github-router 0.3.321 → 0.3.323

package/dist/{server-setup-CXXKmPIE.js → server-setup-CgbHmupe.js}

@@ -1,13 +1,13 @@
-import { $ as EFFORT_ORDER, $t as parseJsonOrDiagnose, An as oneMContextDisabled, B as resolveAdvisorEffort, Bn as CHEAP_PROFILE_MODELS, Bt as getTextTokenCount, Cn as isBudgetClaudeLead, D as toolbeltEnabled, Dn as upstreamMaxConnections, En as upstreamAllowH2, F as ADVISOR_TOOL_INSTRUCTIONS, Fn as CHEAPEST_PROFILE_NATIVE_AGENT_NAMES, G as repairKnownThinkingHistory, Gn as withInstallLock, H as resolveCheapestAdvisorModel, Ht as getTokenizerFromModel, I as FAST_ADVISOR_TOOL_INSTRUCTIONS, J as buildOpenAIErrorEvent, Jt as resolveMcpToolTimeoutMs, K as repairRejectedThinkingHistory, Kt as assembleResponsesPayload, L as buildAdvisorStream, Lt as shimDefaultsToXhigh, Mn as withOneMSuffixForLead, N as searchWeb, Nn as CHEAPEST_PROFILE_ADVISOR_CLIENT_MODEL, On as classifyMessagesRoute, P as ADVISOR_INTERNAL_TOOL_NAME, Pn as CHEAPEST_PROFILE_MODELS, Q as relayAnthropicStream, Qt as readResponseBodyCapped, R as injectAdvisorTool, Rn as CHEAPEST_PROFILE_SUBAGENT_CONTEXT_TOKENS, Rt as countTokens, Sn as generateRandomPort, U as formatThinkingRepairDecline, Ut as findLaunchBySecret, V as resolveAdvisorModel, Vn as CHEAP_PROFILE_NATIVE_AGENT_NAMES, Vt as getTokenCount, W as rememberThinkingHistoryRepair, Wn as CHEAP_PROFILE_SUBAGENT_CONTEXT_TOKENS, X as logStreamError, Xt as createChatCompletions, Y as isControllerClosedError, Yt as createResponses, Z as readIteratorWithTimeout, Zt as MAX_RESPONSE_BODY_BYTES, at as agentToolsEnabled, bn as UPSTREAM_FETCH_TIMEOUT_MS, cn as provisionTreeSitterAssets, en as normalizeOpenAIUsage, et as UNKNOWN_EFFORT_ANCHOR, hn as BUDGET_SMALL_FAST_SLUG, i as assertMcpToolSurfaceConsistent, it as handleMcpPost, jn as withOneMSuffix, kn as catalogAdvertises1M, nt as clampEffort, pn as toolbeltPathOverride, q as buildAnthropicErrorEvent, qt as warnOnTokenPriceDrift, rt as handleMcpDelete, tt as bucketEffort, xn as UPSTREAM_INACTIVITY_TIMEOUT_MS, z as isAdvisorRequested, zn as CHEAP_PROFILE_ADVISOR_CLIENT_MODEL, zt as createMessages } from "./peer-mcp-personas-BerVEkZ9.js";
+import { $ as EFFORT_ORDER, $n as withInstallLock, $t as MAX_RESPONSE_BODY_BYTES, An as upstreamMaxConnections, B as resolveAdvisorEffort, Bn as BALANCED_PROFILE_SUBAGENT_CONTEXT_TOKENS, Bt as countTokens, Cn as UPSTREAM_FETCH_TIMEOUT_MS, D as toolbeltEnabled, En as isBudgetClaudeLead, F as ADVISOR_TOOL_INSTRUCTIONS, Fn as withOneMSuffixForLead, G as repairKnownThinkingHistory, Gt as findLaunchBySecret, H as resolveCheapestAdvisorModel, Hn as CHEAPEST_PROFILE_MODELS, Ht as getTextTokenCount, I as FAST_ADVISOR_TOOL_INSTRUCTIONS, In as BALANCED_PROFILE_MODELS, J as buildOpenAIErrorEvent, Jn as CHEAP_PROFILE_MODELS, Jt as assembleResponsesPayload, K as repairRejectedThinkingHistory, Kn as CHEAPEST_PROFILE_SUBAGENT_CONTEXT_TOKENS, L as buildAdvisorStream, Ln as BALANCED_PROFILE_NATIVE_AGENT_NAMES, Mn as catalogAdvertises1M, N as searchWeb, Nn as oneMContextDisabled, P as ADVISOR_INTERNAL_TOOL_NAME, Pn as withOneMSuffix, Q as relayAnthropicStream, Qn as CHEAP_PROFILE_SUBAGENT_CONTEXT_TOKENS, Qt as createChatCompletions, R as injectAdvisorTool, Tn as generateRandomPort, U as formatThinkingRepairDecline, Un as CHEAPEST_PROFILE_NATIVE_AGENT_NAMES, Ut as getTokenCount, V as resolveAdvisorModel, Vn as CHEAPEST_PROFILE_ADVISOR_CLIENT_MODEL, Vt as createMessages, W as rememberThinkingHistoryRepair, Wt as getTokenizerFromModel, X as logStreamError, Xt as resolveMcpToolTimeoutMs, Y as isControllerClosedError, Yn as CHEAP_PROFILE_NATIVE_AGENT_NAMES, Yt as warnOnTokenPriceDrift, Z as readIteratorWithTimeout, Zt as createResponses, at as agentToolsEnabled, dn as provisionTreeSitterAssets, en as readResponseBodyCapped, et as UNKNOWN_EFFORT_ANCHOR, gn as toolbeltPathOverride, i as assertMcpToolSurfaceConsistent, it as handleMcpPost, jn as classifyMessagesRoute, kn as upstreamAllowH2, nn as normalizeOpenAIUsage, nt as clampEffort, q as buildAnthropicErrorEvent, qn as CHEAP_PROFILE_ADVISOR_CLIENT_MODEL, rt as handleMcpDelete, tn as parseJsonOrDiagnose, tt as bucketEffort, vn as BUDGET_SMALL_FAST_SLUG, wn as UPSTREAM_INACTIVITY_TIMEOUT_MS, z as isAdvisorRequested, zt as shimDefaultsToXhigh } from "./peer-mcp-personas-DcAErb_d.js";
