ghc-proxy 0.5.6 → 0.5.8

package/dist/main.mjs

@@ -2,9 +2,9 @@
 import { createRequire } from "node:module";
 import process$1 from "node:process";
 import { formatWithOptions, parseArgs, promisify } from "node:util";
-import path, { sep } from "node:path";
+import path, { join, sep } from "node:path";
 import * as tty from "node:tty";
-import fs from "node:fs/promises";
+import fs, { mkdir, readdir, unlink, writeFile } from "node:fs/promises";
 import os from "node:os";
 import { randomUUID } from "node:crypto";
 import { execFile, execFileSync } from "node:child_process";
@@ -6340,6 +6340,33 @@ async function cacheVSCodeVersion() {
 	consola.debug(`Using VSCode version: ${response}`);
 }
 
+//#endregion
+//#region src/lib/retry.ts
+/**
+* Retry an async operation with exponential backoff.
+* Delay schedule: baseDelayMs * 2^attempt (0-indexed), e.g. 5s, 10s, 20s, 40s.
+*/
+async function retryWithBackoff(fn, options) {
+	const maxRetries = options?.maxRetries ?? 4;
+	const baseDelayMs = options?.baseDelayMs ?? 5e3;
+	const shouldRetry = options?.shouldRetry ?? (() => true);
+	const onRetry = options?.onRetry;
+	let lastError;
+	for (let attempt = 0; attempt <= maxRetries; attempt++) try {
+		return await fn();
+	} catch (error) {
+		lastError = error;
+		if (!shouldRetry(error) || attempt >= maxRetries) throw error;
+		const delay = baseDelayMs * 2 ** attempt;
+		onRetry?.(error, attempt, delay);
+		await sleep(delay);
+	}
+	throw lastError;
+}
+function formatErrorMessage(error) {
+	return error instanceof Error ? error.message : String(error);
+}
+
 //#endregion
 //#region src/lib/token.ts
 const TRAILING_SLASHES_RE = /\/+$/;
@@ -6354,20 +6381,28 @@ async function setupCopilotToken() {
 	consola.debug("GitHub Copilot Token fetched successfully!");
 	if (state.config.showToken) consola.info("Copilot token:", response.token);
 	const refreshInterval = (response.refresh_in - 60) * 1e3;
-	const refreshCopilotToken = async () => {
-		consola.debug("Refreshing Copilot token");
-		try {
-			const refreshed = await githubClient.getCopilotToken();
-			applyCopilotTokenState(refreshed);
-			consola.debug("Copilot token refreshed");
-			if (state.config.showToken) consola.info("Refreshed Copilot token:", refreshed.token);
-		} catch (error) {
-			consola.error("Failed to refresh Copilot token:", error);
-		}
+	const scheduleRefresh = () => {
+		setTimeout(() => {
+			refreshCopilotToken(githubClient).then(scheduleRefresh);
+		}, refreshInterval);
 	};
-	setInterval(() => {
-		refreshCopilotToken();
-	}, refreshInterval);
+	scheduleRefresh();
+}
+async function refreshCopilotToken(githubClient) {
+	consola.debug("Refreshing Copilot token");
+	try {
+		const refreshed = await retryWithBackoff(() => githubClient.getCopilotToken(), {
+			shouldRetry: (error) => !(error instanceof HTTPError) || isTransientHttpError(error),
+			onRetry: (error, attempt, delayMs) => {
+				consola.warn(`Token refresh failed (attempt ${attempt + 1}), retrying in ${delayMs / 1e3}s:`, formatErrorMessage(error));
+			}
+		});
+		applyCopilotTokenState(refreshed);
+		consola.debug("Copilot token refreshed");
+		if (state.config.showToken) consola.info("Refreshed Copilot token:", refreshed.token);
+	} catch (error) {
+		consola.error("Failed to refresh Copilot token:", error);
+	}
 }
 async function setupGitHubToken(options) {
 	try {
@@ -6416,6 +6451,17 @@ async function setupGitHubToken(options) {
 function isAuthError(error) {
 	return error instanceof HTTPError && (error.status === 401 || error.status === 403);
 }
+const TRANSIENT_HTTP_STATUSES = new Set([
+	408,
+	429,
+	500,
+	502,
+	503,
+	504
+]);
+function isTransientHttpError(error) {
+	return TRANSIENT_HTTP_STATUSES.has(error.status);
+}
 async function logUser() {
 	const user = await createGitHubClient().getGitHubUser();
 	state.cache.githubLogin = user.login;
@@ -6533,7 +6579,7 @@ const checkUsage = defineCommand({
 
 //#endregion
 //#region src/lib/version.ts
-const VERSION = "0.5.6";
