@iowarp/clio-coder 0.4.0 → 0.4.2

package/dist/{chunk-DQITNCXG.js → chunk-BYMNWQ7O.js}

@@ -5,46 +5,54 @@ import {
 } from "./chunk-CFGTUFWB.js";
 import {
   credentialsPresent
-} from "./chunk-P3FOHJT4.js";
+} from "./chunk-72GZI5EV.js";
 import {
   ceilChars,
   estimateAgentMessageTokens,
   toolSchemaChars
-} from "./chunk-UM7N4G5A.js";
+} from "./chunk-CWVRRIEI.js";
 import {
   getSharedBus
-} from "./chunk-ZGVHUX3M.js";
-import {
-  BusChannels
-} from "./chunk-HLW2MRKE.js";
+} from "./chunk-PMZVFWSR.js";
 import {
   readSettings
-} from "./chunk-VFA6GDY5.js";
+} from "./chunk-34BHNEE3.js";
+import {
+  warnLegacyNaming
+} from "./chunk-D7E7WCDF.js";
+import {
+  BusChannels
+} from "./chunk-F4EKGO4N.js";
 import {
   sleep
-} from "./chunk-FQ4SKYE4.js";
+} from "./chunk-EBY63B7G.js";
 import {
   CLIO_CONTEXT_WINDOW_WARN_BELOW,
   CLIO_MIN_CONTEXT_WINDOW,
   CLIO_MIN_MAX_OUTPUT_TOKENS,
+  EMPTY_CAPABILITIES,
+  VALID_THINKING_LEVELS,
+  availableThinkingLevels,
   extractLocalModelQuirks,
+  inferLocalModelFamily,
+  isHarmonyModelId,
   listLmStudioModels,
   lmStudioReasoningEffort,
   lmStudioRootUrl,
   registerBuiltinRuntimes,
   requestLmStudioJson,
   resolveLmStudioInstance
-} from "./chunk-6DB53AJS.js";
+} from "./chunk-IMXMHHMQ.js";
 import {
   listKnownModelsForRuntime
-} from "./chunk-HI63TFOG.js";
+} from "./chunk-CDNVLKUX.js";
 import {
   authNotRequiredStatus,
   openAuthStorage,
   registerClioOAuthProviders,
   resolveAuthTarget,
   targetRequiresAuth
-} from "./chunk-BBVYXMFO.js";
+} from "./chunk-WNP7O5WZ.js";
 import {
   withStateFileLock
 } from "./chunk-IKCO5N3L.js";
@@ -67,17 +75,16 @@ import {
 } from "./chunk-NUGM5KR6.js";
 import {
   capabilitiesFromCatalogModel,
-  catalogThinkingLevelsForRuntime,
   getCatalogModelForRuntime,
   mergeCapabilities,
   resolveCostProvenance
-} from "./chunk-VO2LKSTM.js";
+} from "./chunk-O42A54GG.js";
 import {
   activateExternalPluginApiBridge,
   calculateEngineCost,
   ensurePiAiRegistered,
   registerEngineApiProvider
-} from "./chunk-UOSL25KY.js";
+} from "./chunk-VXMFAE2W.js";
 import {
   __toESM,
   init_esm_shims
@@ -427,7 +434,7 @@ function evictionNotice(facts, entry) {
     facts,
     "swap",
     "warning",
-    `swapping resident '${entry.modelId}' for requested '${facts.keepModelId}' on '${facts.targetId}' (Clio did not load it; recorded transition instead of a silent unload; set CLIO_CODER_RESIDENCY=observe or lifecycle: user-managed to forbid swaps).`,
+    `swapping resident '${entry.modelId}' for requested '${facts.keepModelId}' on '${facts.targetId}' (Clio did not load it; recorded transition instead of a silent unload; set lifecycle: user-managed to forbid swaps).`,
     {
       ...entry.sizeVramBytes !== void 0 ? { freedVramBytes: entry.sizeVramBytes } : {},
       swappedOut: entry.modelId
@@ -730,16 +737,8 @@ async function reconcileResidency(adapter) {
   }
   return plan;
 }
-function residencyManaged(env = process.env) {
-  const opt = (env.CLIO_CODER_RESIDENCY ?? "").trim().toLowerCase();
-  if (opt === "observe" || opt === "off" || opt === "0" || opt === "false" || opt === "user" || opt === "user-managed") {
-    return false;
-  }
-  return true;
-}
-function residencyManagedFor(lifecycle, env = process.env) {
-  if (lifecycle === "user-managed") return false;
-  return residencyManaged(env);
