github-router 0.3.312 → 0.3.314

package/dist/{server-setup-DX27fdcf.js → server-setup-CP8nGop3.js}

@@ -1,4 +1,4 @@
-import { $ as UNKNOWN_EFFORT_ANCHOR, An as CHEAP_PROFILE_MODELS, B as resolveAdvisorEffort, Cn as upstreamMaxConnections, D as toolbeltEnabled, Dn as withOneMSuffix, En as oneMContextDisabled, F as ADVISOR_TOOL_INSTRUCTIONS, Fn as withInstallLock, Ft as createMessages, G as repairRejectedThinkingHistory, Gt as createResponses, H as formatThinkingRepairDecline, Ht as assembleResponsesPayload, I as FAST_ADVISOR_TOOL_INSTRUCTIONS, It as getTextTokenCount, J as isControllerClosedError, Jt as readResponseBodyCapped, K as buildAnthropicErrorEvent, Kt as createChatCompletions, L as buildAdvisorStream, Lt as getTokenCount, N as searchWeb, Nt as shimDefaultsToXhigh, On as withOneMSuffixForLead, P as ADVISOR_INTERNAL_TOOL_NAME, Pn as CHEAP_PROFILE_SUBAGENT_CONTEXT_TOKENS, Pt as countTokens, Q as EFFORT_ORDER, R as injectAdvisorTool, Rt as getTokenizerFromModel, Sn as upstreamAllowH2, Tn as catalogAdvertises1M, U as rememberThinkingHistoryRepair, Ut as warnOnTokenPriceDrift, V as resolveAdvisorModel, W as repairKnownThinkingHistory, Wt as resolveMcpToolTimeoutMs, X as readIteratorWithTimeout, Xt as normalizeOpenAIUsage, Y as logStreamError, Yt as parseJsonOrDiagnose, Z as relayAnthropicStream, _n as UPSTREAM_INACTIVITY_TIMEOUT_MS, dn as BUDGET_SMALL_FAST_SLUG, et as bucketEffort, gn as UPSTREAM_FETCH_TIMEOUT_MS, i as assertMcpToolSurfaceConsistent, in as provisionTreeSitterAssets, it as agentToolsEnabled, jn as CHEAP_PROFILE_NATIVE_AGENT_NAMES, kn as CHEAP_PROFILE_ADVISOR_CLIENT_MODEL, ln as toolbeltPathOverride, nt as handleMcpDelete, q as buildOpenAIErrorEvent, qt as MAX_RESPONSE_BODY_BYTES, rt as handleMcpPost, tt as clampEffort, vn as generateRandomPort, wn as classifyMessagesRoute, yn as isBudgetClaudeLead, z as isAdvisorRequested, zt as findLaunchBySecret } from "./peer-mcp-personas-DklYru_1.js";
+import { $ as UNKNOWN_EFFORT_ANCHOR, An as CHEAP_PROFILE_MODELS, B as resolveAdvisorEffort, Cn as upstreamMaxConnections, D as toolbeltEnabled, Dn as withOneMSuffix, En as oneMContextDisabled, F as ADVISOR_TOOL_INSTRUCTIONS, Fn as withInstallLock, Ft as createMessages, G as repairRejectedThinkingHistory, Gt as createResponses, H as formatThinkingRepairDecline, Ht as assembleResponsesPayload, I as FAST_ADVISOR_TOOL_INSTRUCTIONS, It as getTextTokenCount, J as isControllerClosedError, Jt as readResponseBodyCapped, K as buildAnthropicErrorEvent, Kt as createChatCompletions, L as buildAdvisorStream, Lt as getTokenCount, N as searchWeb, Nt as shimDefaultsToXhigh, On as withOneMSuffixForLead, P as ADVISOR_INTERNAL_TOOL_NAME, Pn as CHEAP_PROFILE_SUBAGENT_CONTEXT_TOKENS, Pt as countTokens, Q as EFFORT_ORDER, R as injectAdvisorTool, Rt as getTokenizerFromModel, Sn as upstreamAllowH2, Tn as catalogAdvertises1M, U as rememberThinkingHistoryRepair, Ut as warnOnTokenPriceDrift, V as resolveAdvisorModel, W as repairKnownThinkingHistory, Wt as resolveMcpToolTimeoutMs, X as readIteratorWithTimeout, Xt as normalizeOpenAIUsage, Y as logStreamError, Yt as