+const VERSION = "0.5.8";
 
 //#endregion
 //#region src/debug.ts
@@ -18492,7 +18538,7 @@ var require_snapshot_utils = /* @__PURE__ */ __commonJSMin$1(((exports, module)
 //#endregion
 //#region node_modules/undici/lib/mock/snapshot-recorder.js
 var require_snapshot_recorder = /* @__PURE__ */ __commonJSMin$1(((exports, module) => {
-	const { writeFile, readFile, mkdir } = __require$1("node:fs/promises");
+	const { writeFile: writeFile$1, readFile, mkdir: mkdir$1 } = __require$1("node:fs/promises");
 	const { dirname, resolve } = __require$1("node:path");
 	const { setTimeout: setTimeout$1, clearTimeout: clearTimeout$1 } = __require$1("node:timers");
 	const { InvalidArgumentError, UndiciError } = require_errors();
@@ -18779,12 +18825,12 @@ var require_snapshot_recorder = /* @__PURE__ */ __commonJSMin$1(((exports, modul
 			const path = filePath || this.#snapshotPath;
 			if (!path) throw new InvalidArgumentError("Snapshot path is required");
 			const resolvedPath = resolve(path);
-			await mkdir(dirname(resolvedPath), { recursive: true });
+			await mkdir$1(dirname(resolvedPath), { recursive: true });
 			const data = Array.from(this.#snapshots.entries()).map(([hash, snapshot]) => ({
 				hash,
 				snapshot
 			}));
-			await writeFile(resolvedPath, JSON.stringify(data, null, 2), { flush: true });
+			await writeFile$1(resolvedPath, JSON.stringify(data, null, 2), { flush: true });
 		}
 		/**
 		* Clears all recorded snapshots
@@ -46879,20 +46925,29 @@ const methodColors = {
 function colorizeMethod(method) {
 	return colorize(methodColors[method] ?? "white", method);
 }
+function getEffectiveModel(info) {
+	return info.steps.length > 0 ? info.steps.at(-1).result : info.originalModel ?? "-";
+}
+function appendModelStep(info, tag, newModel) {
+	if (newModel === getEffectiveModel(info)) return info;
+	return {
+		originalModel: info.originalModel,
+		steps: [...info.steps, {
+			tag,
+			result: newModel
+		}]
+	};
+}
 function formatModelMapping(info) {
 	if (!info) return "";
-	const { originalModel, rewrittenModel, mappedModel } = info;
-	if (!originalModel && !rewrittenModel && !mappedModel) return "";
-	const parts = [];
-	const displayOriginal = originalModel ?? "-";
-	parts.push(colorize("blueBright", displayOriginal));
-	if (rewrittenModel && rewrittenModel !== displayOriginal) {
-		parts.push(colorize("dim", "~>"));
-		parts.push(colorize("cyanBright", rewrittenModel));
-	}
-	if (mappedModel && mappedModel !== (rewrittenModel ?? displayOriginal)) {
-		parts.push(colorize("dim", "→"));
-		parts.push(colorize("greenBright", mappedModel));
+	const { originalModel, steps } = info;
+	if (!originalModel && steps.length === 0) return "";
+	const parts = [colorize("blueBright", originalModel ?? "-")];
+	for (let i = 0; i < steps.length; i++) {
+		const step = steps[i];
+		const isLast = i === steps.length - 1;
+		parts.push(colorize("dim", `-[${step.tag}]->`));
+		parts.push(colorize(isLast ? "greenBright" : "cyanBright", step.result));
 	}
 	return ` ${colorize("dim", "model=")}${parts.join(" ")}`;
 }
@@ -47584,8 +47639,11 @@ function normalizeSystemBlocks(system) {
 	}];
 }
 function normalizeToolResultContent(block) {
-	if (typeof block.content === "string") return [textBlock(block.content)];
-	return block.content.map((contentBlock) => {
+	return normalizeToolResultContentValue(block.content);
+}
+function normalizeToolResultContentValue(content) {
+	if (typeof content === "string") return [textBlock(content)];
+	return content.map((contentBlock) => {
 		switch (contentBlock.type) {
 			case "text": return textBlock(contentBlock.text);
 			case "image": return imageBlock(contentBlock.source.media_type, contentBlock.source.data);
@@ -47593,6 +47651,15 @@ function normalizeToolResultContent(block) {
 		}
 	});
 }
+function normalizeMcpToolResultContent(block) {
+	return normalizeToolResultContentValue(block.content);
+}
+function normalizeServerToolResultContent(block) {
+	return [textBlock(typeof block.content === "string" ? block.content : JSON.stringify(block.content) ?? "")];