+function residencyManagedFor(lifecycle) {
+  return lifecycle !== "user-managed";
 }
 
 // src/core/run-overrides.ts
@@ -787,97 +786,6 @@ async function withRunOverrides(overrides, fn) {
   }
 }
 
-// src/domains/providers/types/capability-flags.ts
-init_esm_shims();
-
-// src/domains/providers/model-family.ts
-init_esm_shims();
-function normalizeModelIdForFamily(modelId) {
-  return (modelId ?? "").trim().toLowerCase().replace(/_/g, "-");
-}
-function isHarmonyModelId(modelId) {
-  return normalizeModelIdForFamily(modelId).includes("gpt-oss");
-}
-function inferLocalModelFamily(modelId) {
-  const normalized = normalizeModelIdForFamily(modelId);
-  if (normalized.length === 0) return "unknown";
-  if (normalized.includes("gpt-oss")) return "openai-gpt-oss";
-  if (normalized.includes("nemotron-cascade-2") || normalized.includes("nemotron-cascade2")) {
-    return "nemotron-cascade-2";
-  }
-  if (normalized.includes("nemotron-3-nano-omni") || normalized.includes("nemotron-nano-omni")) {
-    return "nemotron-3-nano-omni";
-  }
-  if (normalized.includes("qwen3")) return "qwen3";
-  if (normalized.includes("gemma-4") || normalized.includes("gemma4") || normalized.includes("gemopus")) {
-    return "gemma-4";
-  }
-  return "unknown";
-}
-
-// src/domains/providers/types/capability-flags.ts
-var EMPTY_CAPABILITIES = {
-  chat: false,
-  tools: false,
-  reasoning: false,
-  vision: false,
-  audio: false,
-  embeddings: false,
-  rerank: false,
-  fim: false,
-  contextWindow: 0,
-  maxTokens: 0
-};
-var VALID_THINKING_LEVELS = ["off", "minimal", "low", "medium", "high", "xhigh", "max"];
-var THINKING_LEVELS_WITHOUT_XHIGH = ["off", "minimal", "low", "medium", "high"];
-var THINKING_LEVELS_OPENAI_5_1_MINI = ["off", "minimal", "low", "medium", "high"];
-var THINKING_LEVELS_OPENAI_5_2_PLUS = [
-  "off",
-  "minimal",
-  "low",
-  "medium",
-  "high",
-  "xhigh"
-];
-var THINKING_LEVELS_OPENAI_5_6 = [...THINKING_LEVELS_OPENAI_5_2_PLUS, "max"];
-var THINKING_LEVELS_HARMONY = ["low", "medium", "high"];
-function normalizeModelId(modelId) {
-  if (!modelId) return void 0;
-  const trimmed = modelId.trim();
-  if (trimmed.length === 0) return void 0;
-  return trimmed.includes("/") ? trimmed.split("/").pop() ?? trimmed : trimmed;
-}
-function availableOpenAICodexThinkingLevels(modelId) {
-  const normalized = normalizeModelId(modelId);
-  if (!normalized) return VALID_THINKING_LEVELS;
-  if (normalized === "gpt-5.1-codex-mini") return THINKING_LEVELS_OPENAI_5_1_MINI;
-  if (normalized.startsWith("gpt-5.2") || normalized.startsWith("gpt-5.3") || normalized.startsWith("gpt-5.4") || normalized.startsWith("gpt-5.5")) {
-    return THINKING_LEVELS_OPENAI_5_2_PLUS;
-  }
-  if (normalized.startsWith("gpt-5.6")) {
-    return THINKING_LEVELS_OPENAI_5_6;
-  }
-  return THINKING_LEVELS_WITHOUT_XHIGH;
-}
-function availableThinkingLevels(caps, options) {
-  if (!caps.reasoning) return ["off"];
-  if (caps.thinkingFormat === "harmony" || isHarmonyModelId(options?.modelId)) {
-    return THINKING_LEVELS_HARMONY;
-  }
-  const catalogLevels = options?.runtimeId && options.modelId ? catalogThinkingLevelsForRuntime(options.runtimeId, options.modelId) : void 0;
-  if (catalogLevels) {
-    if (!catalogLevels.includes("xhigh")) return catalogLevels.filter((level) => level !== "max");