 import { t as getPackageVersion } from "./version-C8x2hQrZ.js";
-import { i as ensurePaths, t as PATHS } from "./paths-CHBAj_t9.js";
+import { i as ensurePaths, t as PATHS } from "./paths-BnZwolac.js";
 import { t as state } from "./state-BlMpEFAt.js";
-import { T as copilotHeaders, _ as forwardError, a as cacheCopilotVersion, c as filterBetaHeader, d as resolveModel, f as sleep, g as HTTPError, i as tryRefreshAndRetry, l as isNullish, n as setupGitHubAgentToken, o as cacheModels, r as setupGitHubToken, s as cacheVSCodeVersion, t as setupCopilotToken, w as copilotBaseUrl, y as fetchWithTransientRetry } from "./token-86uk6y4P.js";
-import { t as getCopilotUsage } from "./get-copilot-usage-BWCAzmSc.js";
+import { T as copilotHeaders, _ as forwardError, a as cacheCopilotVersion, c as filterBetaHeader, d as resolveModel, f as sleep, g as HTTPError, i as tryRefreshAndRetry, l as isNullish, n as setupGitHubAgentToken, o as cacheModels, r as setupGitHubToken, s as cacheVSCodeVersion, t as setupCopilotToken, w as copilotBaseUrl, y as fetchWithTransientRetry } from "./token-Css-ARGL.js";
+import { t as getCopilotUsage } from "./get-copilot-usage-Bs-FEPMf.js";
 import { c as runCommandVoid, i as parseBoolEnv, o as resolveExecutable, r as killManagedTree, s as runCommandCapture } from "./exec-COUT4QsS.js";
-import { c as FAST_PROFILE_MODELS, l as FAST_PROFILE_NATIVE_AGENT_NAMES, r as FAST_PROFILE_ADVISOR_MODEL } from "./fast-profile-contract-_DhIWank.js";
+import { c as FAST_PROFILE_MODELS, l as FAST_PROFILE_NATIVE_AGENT_NAMES, r as FAST_PROFILE_ADVISOR_MODEL } from "./fast-profile-contract-DmsyQQrw.js";
 import { A as normalizeTrailingOneMSuffix, a as MAX_PROFILE_NATIVE_AGENT_NAMES, d as maxAdvisorPinIsValid, h as maxOpusModel, i as MAX_PROFILE_MODELS, j as stripTrailingOneMSuffix, k as fastEndpointForModel, l as maxAdvisorEffortForModel, n as MAX_PROFILE_ALLOWED_LEAD_MODEL_IDS, t as MAX_PROFILE_ADVISOR_INSTRUCTIONS, u as maxAdvisorModelFromPin, y as validateMaxProfilePrerequisites } from "./max-profile-contract-Bl7I_EOC.js";
-import { a as extractAndRecordPricedAic, f as extractCopilotUsage, i as extractAndRecordAic, l as recordAic, m as nanoAiuToCredits, p as formatAic } from "./aic-ledger-BrDqsuge.js";
+import { a as extractAndRecordPricedAic, f as copilotUsageToWire, h as nanoAiuToCredits, i as extractAndRecordAic, l as recordAic, m as formatAic, p as extractCopilotUsage } from "./aic-ledger-DRU89gJg.js";
 import consola from "consola";
 import fs, { existsSync } from "node:fs";
 import path from "node:path";
@@ -362,8 +362,8 @@ const STANDARD_PROFILE = Object.freeze({
 	hasCoordinator: true
 });
 /**
-* The `-m fast` roster: exactly five native agents (`Explore`, `Plan`,
-* `general-purpose`, `implementer`, `reviewer`), the fast-only `oracle` peer
+* The `-m fast` roster: exactly four native agents (`Explore`, `Plan`,
+* `General-Purpose`, `Reviewer`), the fast-only `oracle` peer
 * tool, no coordinator, and `peers`/`search` plus optional browser and
 * browse-only worker groups. Core workers, `orchestrate`, `decide`, `fleet`,
 * and `first-mate` remain hard denies even when their standard gates pass.
@@ -440,6 +440,25 @@ const CHEAPEST_PROFILE = Object.freeze({
 	hasCoordinator: false
 });
 /**
+* The `-m balanced` roster: the most-complex-tasks tier. A Sol/high lead at
+* the 200K default window, the same four-agent surface as cheap
+* (`Explore`/`Plan`/`General-Purpose`/`Reviewer`, every role at 200K), and
+* the Grok-4.6/medium Oracle-only peer set. Hard-denies match fast's: core
+* workers, `orchestrate`, `decide`, `fleet`, and `first-mate`.