+}
+function describeDocumentBlock(block) {
+	return `[document attachment omitted: ${typeof block.source.type === "string" ? block.source.type : "unknown"}]`;
+}
 function normalizeMessage(message) {
 	if (typeof message.content === "string") return {
 		role: message.role,
@@ -47607,7 +47674,13 @@ function normalizeMessage(message) {
 				thinking: block.thinking,
 				signature: block.signature
 			};
-			case "tool_use": return {
+			case "redacted_thinking": return {
+				kind: "redacted_thinking",
+				data: block.data
+			};
+			case "tool_use":
+			case "server_tool_use":
+			case "mcp_tool_use": return {
 				kind: "tool_use",
 				id: block.id,
 				name: block.name,
@@ -47619,6 +47692,25 @@ function normalizeMessage(message) {
 				content: normalizeToolResultContent(block),
 				isError: block.is_error
 			};
+			case "mcp_tool_result": return {
+				kind: "tool_result",
+				toolUseId: block.tool_use_id,
+				content: normalizeMcpToolResultContent(block),
+				isError: block.is_error
+			};
+			case "server_tool_result":
+			case "web_search_tool_result":
+			case "web_fetch_tool_result":
+			case "code_execution_tool_result":
+			case "bash_code_execution_tool_result":
+			case "text_editor_code_execution_tool_result":
+			case "tool_search_tool_result": return {
+				kind: "tool_result",
+				toolUseId: block.tool_use_id,
+				content: normalizeServerToolResultContent(block),
+				isError: block.is_error
+			};
+			case "document": return textBlock(describeDocumentBlock(block));
 			default: return assertNever(block);
 		}
 	});
@@ -47983,6 +48075,10 @@ function mapBlocks(response) {
 				thinking: block.thinking,
 				signature: block.signature
 			};
+			case "redacted_thinking": return {
+				type: "redacted_thinking",
+				data: block.data
+			};
 			case "tool_use": return {
 				type: "tool_use",
 				id: block.id,
@@ -48157,6 +48253,10 @@ function toConversationBlock(block) {
 			text: block.thinking,
 			signature: block.signature
 		};
+		case "redacted_thinking": return {
+			kind: "redacted_thinking",
+			data: block.data
+		};
 		case "tool_use": return {
 			kind: "tool_use",
 			id: block.id,
@@ -48205,6 +48305,9 @@ function recordAnthropicRequestIssues(request, context) {
 		for (const block of turn.blocks) if (block.kind === "thinking") {
 			sawThinking = true;
 			if (sawToolUse) sawTextOrThinkingAfterTool = true;
+		} else if (block.kind === "redacted_thinking") {
+			sawThinking = true;
+			if (sawToolUse) sawTextOrThinkingAfterTool = true;
 		} else if (block.kind === "text") {
 			if (sawToolUse) sawTextOrThinkingAfterTool = true;
 		} else if (block.kind === "tool_use") sawToolUse = true;
@@ -48569,13 +48672,15 @@ function rewriteModel(modelId) {
 	if (userRules) {
 		for (const rule of userRules) if (matchesGlob(rule.from, modelId)) return {
 			originalModel: modelId,
-			model: normalizeToKnownModel(rule.to) ?? rule.to
+			model: normalizeToKnownModel(rule.to) ?? rule.to,
+			reason: "CONFIG_REWRITE"
 		};
 	}
 	const normalized = normalizeToKnownModel(modelId);
 	if (normalized && normalized !== modelId) return {
 		originalModel: modelId,
-		model: normalized
+		model: normalized,
+		reason: "AUTO_CORRECT"
 	};
 	return {
 		originalModel: modelId,
@@ -48874,12 +48979,30 @@ function estimateContentBlockChars(blocks) {
 		case "thinking":
 			chars += block.thinking.length;
 			break;
+		case "redacted_thinking":
+			chars += block.data.length;
+			break;
 		case "tool_use":
+		case "server_tool_use":
+		case "mcp_tool_use":
 			chars += JSON.stringify(block.input).length;
 			break;
 		case "tool_result":
+		case "mcp_tool_result":
 			chars += typeof block.content === "string" ? block.content.length : JSON.stringify(block.content ?? "").length;
 			break;
+		case "server_tool_result":
+		case "web_search_tool_result":
+		case "web_fetch_tool_result":
+		case "code_execution_tool_result":
+		case "bash_code_execution_tool_result":