-    return catalogLevels;
-  }
-  if (caps.thinkingFormat === "openai-codex" || options?.runtimeId === "openai-codex") {
-    return availableOpenAICodexThinkingLevels(options?.modelId);
-  }
-  if (caps.thinkingFormat === "anthropic-extended") {
-    return THINKING_LEVELS_WITHOUT_XHIGH;
-  }
-  return THINKING_LEVELS_WITHOUT_XHIGH;
-}
-
 // src/domains/providers/model-capabilities.ts
 init_esm_shims();
 function normalizedModelId(wireModelId) {
@@ -1137,14 +1045,15 @@ function supportedBudgetLevels(baseLevels, quirks) {
 function supportedEffortLevels(baseLevels, quirks, harmony) {
   if (harmony) return HARMONY_LEVELS;
   const efforts = quirks?.thinking?.effortByLevel;
-  if (!efforts) return baseLevels;
+  const byLevelValues = (quirks?.chatTemplateKwargs ?? quirks?.thinking?.chatTemplateKwargs)?.byLevel?.values;
+  if (!efforts && !byLevelValues) return baseLevels;
   const out = [];
   if (baseLevels.includes("off")) out.push("off");
-  if (efforts.minimal !== void 0) out.push("minimal");
-  if (efforts.low !== void 0) out.push("low");
-  if (efforts.medium !== void 0) out.push("medium");
-  if (efforts.high !== void 0) out.push("high");
-  if (efforts.xhigh !== void 0) out.push("xhigh");
+  if (efforts?.minimal !== void 0 || byLevelValues?.minimal !== void 0) out.push("minimal");
+  if (efforts?.low !== void 0 || byLevelValues?.low !== void 0) out.push("low");
+  if (efforts?.medium !== void 0 || byLevelValues?.medium !== void 0) out.push("medium");
+  if (efforts?.high !== void 0 || byLevelValues?.high !== void 0) out.push("high");
+  if (efforts?.xhigh !== void 0 || byLevelValues?.xhigh !== void 0) out.push("xhigh");
   return out.length > 0 ? out : baseLevels;
 }
 function restrictThinkingLevelsByMechanism(levels, mechanism, quirks, options) {
@@ -1280,12 +1189,45 @@ function resolveThinkingCapability(input, quirks, parser) {
     budgetEnforcement
   };
 }
-var REASONING_EFFORT_ON_OFF_RUNTIMES = /* @__PURE__ */ new Set(["lmstudio"]);
+var REASONING_EFFORT_ONLY_RUNTIMES = /* @__PURE__ */ new Set(["lmstudio"]);
 function onOffReasoningEffort(thinkingActive) {
   return thinkingActive ? "low" : "none";
 }
-function resolveRequestCapability(thinking, parser, runtimeId) {
+function resolveFamilyChatTemplateKwargs(quirks, effectiveLevel, configuredLevel) {
+  const block = quirks?.chatTemplateKwargs ?? quirks?.thinking?.chatTemplateKwargs;
+  if (!block) return void 0;
+  const out = {};
+  if (block.static) {
+    for (const [k, v] of Object.entries(block.static)) {
+      out[k] = v;
+    }
+  }
+  if (block.byLevel) {
+    const byLevel = block.byLevel;
+    const val = byLevel.values[effectiveLevel] ?? byLevel.values[configuredLevel];
+    if (val !== void 0) {
+      out[byLevel.key] = val;
+    }
+  }
+  return Object.keys(out).length > 0 ? out : void 0;
+}
+function familyKwargsDeclaredUnsupported(quirks, runtimeId) {
+  if (runtimeId !== "lmstudio") return false;
+  const block = quirks?.chatTemplateKwargs ?? quirks?.thinking?.chatTemplateKwargs;
+  return block?.lmstudio === "unsupported" || block?.byLevel?.lmstudio === "unsupported";
+}
+function resolveRequestCapability(thinking, parser, runtimeId, quirks) {
   const request = { budgetEnforcement: thinking.budgetEnforcement };
+  const effortOnly = REASONING_EFFORT_ONLY_RUNTIMES.has(runtimeId);