+*/
+const BALANCED_PROFILE = Object.freeze({
+	id: "balanced",
+	nativeRoster: new Set(BALANCED_PROFILE_NATIVE_AGENT_NAMES),
+	personaAllowlist: /* @__PURE__ */ new Set(["oracle"]),
+	allowedGroups: /* @__PURE__ */ new Set([
+		"peers",
+		"search",
+		"workers",
+		"browser"
+	]),
+	hasCoordinator: false
+});
+/**
 * The `-m max` profile: Sol/Luna-led, browse-only workers, and explicit
 * cross-lab peer names. The descriptor is a hard projection for bound launch
 * requests; unbound/BYO traffic remains standard because it has no registry
@@ -475,6 +494,7 @@ function profileDescriptor(id) {
 	if (id === "cheap1m") return CHEAP1M_PROFILE;
 	if (id === "cheap") return CHEAP_PROFILE;
 	if (id === "cheapest") return CHEAPEST_PROFILE;
+	if (id === "balanced") return BALANCED_PROFILE;
 	return STANDARD_PROFILE;
 }
 /**
@@ -495,6 +515,7 @@ function resolveLaunchProfile(modelArg) {
 	if (arg === "cheap1m") return "cheap1m";
 	if (arg === "cheap") return "cheap";
 	if (arg === "cheapest") return "cheapest";
+	if (arg === "balanced") return "balanced";
 	return "standard";
 }
 function validateMaxProfileLaunch(catalog) {
@@ -543,6 +564,17 @@ const CHEAPEST_PLAN_ALIAS_ID = "gh-router-cheapest-plan-high";
 const CHEAPEST_GENERAL_PURPOSE_ALIAS_ID = "gh-router-cheapest-general-purpose-xhigh";
 const CHEAPEST_IMPLEMENTER_ALIAS_ID = "gh-router-cheapest-implementer-max";
 const CHEAPEST_REVIEWER_ALIAS_ID = "gh-router-cheapest-reviewer-high";
+/**
+* Balanced-profile subagent aliases (`-m balanced`). Same non-catalog
+* mechanism as the cheap aliases above, with the balanced roster's own
+* identities and efforts (`BALANCED_PROFILE_NATIVE_MODELS` /
+* `BALANCED_PROFILE_NATIVE_EFFORTS`, pinned by a drift test). Emitted BARE by
+* `buildBalancedProfileAgentDefinitions`.
+*/
+const BALANCED_EXPLORE_ALIAS_ID = "gh-router-balanced-explore-high";
+const BALANCED_PLAN_ALIAS_ID = "gh-router-balanced-plan-high";
+const BALANCED_GENERAL_PURPOSE_ALIAS_ID = "gh-router-balanced-general-purpose-high";
+const BALANCED_REVIEWER_ALIAS_ID = "gh-router-balanced-reviewer-max";
 const LUNA_SONNET_ALIAS_ID = "gh-router-luna-sonnet-xhigh";
 /**
 * Router-owned alias id for the fast profile's Haiku-tier row
@@ -624,13 +656,13 @@ const MODEL_ALIAS_TABLE = /* @__PURE__ */ new Map([
 	}],
 	[CHEAP_GENERAL_PURPOSE_ALIAS_ID, {
 		aliasId: CHEAP_GENERAL_PURPOSE_ALIAS_ID,
-		realModel: CHEAP_PROFILE_MODELS["general-purpose"],
+		realModel: CHEAP_PROFILE_MODELS["General-Purpose"],
 		absentEffortDefault: "max"
 	}],
 	[CHEAP_IMPLEMENTER_ALIAS_ID, {
 		aliasId: CHEAP_IMPLEMENTER_ALIAS_ID,
-		realModel: CHEAP_PROFILE_MODELS.implementer,
-		absentEffortDefault: "high"
+		realModel: CHEAP_PROFILE_MODELS["General-Purpose"],
+		absentEffortDefault: "max"
 	}],
 	[CHEAP_REVIEWER_ALIAS_ID, {
 		aliasId: CHEAP_REVIEWER_ALIAS_ID,
@@ -649,18 +681,38 @@ const MODEL_ALIAS_TABLE = /* @__PURE__ */ new Map([
 	}],
 	[CHEAPEST_GENERAL_PURPOSE_ALIAS_ID, {
 		aliasId: CHEAPEST_GENERAL_PURPOSE_ALIAS_ID,
-		realModel: CHEAPEST_PROFILE_MODELS["general-purpose"],
-		absentEffortDefault: "xhigh"
+		realModel: CHEAPEST_PROFILE_MODELS["General-Purpose"],
+		absentEffortDefault: "max"
 	}],
 	[CHEAPEST_IMPLEMENTER_ALIAS_ID, {
 		aliasId: CHEAPEST_IMPLEMENTER_ALIAS_ID,
-		realModel: CHEAPEST_PROFILE_MODELS.implementer,
+		realModel: CHEAPEST_PROFILE_MODELS["General-Purpose"],
 		absentEffortDefault: "max"
 	}],
 	[CHEAPEST_REVIEWER_ALIAS_ID, {
 		aliasId: CHEAPEST_REVIEWER_ALIAS_ID,
 		realModel: CHEAPEST_PROFILE_MODELS.reviewer,
 		absentEffortDefault: "high"
+	}],
+	[BALANCED_EXPLORE_ALIAS_ID, {
+		aliasId: BALANCED_EXPLORE_ALIAS_ID,
+		realModel: BALANCED_PROFILE_MODELS.explore,
+		absentEffortDefault: "high"
+	}],
+	[BALANCED_PLAN_ALIAS_ID, {
+		aliasId: BALANCED_PLAN_ALIAS_ID,
+		realModel: BALANCED_PROFILE_MODELS.plan,
+		absentEffortDefault: "high"
+	}],
+	[BALANCED_GENERAL_PURPOSE_ALIAS_ID, {
+		aliasId: BALANCED_GENERAL_PURPOSE_ALIAS_ID,
+		realModel: BALANCED_PROFILE_MODELS["General-Purpose"],
+		absentEffortDefault: "high"
+	}],
+	[BALANCED_REVIEWER_ALIAS_ID, {
+		aliasId: BALANCED_REVIEWER_ALIAS_ID,
+		realModel: BALANCED_PROFILE_MODELS.reviewer,
+		absentEffortDefault: "max"
 	}]
 ]);
 /**
@@ -721,10 +773,10 @@ function hasUsablePromptMetadata(model) {
 * than silently substituting or dropping an agent.
 *
 * Checks, per the fast-launch-profile design:
-*   - Luna lead/Explore/general-purpose: tool calls, >=1M, high+max, Responses.
+*   - Luna Explore: tool calls, >=1M, high+max, Responses.
 *   - Sol Plan: tool calls, >=1M, high, Responses.
-*   - Grok reviewer: tool calls, medium, Responses, usable prompt metadata.
-*   - Gemini implementer/Advisor: tool calls, >=1M, high, chat-completions.
+*   - Sonnet reviewer: tool calls, >=1M, xhigh, Messages, prompt metadata.
+*   - Gemini lead/General-Purpose: tool calls, >=1M, high, chat-completions.