+		case "text_editor_code_execution_tool_result":
+		case "tool_search_tool_result":
+			chars += JSON.stringify(block.content ?? "").length;
+			break;
+		case "document":
+			chars += JSON.stringify(block).length;
+			break;
 		case "image":
 			chars += 1e3;
 			break;
@@ -48970,12 +49093,29 @@ const anthropicThinkingBlockSchema = object({
 	thinking: string(),
 	signature: string().optional()
 }).loose();
+const anthropicRedactedThinkingBlockSchema = object({
+	type: literal("redacted_thinking"),
+	data: string()
+}).loose();
 const anthropicToolUseBlockSchema = object({
 	type: literal("tool_use"),
 	id: string().min(1),
 	name: string().min(1),
 	input: jsonObjectSchema
 }).loose();
+const anthropicServerToolUseBlockSchema = object({
+	type: literal("server_tool_use"),
+	id: string().min(1),
+	name: string().min(1),
+	input: jsonObjectSchema
+}).loose();
+const anthropicMcpToolUseBlockSchema = object({
+	type: literal("mcp_tool_use"),
+	id: string().min(1),
+	name: string().min(1),
+	input: jsonObjectSchema,
+	server_name: string().min(1)
+}).loose();
 const anthropicToolResultContentBlockSchema = union([anthropicTextBlockSchema, anthropicImageBlockSchema]);
 const anthropicToolResultBlockSchema = object({
 	type: literal("tool_result"),
@@ -48983,12 +49123,39 @@ const anthropicToolResultBlockSchema = object({
 	content: union([string(), array(anthropicToolResultContentBlockSchema)]),
 	is_error: boolean().optional()
 }).loose();
+const anthropicMcpToolResultBlockSchema = object({
+	type: literal("mcp_tool_result"),
+	tool_use_id: string().min(1),
+	content: union([string(), array(anthropicToolResultContentBlockSchema)]),
+	is_error: boolean().optional()
+}).loose();
+const anthropicServerToolResultBlockSchema = object({
+	type: _enum([
+		"server_tool_result",
+		"web_search_tool_result",
+		"web_fetch_tool_result",
+		"code_execution_tool_result",
+		"bash_code_execution_tool_result",
+		"text_editor_code_execution_tool_result",
+		"tool_search_tool_result"
+	]),
+	tool_use_id: string().min(1),
+	content: unknown(),
+	is_error: boolean().optional()
+}).loose();
+const anthropicDocumentBlockSchema = object({
+	type: literal("document"),
+	source: jsonObjectSchema
+}).loose();
 const anthropicUserMessageSchema = object({
 	role: literal("user"),
 	content: union([string(), array(union([
 		anthropicTextBlockSchema,
 		anthropicImageBlockSchema,
-		anthropicToolResultBlockSchema
+		anthropicToolResultBlockSchema,
+		anthropicMcpToolResultBlockSchema,
+		anthropicServerToolResultBlockSchema,
+		anthropicDocumentBlockSchema
 	]))])
 }).loose();
 const anthropicAssistantMessageSchema = object({
@@ -48996,7 +49163,12 @@ const anthropicAssistantMessageSchema = object({
 	content: union([string(), array(union([
 		anthropicTextBlockSchema,
 		anthropicThinkingBlockSchema,
-		anthropicToolUseBlockSchema
+		anthropicRedactedThinkingBlockSchema,
+		anthropicToolUseBlockSchema,
+		anthropicServerToolUseBlockSchema,
+		anthropicMcpToolUseBlockSchema,
+		anthropicMcpToolResultBlockSchema,
+		anthropicServerToolResultBlockSchema
 	]))])
 }).loose();
 const anthropicMessageSchema = union([anthropicUserMessageSchema, anthropicAssistantMessageSchema]);
@@ -49279,8 +49451,13 @@ const responsesReasoningSummarySchema = object({
 const responsesReasoningInputSchema = object({
 	id: string().optional(),
 	type: literal("reasoning"),
-	summary: array(responsesReasoningSummarySchema),
-	encrypted_content: string().min(1)
+	summary: array(responsesReasoningSummarySchema).optional(),
+	encrypted_content: string().nullable().optional(),
+	status: _enum([
+		"in_progress",
+		"completed",
+		"incomplete"
+	]).optional()
 }).loose();
 const responsesCompactionInputSchema = object({
 	id: string().min(1),
@@ -49513,7 +49690,11 @@ async function handleCompletionCore({ body, signal, headers }) {
 	const requestContext = normalizeChatRequestContext(payload, headers);
 	consola.debug("Request payload:", JSON.stringify(payload).slice(-400));
 	const rewrite = applyModelRewrite(payload);
