "main" && this.tools.hasUserTool("mcp__personal-memory__memory_search");
 	setPersonalMemoryHostCapability(enabled) {
 		this.personalMemoryHostCapabilityEnabled = enabled === true;
 	}