 *   - Opus Oracle: exact Opus 5, >=1M, adaptive/high, Messages, prompt metadata.
 *
 * Pure over the passed-in catalog snapshot so it's unit-testable without
@@ -758,13 +810,13 @@ function validateFastProfilePrerequisites(catalog) {
 		if (!hasUsablePromptMetadata(sonnet)) missing.push(`${FAST_PROFILE_MODELS.reviewer}: no usable max_prompt_tokens metadata`);
 		if (!supportsEndpoint(sonnet, "messages")) missing.push(`${FAST_PROFILE_MODELS.reviewer}: does not advertise a supported Messages endpoint`);
 	}
-	const gemini = findModel(catalog, FAST_PROFILE_MODELS.implementer);
-	if (!gemini) missing.push(`${FAST_PROFILE_MODELS.implementer}: absent from the live catalog`);
+	const gemini = findModel(catalog, FAST_PROFILE_MODELS.gemini);
+	if (!gemini) missing.push(`${FAST_PROFILE_MODELS.gemini}: absent from the live catalog`);
 	else {
-		if (!hasToolCalls(gemini)) missing.push(`${FAST_PROFILE_MODELS.implementer}: does not advertise tool_calls`);
-		if (!hasContextAtLeast(gemini, FAST_REQUIRED_CONTEXT_TOKENS)) missing.push(`${FAST_PROFILE_MODELS.implementer}: advertised context window is below 1M`);
-		if (!supportsEffort(gemini, "high")) missing.push(`${FAST_PROFILE_MODELS.implementer}: does not advertise a "high" reasoning effort`);
-		if (!supportsEndpoint(gemini, "chat")) missing.push(`${FAST_PROFILE_MODELS.implementer}: does not advertise a supported chat-completions endpoint`);
+		if (!hasToolCalls(gemini)) missing.push(`${FAST_PROFILE_MODELS.gemini}: does not advertise tool_calls`);
+		if (!hasContextAtLeast(gemini, FAST_REQUIRED_CONTEXT_TOKENS)) missing.push(`${FAST_PROFILE_MODELS.gemini}: advertised context window is below 1M`);
+		if (!supportsEffort(gemini, "high")) missing.push(`${FAST_PROFILE_MODELS.gemini}: does not advertise a "high" reasoning effort`);
+		if (!supportsEndpoint(gemini, "chat")) missing.push(`${FAST_PROFILE_MODELS.gemini}: does not advertise a supported chat-completions endpoint`);
 	}
 	const opus = findModel(catalog, FAST_PROFILE_MODELS.oracle);
 	if (!opus) missing.push(`${FAST_PROFILE_MODELS.oracle}: absent from the live catalog`);
@@ -795,7 +847,7 @@ const CHEAP_SUBAGENT_MIN_CONTEXT_TOKENS = CHEAP_PROFILE_SUBAGENT_CONTEXT_TOKENS;
 /**
 * Shared cheap-family roster prerequisite check, parameterized only by the
 * LEAD's context gate. Both `-m cheap` (200K lead floor) and `-m cheap1m`
-* (1M lead window) validate the EXACT same five-agent roster: the lead gate
+* (1M lead window) validate the EXACT same four-agent roster: the lead gate
 * is the only requirement that differs between the two cheap siblings, and
 * every subagent runs at the 200K default window either way. A non-1M
 * luna/sol/sonnet/grok is acceptable as long as the roster models advertise
@@ -932,6 +984,60 @@ function validateCheapestProfilePrerequisites(catalog) {
 function formatCheapestPrerequisiteFailure(missing) {
 	return "github-router claude -m cheapest requires the following live-catalog capabilities, which this account's catalog does not fully provide:\n" + missing.map((m) => `  - ${m}`).join("\n") + "\n\nFalling back or silently dropping an agent is not supported for the cheapest profile's exact roster. Run plain `github-router claude` instead.";
 }
+/** The subagent window floor: every balanced role runs at the 200K default. */
+const BALANCED_SUBAGENT_MIN_CONTEXT_TOKENS = BALANCED_PROFILE_SUBAGENT_CONTEXT_TOKENS;
+/**
+* Validate the live Copilot catalog for `-m balanced`: Sol lead at the 200K
+* default window, Luna Explore/Reviewer roles, Sol Plan, Gemini
+* General-Purpose, and a Grok Oracle — all at the 200K default with their
+* fixed efforts and supported endpoints.