parseJsonOrDiagnose, Z as relayAnthropicStream, _n as UPSTREAM_INACTIVITY_TIMEOUT_MS, dn as BUDGET_SMALL_FAST_SLUG, et as bucketEffort, gn as UPSTREAM_FETCH_TIMEOUT_MS, i as assertMcpToolSurfaceConsistent, in as provisionTreeSitterAssets, it as agentToolsEnabled, jn as CHEAP_PROFILE_NATIVE_AGENT_NAMES, kn as CHEAP_PROFILE_ADVISOR_CLIENT_MODEL, ln as toolbeltPathOverride, nt as handleMcpDelete, q as buildOpenAIErrorEvent, qt as MAX_RESPONSE_BODY_BYTES, rt as handleMcpPost, tt as clampEffort, vn as generateRandomPort, wn as classifyMessagesRoute, yn as isBudgetClaudeLead, z as isAdvisorRequested, zt as findLaunchBySecret } from "./peer-mcp-personas-0_XU6nm9.js";
 import { t as getPackageVersion } from "./version-C8x2hQrZ.js";
 import { i as ensurePaths, t as PATHS } from "./paths-De1Qh9Q9.js";
 import { t as state } from "./state-B8adscCv.js";
@@ -7,6 +7,7 @@ import { t as getCopilotUsage } from "./get-copilot-usage-Cjy8TiFf.js";
 import { a as resolveExecutable, n as killManagedTree, o as runCommandCapture, r as parseBoolEnv, s as runCommandVoid } from "./exec-y8C_MU8A.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 { 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-BcrMlrmo.js";
+import { d as formatAic, f as nanoAiuToCredits, r as extractAndRecordAic, s as recordAic, u as extractCopilotUsage } from "./aic-ledger-Cx_Ubon3.js";
 import consola from "consola";
 import fs, { existsSync } from "node:fs";
 import path from "node:path";
@@ -1986,7 +1987,7 @@ function formatTokens(n) {
 /**
 * Build a context window summary: "in:1.2K out:50 ctx:1.2K/1M (0.1%)"
 */
-function formatTokenInfo(inputTokens, outputTokens, cacheReadTokens, cacheWriteTokens, model) {
+function formatTokenInfo(inputTokens, outputTokens, cacheReadTokens, cacheWriteTokens, model, aicNano) {
 	if (inputTokens === void 0) return void 0;
 	const parts = [];
 	const maxPrompt = model?.capabilities?.limits?.max_prompt_tokens;
@@ -1996,6 +1997,7 @@ function formatTokenInfo(inputTokens, outputTokens, cacheReadTokens, cacheWriteT
 	} else parts.push(`in:${formatTokens(inputTokens)}`);
 	if (outputTokens !== void 0) parts.push(`out:${formatTokens(outputTokens)}`);
 	if ((cacheReadTokens ?? 0) > 0 || (cacheWriteTokens ?? 0) > 0) parts.push(`cache:r${formatTokens(cacheReadTokens ?? 0)}/w${formatTokens(cacheWriteTokens ?? 0)}`);
+	if ((aicNano ?? 0) > 0) parts.push(`aic:${formatAic(nanoAiuToCredits(aicNano))}`);
 	return parts.join(" ");
 }
 /**
@@ -2031,7 +2033,7 @@ function logRequest(info, model, startTime) {
 	parts.push(`${info.method} ${info.path}`);
 	if (info.resolvedModel && info.resolvedModel !== info.model) parts.push(`${info.model}→${info.resolvedModel}`);
 	else if (info.resolvedModel ?? info.model) parts.push(info.resolvedModel ?? info.model);
-	const tokenInfo = formatTokenInfo(info.inputTokens, info.outputTokens, info.cacheReadTokens, info.cacheWriteTokens, model);
+	const tokenInfo = formatTokenInfo(info.inputTokens, info.outputTokens, info.cacheReadTokens, info.cacheWriteTokens, model, info.aicNano);
 	if (tokenInfo) parts.push(tokenInfo);