-	const originalModel = rewrite.originalModel;
+	const steps = [];
+	if (rewrite.reason) steps.push({
+		tag: rewrite.reason,
+		result: rewrite.model
+	});
 	const selectedModel = findModelById(payload.model);
 	try {
 		if (selectedModel) {
@@ -49532,11 +49713,10 @@ async function handleCompletionCore({ body, signal, headers }) {
 	}
 	const upstreamSignal = createUpstreamSignalFromConfig(signal);
 	const plan = adapter.toCapiPlan(payload, { requestContext });
-	const modelMapping = {
-		originalModel,
-		rewrittenModel: rewrite.model,
-		mappedModel: plan.resolvedModel
-	};
+	const modelMapping = appendModelStep({
+		originalModel: rewrite.originalModel,
+		steps
+	}, "MODEL_RESOLVE", plan.resolvedModel);
 	const transport = new CopilotTransport(createCopilotClient());
 	consola.debug("Streaming response");
 	return {
@@ -49847,11 +50027,16 @@ function translateUserMessage(message) {
 	const items = [];
 	const pendingContent = [];
 	for (const block of message.content) {
-		if (block.type === "tool_result") {
+		if (block.type === "tool_result" || block.type === "mcp_tool_result") {
 			flushPendingContent(pendingContent, items, { role: "user" });
 			items.push(createFunctionCallOutput(block));
 			continue;
 		}
+		if (isServerToolResultBlock(block)) {
+			flushPendingContent(pendingContent, items, { role: "user" });
+			items.push(createServerFunctionCallOutput(block));
+			continue;
+		}
 		const converted = translateUserContentBlock(block);
 		if (converted) pendingContent.push(converted);
 	}
@@ -49865,7 +50050,7 @@ function translateAssistantMessage(message) {
 	const items = [];
 	const pendingContent = [];
 	for (const block of message.content) {
-		if (block.type === "tool_use") {
+		if (block.type === "tool_use" || block.type === "server_tool_use" || block.type === "mcp_tool_use") {
 			flushPendingContent(pendingContent, items, {
 				role: "assistant",
 				phase: assistantPhase
@@ -49873,6 +50058,30 @@ function translateAssistantMessage(message) {
 			items.push(createFunctionToolCall(block));
 			continue;
 		}
+		if (block.type === "redacted_thinking") {
+			flushPendingContent(pendingContent, items, {
+				role: "assistant",
+				phase: assistantPhase
+			});
+			items.push(createRedactedReasoningContent(block));
+			continue;
+		}
+		if (block.type === "mcp_tool_result") {
+			flushPendingContent(pendingContent, items, {
+				role: "assistant",
+				phase: assistantPhase
+			});
+			items.push(createFunctionCallOutput(block));
+			continue;
+		}
+		if (isServerToolResultBlock(block)) {
+			flushPendingContent(pendingContent, items, {
+				role: "assistant",
+				phase: assistantPhase
+			});
+			items.push(createServerFunctionCallOutput(block));
+			continue;
+		}
 		if (block.type === "thinking" && block.signature) {
 			const compaction = createCompactionContent(block);
 			if (compaction) {
@@ -49908,6 +50117,7 @@ function translateUserContentBlock(block) {
 	switch (block.type) {
 		case "text": return createTextContent(block.text);
 		case "image": return createImageContent(block);
+		case "document": return createDocumentContent(block);
 		default: return;
 	}
 }
@@ -49937,7 +50147,7 @@ function resolveAssistantPhase(content) {
 	let hasToolUse = false;
 	for (const block of content) {
 		if (block.type === "text") hasText = true;
-		else if (block.type === "tool_use") hasToolUse = true;
+		else if (block.type === "tool_use" || block.type === "server_tool_use" || block.type === "mcp_tool_use") hasToolUse = true;
 		if (hasText && hasToolUse) break;
 	}
 	if (!hasText) return;
@@ -49962,6 +50172,23 @@ function createImageContent(block) {
 		detail: "auto"
 	};
 }
+function createDocumentContent(block) {
+	const source = block.source;
+	if (source.type === "file" && typeof source.file_id === "string") return {
+		type: "input_file",
+		file_id: source.file_id
+	};
+	if (source.type === "url" && typeof source.url === "string") return {
+		type: "input_file",
+		file_url: source.url