+  const familyKwargs = resolveFamilyChatTemplateKwargs(quirks, thinking.effectiveLevel, thinking.configuredLevel);
+  if (familyKwargs && effortOnly) {
+    request.undeliverableChatTemplateKwargs = {
+      keys: Object.keys(familyKwargs),
+      declaredUnsupported: familyKwargsDeclaredUnsupported(quirks, runtimeId)
+    };
+  } else if (familyKwargs) {
+    request.chatTemplateKwargs = { ...familyKwargs };
+  }
   if (thinking.mechanism === "effort-levels" && thinking.effort) {
     request.reasoningEffort = thinking.effort;
   }
@@ -1296,15 +1238,16 @@ function resolveRequestCapability(thinking, parser, runtimeId) {
     request.budgetTokens = thinking.budgetTokens;
   }
   if (thinking.mechanism === "on-off" && thinking.chatTemplateKwargs) {
-    request.chatTemplateKwargs = { ...thinking.chatTemplateKwargs };
-    if (REASONING_EFFORT_ON_OFF_RUNTIMES.has(runtimeId)) {
-      request.reasoningEffort = onOffReasoningEffort(thinking.thinkingActive);
-    }
+    request.chatTemplateKwargs = { ...request.chatTemplateKwargs ?? {}, ...thinking.chatTemplateKwargs };
   }
   if (parser === "harmony" && thinking.effort) {
     request.reasoningEffort = thinking.effort;
     request.chatTemplateKwargs = { ...request.chatTemplateKwargs ?? {}, reasoning_effort: thinking.effort };
   }
+  if (effortOnly && thinking.mechanism !== "none" && thinking.mechanism !== "always-on") {
+    if (!thinking.thinkingActive) request.reasoningEffort = onOffReasoningEffort(false);
+    else if (thinking.mechanism === "on-off") request.reasoningEffort = onOffReasoningEffort(true);
+  }
   return request;
 }
 function resolveModelRuntimeCapabilities(input) {
@@ -1320,7 +1263,7 @@ function resolveModelRuntimeCapabilities(input) {
     family,
     capabilities: input.capabilities,
     thinking,
-    request: resolveRequestCapability(thinking, parser, input.runtimeId),
+    request: resolveRequestCapability(thinking, parser, input.runtimeId, quirks),
     response: {
       parser,
       stripTokenizerSentinels: true
@@ -1404,7 +1347,9 @@ function capabilitiesFromModel(model) {
   return caps;
 }
 function resolveModelRuntimeCapabilitiesForModel(model, configuredThinkingLevel) {
-  const metadata2 = model.clio;
+  const runtimeModel = model;
+  if (runtimeModel.clio !== void 0) warnLegacyNaming("model.clio", "model.clioCoder");
+  const metadata2 = runtimeModel.clioCoder ?? runtimeModel.clio;
   const caps = capabilitiesFromModel(model);
   return resolveModelRuntimeCapabilities({
     targetId: metadata2?.targetId ?? null,
@@ -1816,6 +1761,20 @@ function appendThinkingDiagnostics(diagnostics, resolved, requested) {
   const severity = thinking.noticeKind === "unsupported" || thinking.noticeKind === "always-on" || thinking.noticeKind === "ignored-on-off" ? "warning" : "info";
   diagnostics.push(diagnostic(severity, `thinking-${thinking.noticeKind}`, thinking.notice));
 }
+var CHAT_TEMPLATE_KWARGS_UNDELIVERABLE = "chat-template-kwargs-undeliverable";
+function appendChatTemplateKwargsDiagnostics(diagnostics, resolved) {
+  const undeliverable = resolved.request.undeliverableChatTemplateKwargs;
+  if (!undeliverable || undeliverable.keys.length === 0) return;
+  const keys = undeliverable.keys.join(", ");
+  const declared = undeliverable.declaredUnsupported ? "its family entry marks them unsupported there" : "its family entry does not say so";