 	if (info.bodyBytes !== void 0) parts.push(`body:${formatBytes(info.bodyBytes)}`);
 	if (info.status !== void 0) parts.push(String(info.status));
@@ -2106,7 +2108,7 @@ function collectToolFieldKeys(body) {
 //#endregion
 //#region package.json
 var name = "github-router";
-var version = "0.3.312";
+var version = "0.3.314";
 //#endregion
 //#region src/lib/approval.ts
 const awaitApproval = async () => {
@@ -2801,6 +2803,7 @@ async function handleCompletion$1(c) {
 	const outputTokens = !isStreaming ? response.usage?.completion_tokens : void 0;
 	const rawUsage = !isStreaming ? response.usage : void 0;
 	const responseUsage = rawUsage ? normalizeOpenAIUsage(rawUsage) : void 0;
+	const aicNano = !isStreaming ? extractAndRecordAic(resolvedModel, response) : void 0;
 	logRequest({
 		method: "POST",
 		path: c.req.path,
@@ -2810,6 +2813,7 @@ async function handleCompletion$1(c) {
 		outputTokens,
 		cacheReadTokens: responseUsage?.cacheRead,
 		cacheWriteTokens: responseUsage?.cacheWrite,
+		aicNano,
 		status: 200,
 		streaming: isStreaming
 	}, selectedModel, startTime);
@@ -2823,6 +2827,14 @@ async function handleCompletion$1(c) {
 	let pendingFirstChunk = firstResult.done ? void 0 : firstResult.value;
 	let upstreamFinished = firstResult.done;
 	let consumerCancelled = false;
+	let aicRecorded = false;
+	const tapAic = (chunk) => {
+		if (aicRecorded || !chunk.data || chunk.data === "[DONE]") return;
+		try {
+			const parsed = JSON.parse(chunk.data);
+			if (extractAndRecordAic(resolvedModel, parsed) !== void 0) aicRecorded = true;
+		} catch {}
+	};
 	const safeClose = (controller) => {
 		try {
 			controller.close();
@@ -2854,6 +2866,7 @@ async function handleCompletion$1(c) {
 				const chunk = pendingFirstChunk;
 				pendingFirstChunk = void 0;
 				if (debugEnabled) consola.debug("Streaming chunk:", JSON.stringify(chunk));
+				tapAic(chunk);
 				safeEnqueue(controller, ENCODER$1.encode(formatSSE$1(chunk)));
 				return;
 			}
@@ -2870,6 +2883,7 @@ async function handleCompletion$1(c) {
 				}
 				if (result.value === void 0 || result.value === null) return;
 				if (debugEnabled) consola.debug("Streaming chunk:", JSON.stringify(result.value));
+				tapAic(result.value);
 				safeEnqueue(controller, ENCODER$1.encode(formatSSE$1(result.value)));
 			} catch (error) {
 				upstreamFinished = true;
@@ -3285,6 +3299,63 @@ function sanitizeAnthropicBody(rawBody) {
 	return JSON.stringify(parsed);
 }
 //#endregion
+//#region src/lib/aic-sse-tap.ts
+/**
+* Incremental SSE tap that extracts upstream `copilot_usage` (AIC) from a
+* `/v1/messages` byte stream without modifying it.
+*
+* Verified live: `copilot_usage` rides inside the terminal `message_delta`
+* event beside `usage`. Chunks are arbitrary byte splits, so the tap buffers
+* text, splits on blank-line event boundaries, and keeps the tail. Recording
+* fires at most once per tap instance (one terminal event per stream).
+*
+* Total and never throws — a malformed frame is skipped silently. The relay
+* path additionally wraps the call in try/catch, so a tap bug can never break
+* byte delivery.