+*/
+function validateBalancedProfilePrerequisites(catalog) {
+	const missing = [];
+	const sol = findModel(catalog, BALANCED_PROFILE_MODELS.lead);
+	if (!sol) missing.push(`${BALANCED_PROFILE_MODELS.lead}: absent from the live catalog`);
+	else {
+		if (!hasToolCalls(sol)) missing.push(`${BALANCED_PROFILE_MODELS.lead}: does not advertise tool_calls`);
+		if (!hasContextAtLeast(sol, BALANCED_SUBAGENT_MIN_CONTEXT_TOKENS)) missing.push(`${BALANCED_PROFILE_MODELS.lead}: advertised context window is below the 200K subagent floor`);
+		if (!supportsEffort(sol, "high")) missing.push(`${BALANCED_PROFILE_MODELS.lead}: does not advertise a "high" reasoning effort`);
+		if (!supportsEndpoint(sol, "responses")) missing.push(`${BALANCED_PROFILE_MODELS.lead}: does not advertise a supported Responses endpoint`);
+	}
+	const luna = findModel(catalog, BALANCED_PROFILE_MODELS.explore);
+	if (!luna) missing.push(`${BALANCED_PROFILE_MODELS.explore}: absent from the live catalog`);
+	else {
+		if (!hasToolCalls(luna)) missing.push(`${BALANCED_PROFILE_MODELS.explore}: does not advertise tool_calls`);
+		if (!hasContextAtLeast(luna, BALANCED_SUBAGENT_MIN_CONTEXT_TOKENS)) missing.push(`${BALANCED_PROFILE_MODELS.explore}: advertised context window is below the 200K subagent floor`);
+		if (!supportsEffort(luna, "high") || !supportsEffort(luna, "max")) missing.push(`${BALANCED_PROFILE_MODELS.explore}: does not advertise both "high" and "max" reasoning effort`);
+		if (!supportsEndpoint(luna, "responses")) missing.push(`${BALANCED_PROFILE_MODELS.explore}: does not advertise a supported Responses endpoint`);
+	}
+	const gemini = findModel(catalog, BALANCED_PROFILE_MODELS["General-Purpose"]);
+	if (!gemini) missing.push(`${BALANCED_PROFILE_MODELS["General-Purpose"]}: absent from the live catalog`);
+	else {
+		if (!hasToolCalls(gemini)) missing.push(`${BALANCED_PROFILE_MODELS["General-Purpose"]}: does not advertise tool_calls`);
+		if (!hasContextAtLeast(gemini, BALANCED_SUBAGENT_MIN_CONTEXT_TOKENS)) missing.push(`${BALANCED_PROFILE_MODELS["General-Purpose"]}: advertised context window is below the 200K subagent floor`);
+		if (!supportsEffort(gemini, "high")) missing.push(`${BALANCED_PROFILE_MODELS["General-Purpose"]}: does not advertise a "high" reasoning effort`);
+		if (!supportsEndpoint(gemini, "chat")) missing.push(`${BALANCED_PROFILE_MODELS["General-Purpose"]}: does not advertise a supported chat-completions endpoint`);
+	}
+	const grok = findModel(catalog, BALANCED_PROFILE_MODELS.oracle);
+	if (!grok) missing.push(`${BALANCED_PROFILE_MODELS.oracle}: absent from the live catalog`);
+	else {
+		if (!hasContextAtLeast(grok, BALANCED_SUBAGENT_MIN_CONTEXT_TOKENS)) missing.push(`${BALANCED_PROFILE_MODELS.oracle}: advertised context window is below the 200K subagent floor`);
+		if (!supportsEffort(grok, "medium")) missing.push(`${BALANCED_PROFILE_MODELS.oracle}: does not advertise a "medium" reasoning effort`);
+		if (!hasUsablePromptMetadata(grok)) missing.push(`${BALANCED_PROFILE_MODELS.oracle}: no usable max_prompt_tokens metadata`);
+		if (!supportsEndpoint(grok, "responses")) missing.push(`${BALANCED_PROFILE_MODELS.oracle}: does not advertise a supported Responses endpoint`);
+	}
+	return {
+		ok: missing.length === 0,
+		missing
+	};
+}
+/**
+* Format `validateBalancedProfilePrerequisites`'s failure list into the
+* launch error message.
+*/
+function formatBalancedPrerequisiteFailure(missing) {
+	return "github-router claude -m balanced requires the following live-catalog capabilities, which this account's catalog does not fully provide:\n" + missing.map((m) => `  - ${m}`).join("\n") + "\n\nFalling back or silently dropping an agent is not supported for the balanced profile's exact roster. Run plain `github-router claude` instead.";
+}
 //#endregion
 //#region src/lib/file-log-reporter.ts
 const MAX_LOG_BYTES = 1048576;
@@ -1316,6 +1422,32 @@ const CHEAPEST_PICKER_MODELS = Object.freeze([
 		neverOneM: true
 	}
 ]);
+const BALANCED_PICKER_MODELS = Object.freeze([
+	{
+		id: "gpt-5.6-sol",
+		label: "GPT-5.6 Sol",
+		behavesAs: "claude-opus-5",
+		neverOneM: true
+	},
+	{
+		id: "gpt-5.6-luna",
+		label: "GPT-5.6 Luna",
+		behavesAs: "claude-opus-5",
+		neverOneM: true
+	},
+	{
+		id: "gemini-3.8-flash",
+		label: "Gemini 3.8 Flash",
+		behavesAs: "claude-sonnet-5",
+		neverOneM: true
+	},
+	{
+		id: "grok-4.6",
+		label: "Grok 4.6",
+		behavesAs: "claude-sonnet-5",
+		neverOneM: true
+	}
+]);
 /**
 * Return the ordered, profile-specific `/model` rows supported by the live
 * Copilot catalog. Fast intentionally shares Standard's four-row inventory;
@@ -1333,7 +1465,7 @@ function selectableModelsInCatalog(profile) {
 	const catalog = state.models?.data;
 	if (!catalog || catalog.length === 0) return [];
 	const present = new Set(catalog.map((entry) => entry.id));
-	return (profile === "max" ? MAX_PICKER_MODELS : profile === "cheap" || profile === "cheap1m" ? CHEAP_PICKER_MODELS : profile === "cheapest" ? CHEAPEST_PICKER_MODELS : STANDARD_PICKER_MODELS).filter((entry) => present.has(entry.id)).map((entry) => ({
+	return (profile === "max" ? MAX_PICKER_MODELS : profile === "cheap" || profile === "cheap1m" ? CHEAP_PICKER_MODELS : profile === "cheapest" ? CHEAPEST_PICKER_MODELS : profile === "balanced" ? BALANCED_PICKER_MODELS : STANDARD_PICKER_MODELS).filter((entry) => present.has(entry.id)).map((entry) => ({
 		model: entry.neverOneM ? entry.id : withOneMSuffix(entry.id),
 		label: entry.label,
 		behavesAs: entry.behavesAs
@@ -1371,7 +1503,7 @@ async function renameSettingsWithRetry(temp, target) {
 * 200K session's `/model` UI.