+  diagnostics.push(
+    diagnostic(
+      "warning",
+      CHAT_TEMPLATE_KWARGS_UNDELIVERABLE,
+      `chat-template kwargs ${keys} cannot reach ${resolved.runtimeId}, which ignores chat_template_kwargs; ${resolved.modelId} runs without them and ${declared}`
+    )
+  );
+}
 function nonNegativeFiniteNumber(value) {
   return typeof value === "number" && Number.isFinite(value) && value >= 0 ? value : void 0;
 }
@@ -1943,6 +1902,7 @@ function resolveRuntimeTarget(providers, input) {
   const decisions = capabilityDecisions(runtime, capabilities);
   appendCapabilityDiagnostics(diagnostics, input, capabilities, decisions, targetId);
   appendThinkingDiagnostics(diagnostics, modelRuntime, requestedThinkingLevel);
+  appendChatTemplateKwargsDiagnostics(diagnostics, modelRuntime);
   if (hasError(diagnostics)) return { ok: false, diagnostics };
   const resolved = {
     targetId: target.id,
@@ -1982,6 +1942,7 @@ function modelHintPatch(target, model) {
 function withoutStaleRuntimeDiagnostics(diagnostics, decisions) {
   return diagnostics.filter((entry) => {
     if (entry.code.startsWith("thinking-")) return false;
+    if (entry.code === CHAT_TEMPLATE_KWARGS_UNDELIVERABLE) return false;
     if (entry.code === "output-budget-unknown" && decisions.maxTokens > 0) return false;
     if (entry.code === "tools-unsupported" && decisions.tools) return false;
     return true;
@@ -2020,6 +1981,7 @@ function refineRuntimeTargetWithModelHints(target, model, knowledgeBase) {
   const decisions = capabilityDecisions(target.runtime, capabilities);
   const diagnostics = withoutStaleRuntimeDiagnostics(target.diagnostics, decisions);
   appendThinkingDiagnostics(diagnostics, modelRuntime, target.requestedThinkingLevel);
+  appendChatTemplateKwargsDiagnostics(diagnostics, modelRuntime);
   return {
     ...target,
     capabilities,
@@ -2147,6 +2109,15 @@ init_esm_shims();
 function normalizeCostProvenance(value) {
   return value ?? "unknown";
 }
+var COST_PROVENANCE_VALUES = /* @__PURE__ */ new Set([
+  "known",
+  "known_free",
+  "estimated",
+  "unknown"
+]);
+function resolveCostProvenance2(value, fallback) {
+  return typeof value === "string" && COST_PROVENANCE_VALUES.has(value) ? value : fallback;
+}
 
 // src/domains/providers/index.ts
 init_esm_shims();
@@ -2180,6 +2151,9 @@ function residencyTargetKey(runtimeId, baseUrl) {
 // src/engine/gemma-channel-filter.ts
 init_esm_shims();
 import { createAssistantMessageEventStream } from "@earendil-works/pi-ai";
+function usesGemmaChannelMarkers(modelId) {
+  return inferLocalModelFamily(modelId) === "gemma-4";
+}
 var THOUGHT_START = /<\|channel>[^\n]*\n/;
 var BARE_THOUGHT_START = /^\s*(?:thought|own[- ]?(?:thought|think))\s*\n/i;
 var BARE_THOUGHT_ONLY = /^\s*(?:thought|own[- ]?(?:thought|think))\s*$/i;
@@ -2379,7 +2353,6 @@ var globalDefaultMaxOutputTokens = 0;
 function setGlobalDefaultMaxOutputTokens(value) {
   globalDefaultMaxOutputTokens = Number.isFinite(value) && value > 0 ? Math.floor(value) : 0;
 }
-var LOCAL_TOOL_TURN_MAX_OUTPUT_TOKENS = CLIO_MIN_MAX_OUTPUT_TOKENS;
 function resolveReservedOutputTokens(maxOutputTokens) {
   const limit = typeof maxOutputTokens === "number" && Number.isFinite(maxOutputTokens) && maxOutputTokens > 0 ? maxOutputTokens : DEFAULT_MAX_OUTPUT_TOKENS;
   return Math.min(limit, DEFAULT_MAX_OUTPUT_TOKENS);