+*/
+function createAnthropicAicTap(record) {
+	const decoder = new TextDecoder();
+	let buffer = "";
+	let recorded = false;
+	const handleEvent = (rawEvent) => {
+		if (recorded) return;
+		const lines = rawEvent.split("\n");
+		let eventName = "";
+		const dataLines = [];
+		for (const line of lines) if (line.startsWith("event:")) eventName = line.slice(6).trim();
+		else if (line.startsWith("data:")) dataLines.push(line.slice(5).trimStart());
+		if (eventName !== "" && eventName !== "message_delta") return;
+		if (dataLines.length === 0) return;
+		if (!(dataLines[0] ?? "").startsWith("{")) return;
+		try {
+			const parsed = JSON.parse(dataLines.join("\n"));
+			const usage = extractCopilotUsage(parsed.copilot_usage);
+			if (usage) {
+				recorded = true;
+				try {
+					record(usage);
+				} catch {}
+			}
+		} catch {}
+	};
+	return { onBytes(bytes) {
+		if (recorded || bytes.length === 0) return;
+		buffer += decoder.decode(bytes, { stream: true });
+		let boundary = buffer.indexOf("\n\n");
+		while (boundary >= 0) {
+			const rawEvent = buffer.slice(0, boundary);
+			buffer = buffer.slice(boundary + 2);
+			handleEvent(rawEvent);
+			if (recorded) {
+				buffer = "";
+				return;
+			}
+			boundary = buffer.indexOf("\n\n");
+		}
+		if (buffer.length > 1048576) buffer = "";
+	} };
+}
+//#endregion
 //#region src/lib/anthropic-translate/anthropic-sse.ts
 /**
 * Anthropic Messages SSE egress: frame builders + a Bun-safe ReadableStream
@@ -4215,6 +4286,7 @@ function chatStopReason(finishReason, sawTool) {
 * and each `message.tool_calls[]` becomes a tool_use block.
 */
 function chatResponseToAnthropicMessage(resp, modelId) {
+	extractAndRecordAic(modelId, resp);
 	const choice = resp.choices?.[0];
 	const content = [];
 	let sawTool = false;
@@ -4277,6 +4349,7 @@ async function* synthAnthropicFromChat(upstream, opts) {
 	let usageOut = 0;
 	let usageCacheRead = 0;
 	let usageCacheWrite = 0;
+	let aicRecorded = false;
 	let finishReason = null;
 	let sawDone = false;
 	yield makeMessageStart(messageId, opts.modelId);
@@ -4293,6 +4366,9 @@ async function* synthAnthropicFromChat(upstream, opts) {
 		} catch {
 			continue;
 		}
+		if (!aicRecorded) {
+			if (extractAndRecordAic(opts.modelId, chunk) !== void 0) aicRecorded = true;
+		}
 		if (chunk.usage) {
 			usageIn = Math.max(usageIn, chunk.usage.prompt_tokens ?? 0);
 			usageOut = Math.max(usageOut, chunk.usage.completion_tokens ?? 0);
@@ -4455,6 +4531,7 @@ function parseToolArgs(raw) {
 * else `end_turn`.
 */
 function responsesResponseToAnthropicMessage(resp, modelId) {
+	extractAndRecordAic(modelId, resp);
 	const output = Array.isArray(resp.output) ? resp.output : [];
 	const content = [];
 	let sawToolUse = false;
@@ -4532,6 +4609,7 @@ async function* synthAnthropicFromResponses(upstream, opts) {
 	let sawTool = false;
 	let hitMaxTokens = false;
 	let sawTerminal = false;
+	let aicRecorded = false;
 	const closeCurrent = () => {
 		if (!current) return;
 		q.push(makeContentBlockStop(current.index));
@@ -4701,6 +4779,9 @@ async function* synthAnthropicFromResponses(upstream, opts) {
 			case "response.completed":
 			case "response.incomplete": {
 				sawTerminal = true;
+				if (!aicRecorded) {
+					if (extractAndRecordAic(opts.modelId, ev) !== void 0) aicRecorded = true;
+				}
 				const u = ev.response?.usage;
 				if (u) {
 					usageIn = Math.max(usageIn, u.input_tokens ?? 0);
@@ -5891,7 +5972,10 @@ async function handleCompletion(c) {
 				headers: streamHeaders
 			});
 		}
-		return new Response(response.body ? relayAnthropicStream(response.body, { routePath: c.req.path }) : null, {
+		return new Response(response.body ? relayAnthropicStream(response.body, {
+			routePath: c.req.path,
+			onBytes: createAnthropicAicTap((usage) => recordAic(resolvedModel, usage)).onBytes