+	};
+	if (source.type === "base64" && typeof source.media_type === "string" && typeof source.data === "string") return {
+		type: "input_file",
+		file_data: `data:${source.media_type};base64,${source.data}`
+	};
+	if (source.type === "text" && typeof source.data === "string") return createTextContent(source.data);
+	return createTextContent("[document attachment omitted]");
+}
 function createReasoningContent(block) {
 	const { encryptedContent, id } = SignatureCodec.decodeReasoning(block.signature ?? "");
 	const thinking = block.thinking === THINKING_TEXT ? "" : block.thinking;
@@ -49975,6 +50202,13 @@ function createReasoningContent(block) {
 		encrypted_content: encryptedContent
 	};
 }
+function createRedactedReasoningContent(block) {
+	return {
+		type: "reasoning",
+		summary: [],
+		encrypted_content: block.data
+	};
+}
 function createCompactionContent(block) {
 	const compaction = decodeCompactionCarrierSignature(block.signature ?? "");
 	if (!compaction) return;
@@ -50001,6 +50235,17 @@ function createFunctionCallOutput(block) {
 		status: block.is_error ? "incomplete" : "completed"
 	};
 }
+function createServerFunctionCallOutput(block) {
+	return {
+		type: "function_call_output",
+		call_id: block.tool_use_id,
+		output: typeof block.content === "string" ? block.content : JSON.stringify(block.content) ?? "",
+		status: block.is_error ? "incomplete" : "completed"
+	};
+}
+function isServerToolResultBlock(block) {
+	return block.type === "server_tool_result" || block.type === "web_search_tool_result" || block.type === "web_fetch_tool_result" || block.type === "code_execution_tool_result" || block.type === "bash_code_execution_tool_result" || block.type === "text_editor_code_execution_tool_result" || block.type === "tool_search_tool_result";
+}
 function translateSystemPrompt(system) {
 	if (!system) return null;
 	if (typeof system === "string") return system;
@@ -50913,10 +51158,6 @@ const responsesApiEntry = {
 			if (error instanceof TranslationFailure) throw fromTranslationFailure(error);
 			throw error;
 		}
-		const modelMapping = {
-			originalModel: ctx.modelMapping.originalModel,
-			mappedModel: responsesPayload.model
-		};
 		applyContextManagement(responsesPayload, ctx.selectedModel?.capabilities.limits.max_prompt_tokens);
 		compactInputByLatestCompaction(responsesPayload);
 		const { vision, initiator } = getResponsesRequestOptions(responsesPayload);
@@ -50927,7 +51168,7 @@ const responsesApiEntry = {
 				signal: ctx.upstreamSignal.signal,
 				requestContext: ctx.requestContext
 			}), ctx.upstreamSignal),
-			modelMapping
+			modelMapping: ctx.modelMapping
 		};
 	}
 };
@@ -50943,10 +51184,7 @@ const chatCompletionsEntry = {
 			if (error instanceof TranslationFailure) throw fromTranslationFailure(error);
 			throw error;
 		}
-		const modelMapping = {
-			originalModel: ctx.modelMapping.originalModel,
-			mappedModel: plan.resolvedModel
-		};
+		const modelMapping = appendModelStep(ctx.modelMapping, "MODEL_RESOLVE", plan.resolvedModel);
 		consola.debug("Claude Code requested model:", ctx.anthropicPayload.model, "-> Copilot model:", plan.resolvedModel);
 		if (consola.level >= 4) consola.debug("Planned Copilot request payload:", JSON.stringify(plan.payload));
 		return {
@@ -50996,17 +51234,33 @@ async function handleMessagesCore({ body, signal, headers }) {
 	const requestContext = normalizeAnthropicRequestContext(anthropicPayload, headers);
 	if (consola.level >= 4) consola.debug("Anthropic request payload:", JSON.stringify(anthropicPayload));
 	const rewrite = applyModelRewrite(anthropicPayload);
+	const steps = [];
+	if (rewrite.reason) steps.push({
+		tag: rewrite.reason,
+		result: rewrite.model
+	});
 	const betaResult = processAnthropicBetaHeader(headers.get("anthropic-beta"), anthropicPayload.model);
 	if (betaResult.upgradeTarget) {
 		consola.debug(`Beta header context upgrade: ${anthropicPayload.model} → ${betaResult.upgradeTarget}`);
 		anthropicPayload.model = betaResult.upgradeTarget;