@@ -2398,7 +2371,7 @@ function remainingContextMaxTokens(model, context, options, limits) {
   const contextWindow = Math.min(configuredContextWindow, loadedContextWindow);
   const budget = Number.isFinite(contextWindow) ? Math.max(1, contextWindow - inputTokens - safety) : Number.POSITIVE_INFINITY;
   const modelLimit = model.maxTokens > 0 ? model.maxTokens : Number.POSITIVE_INFINITY;
-  const defaultLimit = limits?.maxOutputTokens !== void 0 && limits.maxOutputTokens > 0 ? limits.maxOutputTokens : globalDefaultMaxOutputTokens > 0 ? globalDefaultMaxOutputTokens : modelLimit;
+  const defaultLimit = limits?.maxOutputTokens !== void 0 && limits.maxOutputTokens > 0 ? limits.maxOutputTokens : globalDefaultMaxOutputTokens > 0 ? globalDefaultMaxOutputTokens : model.maxTokens > 0 ? modelLimit : DEFAULT_MAX_OUTPUT_TOKENS;
   const requested = options?.maxTokens ?? defaultLimit;
   const resolved = Math.min(requested, modelLimit, budget);
   return Number.isFinite(resolved) ? resolved : DEFAULT_MAX_OUTPUT_TOKENS;
@@ -2436,15 +2409,15 @@ function mergeSamplingOverride(profile, override = runSamplingOverrides()) {
 // src/engine/apis/ollama-native.ts
 var REASONING_CHARS_PER_TOKEN = 4;
 function clioQuirks(model) {
-  return model.clio?.quirks;
+  return model.clioCoder?.quirks;
 }
 function ollamaTargetId(model) {
-  return model.clio?.targetId ?? model.provider;
+  return model.clioCoder?.targetId ?? model.provider;
 }
 async function reconcileOllamaResidency(model, headers) {
   const baseUrl = model.baseUrl;
   if (!baseUrl) return;
-  const metadata2 = model.clio;
+  const metadata2 = model.clioCoder;
   const adapter = {
     targetKey: residencyTargetKey("ollama-native", baseUrl),
     targetId: ollamaTargetId(model),
@@ -2676,7 +2649,7 @@ function runStream(model, context, options, thinkingLevel) {
       let doneReason;
       const sentinelStripper = createSentinelStripper();
       const resolved = resolveModelRuntimeCapabilitiesForModel(model, thinkingLevel);
-      const channelFilter = resolved.family === "gemma-4" ? createGemmaChannelFilter() : null;
+      const channelFilter = usesGemmaChannelMarkers(resolved.modelId) ? createGemmaChannelFilter() : null;
       const closeActiveThinking = () => {
         if (!activeThinking) return;
         stream.push({
@@ -3108,7 +3081,7 @@ async function ensureLlamaCppResidency(input) {
     targetId: input.targetId,
     runtimeId: input.runtimeId,
     keepModelId: input.keepModelId,
-    managed: input.managed ?? residencyManaged(),
+    managed: input.managed ?? true,
     strategy: "router",
     listResident: async () => {
       const models = await fetchRouterModels(input, fetchImpl);
@@ -3188,7 +3161,7 @@ function responseInstanceId(data) {
   return typeof id === "string" && id.trim().length > 0 ? id.trim() : void 0;
 }
 function metadata(model) {
-  return model.clio;
+  return model.clioCoder;
 }
 function targetForModel(model) {
   const info = metadata(model);
@@ -3385,10 +3358,10 @@ async function ensureLmStudioResidency(model, options = {}) {
 var piOpenAICompletions = openAICompletionsApi();
 var REASONING_CHARS_PER_TOKEN2 = 4;
 function chatTemplateKwargsUnsupported(model) {
-  return model.clio?.chatTemplateKwargsUnsupported === true;
+  return model.clioCoder?.chatTemplateKwargsUnsupported === true;
 }