+		}) : null, {
 			status: response.status,
 			headers: streamHeaders
 		});
@@ -5900,6 +5984,7 @@ async function handleCompletion(c) {
 	if (!cappedResult.ok) return c.json(cappedResult.errorResponse, cappedResult.status);
 	const responseBody = cappedResult.value;
 	const usage = responseBody.usage;
+	const aicNano = extractAndRecordAic(resolvedModel, responseBody);
 	logRequest({
 		method: "POST",
 		path: c.req.path,
@@ -5909,6 +5994,7 @@ async function handleCompletion(c) {
 		outputTokens: usage?.output_tokens,
 		cacheReadTokens: usage?.cache_read_input_tokens,
 		cacheWriteTokens: usage?.cache_creation_input_tokens,
+		aicNano,
 		status: response.status
 	}, selectedModel, startTime);
 	if (debugEnabled) consola.debug("Non-streaming response from Copilot /v1/messages:", JSON.stringify(responseBody).slice(0, 2e3));
@@ -6470,6 +6556,7 @@ async function handleResponses(c) {
 	const isStreaming = !isNonStreaming(response);
 	const rawUsage = !isStreaming ? response.usage : void 0;
 	const responseUsage = rawUsage && typeof rawUsage === "object" && !Array.isArray(rawUsage) ? normalizeOpenAIUsage(rawUsage) : void 0;
+	const aicNano = !isStreaming ? extractAndRecordAic(resolvedModel, response) : void 0;
 	logRequest({
 		method: "POST",
 		path: c.req.path,
@@ -6479,6 +6566,7 @@ async function handleResponses(c) {
 		outputTokens: responseUsage?.output,
 		cacheReadTokens: responseUsage?.cacheRead,
 		cacheWriteTokens: responseUsage?.cacheWrite,
+		aicNano,
 		status: 200,
 		streaming: isStreaming
 	}, selectedModel, startTime);
@@ -6514,6 +6602,15 @@ async function handleResponses(c) {
 	if (firstChunk === void 0) consola.warn(`Upstream /responses returned no payload events at ${c.req.path}`);
 	let pendingFirstChunk = firstChunk;
 	let consumerCancelled = false;
+	let aicRecorded = false;
+	const tapAic = (chunk) => {
+		if (aicRecorded || !chunk.data || chunk.data === "[DONE]") return;
+		try {
+			const parsed = JSON.parse(chunk.data);
+			if (extractAndRecordAic(resolvedModel, parsed) !== void 0) aicRecorded = true;
+		} catch {}
+	};
+	if (pendingFirstChunk !== void 0) tapAic(pendingFirstChunk);
 	const safeClose = (controller) => {
 		try {
 			controller.close();
@@ -6545,6 +6642,7 @@ async function handleResponses(c) {
 				const chunk = pendingFirstChunk;
 				pendingFirstChunk = void 0;
 				if (debugEnabled) consola.debug("Streaming chunk:", JSON.stringify(chunk));
+				tapAic(chunk);
 				safeEnqueue(controller, ENCODER.encode(formatSSE(chunk)));
 				return;
 			}
@@ -6567,6 +6665,7 @@ async function handleResponses(c) {
 				}
 				if (!result.value.data) return;
 				if (debugEnabled) consola.debug("Streaming chunk:", JSON.stringify(result.value));
+				tapAic(result.value);
 				safeEnqueue(controller, ENCODER.encode(formatSSE(result.value)));
 			} catch (error) {
 				upstreamFinished = true;
@@ -7425,4 +7524,4 @@ function getCodexEnvVars(serverUrl) {
 //#endregion
 export { checkClaudeVersion as C, runSelfUpdate as S, resolveLaunchProfile as _, sharedServerArgs as a, validateFastProfilePrerequisites as b, stopKeepAwake as c, enableFileLogging as d, LUNA_SCOUT_ALIAS_ID as f, profileDescriptor as g, formatFastPrerequisiteFailure as h, setupAndServe as i, injectModelPickerSettingsFile as l, formatCheapPrerequisiteFailure as m, getCodexEnvVars as n, LAUNCH_SECRET_HEADER as o, formatCheap1mPrerequisiteFailure as p, parseSharedArgs as r, startKeepAwake as s, getClaudeCodeEnvVars as t, listModelsForEndpoint as u, validateCheap1mProfilePrerequisites as v, updateClaude as w, validateMaxProfileLaunch as x, validateCheapProfilePrerequisites as y };
 
-//# sourceMappingURL=server-setup-DX27fdcf.js.map
+//# sourceMappingURL=server-setup-CP8nGop3.js.map