 */
 function routerWinsPicker(profile) {
-	return profile === "fast" || profile === "cheap" || profile === "cheap1m" || profile === "cheapest";
+	return profile === "fast" || profile === "cheap" || profile === "cheap1m" || profile === "cheapest" || profile === "balanced";
 }
 function modelIdsFromExistingPicker(value) {
 	if (!value || typeof value !== "object" || Array.isArray(value)) return [];
@@ -2291,7 +2423,7 @@ function collectToolFieldKeys(body) {
 //#endregion
 //#region package.json
 var name = "github-router";
-var version = "0.3.321";
+var version = "0.3.323";
 //#endregion
 //#region src/lib/approval.ts
 const awaitApproval = async () => {
@@ -3656,7 +3788,7 @@ function makeContentBlockStop(index) {
 		index
 	};
 }
-function makeMessageDelta(stopReason, stopSequence, usage) {
+function makeMessageDelta(stopReason, stopSequence, usage, copilotUsage) {
 	return {
 		type: "message_delta",
 		delta: {
@@ -3668,7 +3800,8 @@ function makeMessageDelta(stopReason, stopSequence, usage) {
 			output_tokens: usage.output_tokens ?? 0,
 			cache_read_input_tokens: usage.cache_read_input_tokens ?? 0,
 			cache_creation_input_tokens: usage.cache_creation_input_tokens ?? 0
-		}
+		},
+		...copilotUsage ? { copilot_usage: copilotUsage } : {}
 	};
 }
 function makeMessageStop() {
@@ -4623,6 +4756,7 @@ async function* synthAnthropicFromChat(upstream, opts) {
 	let usageCacheRead = 0;
 	let usageCacheWrite = 0;
 	let aicRecorded = false;
+	let aicCopilotUsageWire;
 	let finishReason = null;
 	let sawDone = false;
 	yield makeMessageStart(messageId, opts.modelId);
@@ -4640,7 +4774,12 @@ async function* synthAnthropicFromChat(upstream, opts) {
 			continue;
 		}
 		if (!aicRecorded) {
-			if (extractAndRecordPricedAic(opts.modelId, chunk) !== void 0) aicRecorded = true;
+			const pricedUsage = extractCopilotUsage(chunk.copilot_usage);
+			if (pricedUsage && pricedUsage.totalNanoAiu > 0) {
+				recordAic(opts.modelId, pricedUsage);
+				aicRecorded = true;
+				aicCopilotUsageWire = copilotUsageToWire(pricedUsage);
+			}
 		}
 		if (chunk.usage) {
 			usageIn = Math.max(usageIn, chunk.usage.prompt_tokens ?? 0);
@@ -4726,7 +4865,7 @@ async function* synthAnthropicFromChat(upstream, opts) {
 		output_tokens: usage.output,
 		cache_read_input_tokens: usage.cacheRead,
 		cache_creation_input_tokens: usage.cacheWrite
-	});
+	}, aicCopilotUsageWire);
 	yield makeMessageStop();
 }
 //#endregion
@@ -4882,7 +5021,7 @@ async function* synthAnthropicFromResponses(upstream, opts) {
 	let sawTool = false;
 	let hitMaxTokens = false;
 	let sawTerminal = false;
-	let aicRecorded = false;
+	let aicCopilotUsageWire = void 0;
 	const closeCurrent = () => {
 		if (!current) return;
 		q.push(makeContentBlockStop(current.index));
@@ -5052,8 +5191,12 @@ async function* synthAnthropicFromResponses(upstream, opts) {
 			case "response.completed":
 			case "response.incomplete": {
 				sawTerminal = true;
-				if (!aicRecorded) {
-					if (extractAndRecordPricedAic(opts.modelId, ev) !== void 0) aicRecorded = true;
+				if (aicCopilotUsageWire === void 0) {
+					const usage = extractCopilotUsage(ev.copilot_usage);
+					if (usage && usage.totalNanoAiu > 0) {
+						recordAic(opts.modelId, usage);
+						aicCopilotUsageWire = copilotUsageToWire(usage);
+					}
 				}
 				const u = ev.response?.usage;
 				if (u) {
@@ -5087,7 +5230,7 @@ async function* synthAnthropicFromResponses(upstream, opts) {
 		output_tokens: usage.output,
 		cache_read_input_tokens: usage.cacheRead,
 		cache_creation_input_tokens: usage.cacheWrite
-	}));
+	}, aicCopilotUsageWire));
 	q.push(makeMessageStop());
 	for (const e of q) yield e;
 	q.length = 0;
@@ -5432,7 +5575,7 @@ function preprocessMaxRequest(rawBody, launch, subagentRequest = false) {
 Object.freeze([MAX_LUNA_HIGH_ALIAS_ID, MAX_LUNA_MAX_ALIAS_ID]);
 //#endregion
 //#region src/lib/fast-request-preprocess.ts
-const cheapFamily = (profileId) => profileId === "cheap" || profileId === "cheap1m" || profileId === "cheapest";
+const cheapFamily = (profileId) => profileId === "cheap" || profileId === "cheap1m" || profileId === "cheapest" || profileId === "balanced";
 const pinnedProfile = (profileId) => profileId === "fast" || cheapFamily(profileId);
 /**
 * Profiles whose every selectable `/model` row must stay at Claude Code's
@@ -5440,24 +5583,25 @@ const pinnedProfile = (profileId) => profileId === "fast" || cheapFamily(profile
 * `fast` keeps 1M on lead and subagents by design; `cheap1m` keeps 1M on
 * the lead only (its subagents are bare-by-construction).
 */
-const bareEnforcedProfile = (profileId) => profileId === "cheap" || profileId === "cheapest";
+const bareEnforcedProfile = (profileId) => profileId === "cheap" || profileId === "cheapest" || profileId === "balanced";
 function explicitEffortOf(value) {
 	return typeof value === "string" && value.trim() !== "" ? value : void 0;
 }
 /**
-* Apply authenticated fast/cheap-profile model and effort policy before
+* Apply authenticated fast/cheap/balanced-profile model and effort policy before
 * ordinary model resolution. Synthetic aliases are refused outside an
 * authenticated pinned launch, so raw/BYO traffic cannot opt itself into
-* private profile semantics. The cheap family (including cheapest) shares
-* fast's effort mapping (same model-to-effort rows, just bare subagent
-* slugs at the wiring layer), so this preprocess is shared.
+* private profile semantics. The cheap family (including cheapest and
+* balanced) shares fast's effort mapping (same model-to-effort rows, just
+* bare subagent slugs at the wiring layer), so this preprocess is shared.