+		steps.push({
+			tag: "BETA_UPGRADE",
+			result: betaResult.upgradeTarget
+		});
 	}
 	const anthropicBetaHeader = betaResult.header;
 	const modelRouting = applyMessagesModelPolicy(anthropicPayload, { betaUpgraded: !!betaResult.upgradeTarget });
+	if (modelRouting.reason === "context-upgrade") steps.push({
+		tag: "CONTEXT_UPGRADE",
+		result: modelRouting.routedModel
+	});
+	else if (modelRouting.reason === "compact") steps.push({
+		tag: "COMPACT",
+		result: modelRouting.routedModel
+	});
 	const modelMapping = {
 		originalModel: rewrite.originalModel,
-		rewrittenModel: rewrite.model,
-		mappedModel: modelRouting.routedModel
+		steps
 	};
 	if (modelRouting.reason) consola.debug(`Routed anthropic request via ${modelRouting.reason}:`, `${modelRouting.originalModel} -> ${modelRouting.routedModel}`);
 	const selectedModel = findModelById(anthropicPayload.model);
@@ -51040,8 +51294,10 @@ async function handleMessagesCore({ body, signal, headers }) {
 			upstreamSignal: retrySignal,
 			modelMapping: {
 				originalModel: rewrite.originalModel,
-				rewrittenModel: rewrite.model,
-				mappedModel: upgradeTarget
+				steps: [...steps, {
+					tag: "RETRY_UPGRADE",
+					result: upgradeTarget
+				}]
 			}
 		});
 	}
@@ -51376,8 +51632,13 @@ function fixStreamIds(rawData, eventName, state) {
 function createResponsesPassthroughStrategy(copilotClient, payload, options) {
 	const tracker = createStreamIdTracker();
 	return {
-		execute() {
-			return copilotClient.createResponses(payload, options);
+		async execute() {
+			try {
+				return await copilotClient.createResponses(payload, options);
+			} catch (error) {
+				if (error instanceof HTTPError && error.status === 400) dumpFailedPayload(payload, error).catch(() => {});
+				throw error;
+			}
 		},
 		isStream(result) {
 			return Boolean(payload.stream) && isAsyncIterable(result);
@@ -51414,6 +51675,29 @@ function tryExtractTerminalResponse(rawData) {
 		if (response && typeof response === "object") return response;
 	} catch {}
 }
+const DUMP_DIR = join(PATHS.APP_DIR, "dumps");
+const MAX_DUMPS = 20;
+const TIMESTAMP_CHARS_RE = /[:.]/g;
+const DUMP_FILE_RE = /^\d{3}-/;
+async function dumpFailedPayload(payload, error) {
+	try {
+		await mkdir(DUMP_DIR, { recursive: true });
+		const now = (/* @__PURE__ */ new Date()).toISOString();
+		const ts = now.replace(TIMESTAMP_CHARS_RE, "-");
+		const file = join(DUMP_DIR, `${error.status}-${ts}.json`);
+		await writeFile(file, JSON.stringify({
+			timestamp: now,
+			error: {
+				status: error.status,
+				message: error.message
+			},
+			payload
+		}, null, 2));
+		consola.warn(`Dumped failed /responses payload → ${file}`);
+		const dumps = (await readdir(DUMP_DIR)).filter((f) => DUMP_FILE_RE.test(f) && f.endsWith(".json")).sort();
+		if (dumps.length > MAX_DUMPS) await Promise.all(dumps.slice(0, dumps.length - MAX_DUMPS).map((f) => unlink(join(DUMP_DIR, f)).catch(() => {})));
+	} catch {}
+}
 
 //#endregion
 //#region src/routes/responses/handler.ts
@@ -51458,8 +51742,10 @@ async function handleResponsesCore({ body, signal, headers }) {
 		result,
 		modelMapping: {
 			originalModel: rewrite.originalModel,
-			rewrittenModel: rewrite.model,
-			mappedModel: effectivePayload.model
+			steps: rewrite.reason ? [{
+				tag: rewrite.reason,
+				result: rewrite.model
+			}] : []
 		}
 	};
 }
@@ -51508,8 +51794,52 @@ function rejectUnsupportedBuiltinTools(payload) {
 	for (const tool of payload.tools) if (tool.type === "web_search") throwInvalidRequestError("The selected Copilot endpoint does not support the Responses web_search tool.", "tools", "unsupported_tool_web_search");
 }
 function applyResponsesInputPolicies(payload) {
+	payload.store = false;
+	stripUnresolvableInputItems(payload);
+	stripPhaseFromInputMessages(payload);
 	rejectUnsupportedRemoteImageUrls(payload);
 }
+/**
+* Strip `phase` from input message items. The `phase` field is an output
+* annotation that some models may reject when sent back as input.
