blun-king-cli 9.1.164 → 9.1.165

package/bin/turn-thinking-policy.cjs

@@ -8,6 +8,7 @@ const STATUS_ONLY_CHECK_IN = /(?:\bstandabfrage\b|\bstatusabfrage\b|\bkurzes?\s+
 const STATUS_FOLLOW_UP_ACTION = /\b(?:aendere|arbeite|baue|build|change|delete|deploy|deploye|fahr|fix|fixe|implement|implementiere|install|installiere|loesche|mach|patch|pruefe|repariere|run|schreibe|setze|start|starte|stop|stoppe|test|teste|write)\b/iu;
 const LIGHTWEIGHT_STEER = /\b(?:cool|danke|dankeschoen|gute arbeit|passt|perfekt|stolz|super|freut mich)\b/iu;
 const ADAPTIVE_LOW_COMPLETION_TOKENS = 8192;
+const STANDARD_TURN_MAX_COMPLETION_TOKENS = 24_576;
 
 function selectThinkingEffortForTurn(text, originKind) {
   if (originKind !== "user") return undefined;
@@ -52,6 +53,13 @@ function capCompletionBudgetForAdaptiveEffort(budget, thinkingEffort) {
   };
 }
 
+function capTurnCompletionTokens(maxOutputSize) {
+  if (!Number.isFinite(maxOutputSize) || maxOutputSize <= 0) {
+    return STANDARD_TURN_MAX_COMPLETION_TOKENS;
+  }
+  return Math.min(Math.floor(maxOutputSize), STANDARD_TURN_MAX_COMPLETION_TOKENS);
+}
+
 function selectThinkingEffortForBufferedSteer(text, originKind) {
   const classified = selectThinkingEffortForTurn(text, originKind);
   if (classified !== undefined) return classified;
@@ -73,7 +81,9 @@ module.exports = {
   ACTIONABLE_PROMPT,
   ADAPTIVE_LOW_COMPLETION_TOKENS,
   EXPLICIT_THINKING_INTENT,
+  STANDARD_TURN_MAX_COMPLETION_TOKENS,
   capCompletionBudgetForAdaptiveEffort,
+  capTurnCompletionTokens,
   selectThinkingEffortForBufferedSteer,
   selectThinkingEffortForWorkStep,
   selectThinkingEffortForTurn,

package/blun.mjs

@@ -264540,7 +264540,7 @@ var init_llm_request_logger = __esmMin((() => {
 function normalizeRuntimeSystemPromptAppend(value) {
 	return value?.trim().slice(0, 16e3) ?? "";
 }
-var PINNED_MODEL_ALIASES, resolveProviderIdleTimeoutMs, capCompletionBudgetForAdaptiveEffort, selectThinkingEffortForBufferedSteer, selectThinkingEffortForWorkStep, selectThinkingEffortForTurn, Agent$1;
+var PINNED_MODEL_ALIASES, resolveProviderIdleTimeoutMs, capCompletionBudgetForAdaptiveEffort, capTurnCompletionTokens, selectThinkingEffortForBufferedSteer, selectThinkingEffortForWorkStep, selectThinkingEffortForTurn, Agent$1;
 var init_agent = __esmMin((() => {
 	init_dist$6();
 	init_config$4();
@@ -264578,7 +264578,7 @@ var init_agent = __esmMin((() => {
 	init_session_loop();
 	init_error_memory();
 	({ resolveProviderIdleTimeoutMs } = createRequire(import.meta.url)("./bin/provider-idle-timeout-policy.cjs"));
-	({ capCompletionBudgetForAdaptiveEffort, selectThinkingEffortForBufferedSteer, selectThinkingEffortForWorkStep, selectThinkingEffortForTurn } = createRequire(import.meta.url)("./bin/turn-thinking-policy.cjs"));
+	({ capCompletionBudgetForAdaptiveEffort, capTurnCompletionTokens, selectThinkingEffortForBufferedSteer, selectThinkingEffortForWorkStep, selectThinkingEffortForTurn } = createRequire(import.meta.url)("./bin/turn-thinking-policy.cjs"));
 	PINNED_MODEL_ALIASES = new Set(["king-blun", "pruefstand"]);
 	Agent$1 = class {
 		type;
@@ -264769,7 +264769,7 @@ var init_agent = __esmMin((() => {
 			const provider = thinkingEffort === void 0 ? runtimeModel.provider : runtimeModel.provider.withThinking(thinkingEffort);
 			const loopControl = this.blunConfig?.loopControl;
 			const completionBudgetConfig = resolveCompletionBudget({
-				maxOutputSize: runtimeModel.maxOutputSize,
+				maxOutputSize: capTurnCompletionTokens(runtimeModel.maxOutputSize),
 				reservedContextSize: loopControl?.reservedContextSize
 			});
 			return new KosongLLM({

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "blun-king-cli",
-  "version": "9.1.164",
+  "version": "9.1.165",
   "description": "BLUN CLI - your own AI agent with a Telegram channel. Get it done. With BLUN.",
   "license": "MIT",
   "bin": {