 * Note the reviewer differs: fast reviews on Sonnet 5/xhigh while cheap
-* reviews on Luna/max and cheapest reviews on Gemini/high — all rows exist
-* here, so each profile's reviewer resolves to its fixed effort.
+* reviews on Luna/max, cheapest reviews on Gemini/high, and balanced reviews
+* on Luna/max — all rows exist here, so each profile's reviewer resolves to
+* its fixed effort.
 *
-* Pinned-profile isolation (fast/cheap/cheap1m/cheapest): only
+* Pinned-profile isolation (fast/cheap/cheap1m/cheapest/balanced): only
 * router-provided models are accepted (anything else is `rejectedModel`).
-* Context: cheap/cheapest traffic is stripped to the bare 200K id;
+* Context: cheap/cheapest/balanced traffic is stripped to the bare 200K id;
 * cheap1m subagent traffic is stripped while its lead keeps 1M; fast
 * keeps 1M on both. Effort: picker changes apply to the lead session
 * only — lead requests keep an explicit `output_config.effort` when
@@ -5994,21 +6138,25 @@ async function handleCompletion(c) {
 	const maxProfileRequest = identity.launch?.profileId === "max";
 	const cheapProfileRequest = identity.launch?.profileId === "cheap" || identity.launch?.profileId === "cheap1m";
 	const cheapestProfileRequest = identity.launch?.profileId === "cheapest";
+	const balancedProfileRequest = identity.launch?.profileId === "balanced";
 	const subagentRequest = Boolean(c.req.header("x-claude-code-agent-id"));
 	const fastSubagentRequest = fastProfileRequest && subagentRequest;
 	const maxSubagentRequest = maxProfileRequest && subagentRequest;
 	const cheapSubagentRequest = cheapProfileRequest && subagentRequest;
 	const cheapestSubagentRequest = cheapestProfileRequest && subagentRequest;
+	const balancedSubagentRequest = balancedProfileRequest && subagentRequest;
 	const fastLeadAdvisor = fastProfileRequest && !fastSubagentRequest;
 	const maxLeadAdvisor = maxProfileRequest && !maxSubagentRequest;
 	const cheapLeadAdvisor = cheapProfileRequest && !cheapSubagentRequest;
 	const cheapestLeadAdvisor = cheapestProfileRequest && !cheapestSubagentRequest;
-	const advisorEnabled = advisorRequested && !fastSubagentRequest && !maxSubagentRequest && !cheapSubagentRequest && !cheapestSubagentRequest;
+	const balancedLeadAdvisor = balancedProfileRequest && !balancedSubagentRequest;
+	const advisorEnabled = advisorRequested && !fastSubagentRequest && !maxSubagentRequest && !cheapSubagentRequest && !cheapestSubagentRequest && !balancedSubagentRequest;
 	const fastAdvisorEnabled = fastLeadAdvisor && advisorRequested && hasNonEmptyTools(rawBody);
 	const maxAdvisorEnabled = maxLeadAdvisor && advisorRequested && hasNonEmptyTools(rawBody);
 	const cheapAdvisorEnabled = cheapLeadAdvisor && advisorRequested && hasNonEmptyTools(rawBody);
 	const cheapestAdvisorEnabled = cheapestLeadAdvisor && advisorRequested && hasNonEmptyTools(rawBody);
-	const advisorBehaviorEnabled = fastLeadAdvisor ? fastAdvisorEnabled : cheapLeadAdvisor ? cheapAdvisorEnabled : cheapestLeadAdvisor ? cheapestAdvisorEnabled : maxLeadAdvisor ? maxAdvisorEnabled : advisorEnabled;
+	const balancedAdvisorEnabled = balancedLeadAdvisor && advisorRequested && hasNonEmptyTools(rawBody);
+	const advisorBehaviorEnabled = fastLeadAdvisor ? fastAdvisorEnabled : cheapLeadAdvisor ? cheapAdvisorEnabled : cheapestLeadAdvisor ? cheapestAdvisorEnabled : balancedLeadAdvisor ? balancedAdvisorEnabled : maxLeadAdvisor ? maxAdvisorEnabled : advisorEnabled;
 	let fastAdvisorChoice;
 	let maxAdvisorChoice;
 	if (maxAdvisorEnabled) {
@@ -6047,9 +6195,9 @@ async function handleCompletion(c) {
 			}, 503);
 		}
 	}
-	if (cheapAdvisorEnabled) {
+	if (cheapAdvisorEnabled || balancedAdvisorEnabled) {
 		const mismatch = fastAdvisorMetadataMismatch(rawBody);
-		const relaunchProfile = identity.launch?.profileId === "cheap1m" ? "cheap1m" : "cheap";
+		const relaunchProfile = identity.launch?.profileId === "cheap1m" ? "cheap1m" : identity.launch?.profileId === "balanced" ? "balanced" : "cheap";
 		if (mismatch) return c.json({
 			type: "error",
 			error: {
@@ -6140,7 +6288,7 @@ async function handleCompletion(c) {
 			parsedBase = JSON.parse(promptWindowBody);
 		} catch {}
 		const wantsStream = parsedBase?.stream === true;
-		if ((fastAdvisorEnabled || cheapAdvisorEnabled || cheapestAdvisorEnabled || maxAdvisorEnabled) && wantsStream) {
+		if ((fastAdvisorEnabled || cheapAdvisorEnabled || cheapestAdvisorEnabled || balancedAdvisorEnabled || maxAdvisorEnabled) && wantsStream) {
 			const initialConversation = Array.isArray(parsedBase.messages) ? parsedBase.messages : [];