 function clioQuirks2(model) {
-  return model.clio?.quirks;
+  return model.clioCoder?.quirks;
 }
 function pickSamplingProfile2(quirks, thinkingActive) {
   const sampling = quirks?.sampling;
@@ -3419,7 +3392,7 @@ function backendCompletionTimings(value, source) {
   };
 }
 function backendTimingsSourceForModel(model) {
-  const metadata2 = model.clio;
+  const metadata2 = model.clioCoder;
   if (model.provider === "llamacpp" && metadata2?.runtimeId === "llamacpp") return "llamacpp-timings";
   if (model.provider === "lmstudio" && metadata2?.runtimeId === "lmstudio") return "lmstudio-timings";
   return null;
@@ -3597,8 +3570,23 @@ function withStrippedPartial(event) {
   return { ...event, partial: stripThinkingFromMessage(event.partial) };
 }
 function applyThinkingPayload(payload, applied, resolved, model) {
-  if (applied.mechanism === "none") return stripThinkingRequestFields(payload);
-  if (applied.mechanism === "always-on") return payload;
+  if (applied.mechanism === "none") {
+    const next2 = stripThinkingRequestFields(payload);
+    if (resolved.request.chatTemplateKwargs && !chatTemplateKwargsUnsupported(model)) {
+      const existing = isPlainRecord(next2.chat_template_kwargs) ? next2.chat_template_kwargs : {};
+      next2.chat_template_kwargs = { ...existing, ...resolved.request.chatTemplateKwargs };
+    }
+    return next2;
+  }
+  if (applied.mechanism === "always-on") {
+    if (resolved.request.chatTemplateKwargs && !chatTemplateKwargsUnsupported(model)) {
+      const next2 = { ...payload };
+      const existing = isPlainRecord(next2.chat_template_kwargs) ? next2.chat_template_kwargs : {};
+      next2.chat_template_kwargs = { ...existing, ...resolved.request.chatTemplateKwargs };
+      return next2;
+    }
+    return payload;
+  }
   const next = { ...payload };
   if (resolved.request.reasoningEffort && (next.reasoning_effort === void 0 || resolved.response.parser === "harmony")) {
     next.reasoning_effort = resolved.request.reasoningEffort;
@@ -3729,16 +3717,9 @@ function thinkingLevelFromSimple2(options) {
   return reasoning;
 }
 function withRemainingContextBudget(model, context, options) {
-  const metadata2 = runtimeMetadata(model);
-  const localToolOutputLimit = options?.maxTokens === void 0 && (context.tools?.length ?? 0) > 0 && (model.provider === "llamacpp" || metadata2?.runtimeId === "llamacpp") ? LOCAL_TOOL_TURN_MAX_OUTPUT_TOKENS : void 0;
   return {
     ...options ?? {},
-    maxTokens: remainingContextMaxTokens(
-      model,
-      context,
-      options,
-      localToolOutputLimit === void 0 ? void 0 : { maxOutputTokens: localToolOutputLimit }
-    )
+    maxTokens: remainingContextMaxTokens(model, context, options)
   };
 }
 function requiredToolArguments(tool) {
@@ -3752,7 +3733,7 @@ function hasEmptyArguments(args) {
   return Object.keys(args).length === 0;
 }
 function runtimeMetadata(model) {
-  return model.clio;
+  return model.clioCoder;
 }
 function emptyErrorMessage(model, message) {
   return {
@@ -4026,7 +4007,7 @@ var openAICompletionsApiProvider = {
                   (requestModel) => piOpenAICompletions.stream(requestModel, effectiveContext, capturedOptions)
                 )
               ),
-              resolved.family === "gemma-4"
+              usesGemmaChannelMarkers(resolved.modelId)
             ),