 			const parsedInitial = parseAnthropicRequest(parsedBase, modelId, selectedModel);
 			const translatedAdvisorAborter = new AbortController();
@@ -6163,7 +6311,7 @@ async function handleCompletion(c) {
 				effort: maxAdvisorChoice.effort,
 				escalated: false,
 				fastProfile: false
-			} : cheapAdvisorEnabled || cheapestAdvisorEnabled ? {
+			} : cheapAdvisorEnabled || cheapestAdvisorEnabled || balancedAdvisorEnabled ? {
 				model: fastAdvisorChoice.model,
 				effort: resolveAdvisorEffort(rawBody, fastAdvisorChoice.model, true),
 				escalated: fastAdvisorChoice.escalated,
@@ -6183,7 +6331,7 @@ async function handleCompletion(c) {
 				advisorEscalated: advisorChoice.escalated,
 				advisorFastProfile: advisorChoice.fastProfile,
 				advisorMaxProfile: maxAdvisorEnabled,
-				advisorCheapProfile: cheapAdvisorEnabled || cheapestAdvisorEnabled,
+				advisorCheapProfile: cheapAdvisorEnabled || cheapestAdvisorEnabled || balancedAdvisorEnabled,
 				advisorEffort: advisorChoice.effort,
 				externalAborter: translatedAdvisorAborter,
 				continueTurn: makeShimContinueTurn(endpoint, {
@@ -6306,10 +6454,10 @@ async function handleCompletion(c) {
 				requestHeaders,
 				advisorModel: advisorChoice.model,
 				advisorEscalated: advisorChoice.escalated,
-				advisorFastProfile: fastLeadAdvisor || cheapLeadAdvisor || cheapestLeadAdvisor,
+				advisorFastProfile: fastLeadAdvisor || cheapLeadAdvisor || cheapestLeadAdvisor || balancedLeadAdvisor,
 				advisorMaxProfile: maxAdvisorEnabled,
-				advisorCheapProfile: cheapLeadAdvisor || cheapestLeadAdvisor,
-				advisorEffort: maxAdvisorChoice ? maxAdvisorChoice.effort : resolveAdvisorEffort(rawBody, advisorChoice.model, fastLeadAdvisor || cheapLeadAdvisor || cheapestLeadAdvisor),
+				advisorCheapProfile: cheapLeadAdvisor || cheapestLeadAdvisor || balancedLeadAdvisor,
+				advisorEffort: maxAdvisorChoice ? maxAdvisorChoice.effort : resolveAdvisorEffort(rawBody, advisorChoice.model, fastLeadAdvisor || cheapLeadAdvisor || cheapestLeadAdvisor || balancedLeadAdvisor),
 				externalAborter: advisorAborter
 			}), {
 				status: response.status,
@@ -7899,6 +8047,6 @@ function getCodexEnvVars(serverUrl) {
 	return vars;
 }
 //#endregion
-export { validateCheapProfilePrerequisites as A, formatCheap1mPrerequisiteFailure as C, profileDescriptor as D, formatFastPrerequisiteFailure as E, checkClaudeVersion as F, updateClaude as I, validateFastProfilePrerequisites as M, validateMaxProfileLaunch as N, resolveLaunchProfile as O, runSelfUpdate as P, LUNA_SCOUT_ALIAS_ID as S, formatCheapestPrerequisiteFailure as T, CHEAP_EXPLORE_ALIAS_ID as _, sharedServerArgs as a, CHEAP_PLAN_ALIAS_ID as b, stopKeepAwake as c, enableFileLogging as d, CHEAPEST_EXPLORE_ALIAS_ID as f, CHEAPEST_REVIEWER_ALIAS_ID as g, CHEAPEST_PLAN_ALIAS_ID as h, setupAndServe as i, validateCheapestProfilePrerequisites as j, validateCheap1mProfilePrerequisites as k, injectModelPickerSettingsFile as l, CHEAPEST_IMPLEMENTER_ALIAS_ID as m, getCodexEnvVars as n, LAUNCH_SECRET_HEADER as o, CHEAPEST_GENERAL_PURPOSE_ALIAS_ID as p, parseSharedArgs as r, startKeepAwake as s, getClaudeCodeEnvVars as t, listModelsForEndpoint as u, CHEAP_GENERAL_PURPOSE_ALIAS_ID as v, formatCheapPrerequisiteFailure as w, CHEAP_REVIEWER_ALIAS_ID as x, CHEAP_IMPLEMENTER_ALIAS_ID as y };
+export { profileDescriptor as A, updateClaude as B, CHEAP_REVIEWER_ALIAS_ID as C, formatCheapPrerequisiteFailure as D, formatCheap1mPrerequisiteFailure as E, validateCheapestProfilePrerequisites as F, validateFastProfilePrerequisites as I, validateMaxProfileLaunch as L, validateBalancedProfilePrerequisites as M, validateCheap1mProfilePrerequisites as N, formatCheapestPrerequisiteFailure as O, validateCheapProfilePrerequisites as P, runSelfUpdate as R, CHEAP_PLAN_ALIAS_ID as S, formatBalancedPrerequisiteFailure as T, CHEAPEST_GENERAL_PURPOSE_ALIAS_ID as _, sharedServerArgs as a, CHEAP_EXPLORE_ALIAS_ID as b, stopKeepAwake as c, enableFileLogging as d, BALANCED_EXPLORE_ALIAS_ID as f, CHEAPEST_EXPLORE_ALIAS_ID as g, BALANCED_REVIEWER_ALIAS_ID as h, setupAndServe as i, resolveLaunchProfile as j, formatFastPrerequisiteFailure as k, injectModelPickerSettingsFile as l, BALANCED_PLAN_ALIAS_ID as m, getCodexEnvVars as n, LAUNCH_SECRET_HEADER as o, BALANCED_GENERAL_PURPOSE_ALIAS_ID as p, parseSharedArgs as r, startKeepAwake as s, getClaudeCodeEnvVars as t, listModelsForEndpoint as u, CHEAPEST_PLAN_ALIAS_ID as v, LUNA_SCOUT_ALIAS_ID as w, CHEAP_GENERAL_PURPOSE_ALIAS_ID as x, CHEAPEST_REVIEWER_ALIAS_ID as y, checkClaudeVersion as z };
 
-//# sourceMappingURL=server-setup-CXXKmPIE.js.map
+//# sourceMappingURL=server-setup-CgbHmupe.js.map