             resolved
           ),
@@ -4058,7 +4039,7 @@ var openAICompletionsApiProvider = {
                   (requestModel) => piOpenAICompletions.streamSimple(requestModel, effectiveContext, capturedOptions)
                 )
               ),
-              resolved.family === "gemma-4"
+              usesGemmaChannelMarkers(resolved.modelId)
             ),
             resolved
           ),
@@ -4309,6 +4290,16 @@ function normalizeEntry(raw, file) {
     }
     entry.quirks = candidate.quirks;
   }
+  const canonicalMetadata = candidate["clio-coder"];
+  const legacyMetadata = candidate.clio;
+  if (legacyMetadata !== void 0) warnLegacyNaming("clio: model metadata", "clio-coder: model metadata");
+  const metadata2 = canonicalMetadata ?? legacyMetadata;
+  if (metadata2 !== void 0) {
+    if (typeof metadata2 !== "object" || metadata2 === null || Array.isArray(metadata2)) {
+      throw new Error(`knowledge base file ${file}: entry '${family}' clio-coder metadata must be an object`);
+    }
+    entry.clioCoder = metadata2;
+  }
   return entry;
 }
 
@@ -4554,7 +4545,7 @@ function createProvidersBundle(context) {
     }
     if (probeResult.ok && options?.reasoning !== false && typeof desc.probeReasoning === "function") {
       const settings = readConfig();
-      const orchestratorTarget = settings.orchestrator.target === target.id ? settings.orchestrator.model : null;
+      const orchestratorTarget = settings.chat.target === target.id ? settings.chat.model : null;
       const candidateModelId = orchestratorTarget ?? target.defaultModel ?? null;
       if (candidateModelId) {
         try {
@@ -4632,7 +4623,7 @@ function createProvidersBundle(context) {
       registerClioOAuthProviders();
       registerBuiltinRuntimes(registry);
       const settings = readConfig();
-      setGlobalDefaultMaxOutputTokens(settings.defaults.maxTokens);
+      setGlobalDefaultMaxOutputTokens(settings.chat.maxOutputTokens);
       await loadPluginRuntimes(registry, settings);
       await probeAll();
       const config = context.getContract("config");
@@ -4768,8 +4759,6 @@ export {
   setResidencyNoticeSink,
   setProtectedModelsProvider,
   withRunOverrides,
-  EMPTY_CAPABILITIES,
-  VALID_THINKING_LEVELS,
   applyModelCapabilityPatch,
   probeCapabilitiesForModel,
   resolveModelCapabilities,
@@ -4798,6 +4787,7 @@ export {
   runtimeResolutionWarnings,
   runtimeResolutionWarningsBesideThinkingNotice,
   normalizeCostProvenance,
+  resolveCostProvenance2 as resolveCostProvenance,
   ProvidersDomainModule
 };
-//# sourceMappingURL=chunk-DQITNCXG.js.map
+//# sourceMappingURL=chunk-BYMNWQ7O.js.map

package/dist/{chunk-HI63TFOG.js → chunk-CDNVLKUX.js}

@@ -2,14 +2,14 @@ import { createRequire as __clioCreateRequire } from "node:module"; const requir
 import {
   resolveAuthTarget,
   resolveRuntimeAuthTarget
-} from "./chunk-BBVYXMFO.js";
+} from "./chunk-WNP7O5WZ.js";
 import {
   getRuntimeRegistry
 } from "./chunk-NUGM5KR6.js";
 import {
   catalogProviderForRuntime,
   listCatalogModelsForRuntime
-} from "./chunk-VO2LKSTM.js";
+} from "./chunk-O42A54GG.js";
 import {
   init_esm_shims
 } from "./chunk-3R73A4XB.js";
@@ -170,4 +170,4 @@ export {
   configuredTargetsForRuntime,
   resolveProviderReference
 };
-//# sourceMappingURL=chunk-HI63TFOG.js.map
+//# sourceMappingURL=chunk-CDNVLKUX.js.map