+*/
+function stripPhaseFromInputMessages(payload) {
+	if (!Array.isArray(payload.input)) return;
+	let stripped = 0;
+	for (const item of payload.input) {
+		if (typeof item !== "object" || item === null) continue;
+		const rec = item;
+		if ((!("type" in rec) || rec.type === "message") && "phase" in rec) {
+			delete rec.phase;
+			stripped++;
+		}
+	}
+	if (stripped > 0) consola.debug(`Stripped phase from ${stripped} input message item(s)`);
+}
+/**
+* Remove input items that Copilot cannot resolve and would trigger 404:
+* - `item_reference` items (opaque IDs from store=true sessions)
+* - `function_call_output` items whose `call_id` has no matching prior
+*   `function_call` in the same input array (orphaned outputs)
+*/
+function stripUnresolvableInputItems(payload) {
+	if (!Array.isArray(payload.input)) return;
+	const functionCallIds = /* @__PURE__ */ new Set();
+	for (const item of payload.input) {
+		if (typeof item !== "object" || item === null) continue;
+		const rec = item;
+		if (rec.type === "function_call" && typeof rec.call_id === "string") functionCallIds.add(rec.call_id);
+	}
+	const originalLength = payload.input.length;
+	payload.input = payload.input.filter((item) => {
+		if (typeof item !== "object" || item === null) return true;
+		const rec = item;
+		if (rec.type === "item_reference") return false;
+		if (rec.type === "function_call_output" && typeof rec.call_id === "string" && !functionCallIds.has(rec.call_id)) return false;
+		return true;
+	});
+	if (payload.input.length !== originalLength) consola.debug(`Stripped ${originalLength - payload.input.length} unresolvable input items (item_reference / orphaned function_call_output)`);
+}
 function rejectUnsupportedRemoteImageUrls(payload) {
 	if (!Array.isArray(payload.input) || !containsRemoteImageUrl(payload.input)) return;
 	throwInvalidRequestError("The selected Copilot endpoint does not support external image URLs on the Responses API. Use file_id or data URL image input instead.", "input", "unsupported_input_image_remote_url");
@@ -51624,7 +51954,7 @@ function parseBooleanParam(value) {
 //#region src/routes/responses/route.ts
 function createResponsesRoutes() {
 	return new Elysia().use(requestGuardPlugin).post("/responses", async function* ({ body, request, server }) {
-		if (hasStreamingFlag(body)) disableIdleTimeout(server, request);
+		disableIdleTimeout(server, request);
 		const { result, modelMapping } = await handleResponsesCore({
 			body,
 			signal: request.signal,
@@ -51633,7 +51963,8 @@ function createResponsesRoutes() {
 		if (modelMapping) setRequestModelMapping(request, modelMapping);
 		if (result.kind === "json") return result.data;
 		yield* sseAdapter(result.generator);
-	}, { guarded: true }).post("/responses/input_tokens", async ({ body, request }) => {
+	}, { guarded: true }).post("/responses/input_tokens", async ({ body, request, server }) => {
+		disableIdleTimeout(server, request);
 		return handleCreateResponseInputTokensCore({
 			body,
 			headers: request.headers,
@@ -51705,7 +52036,7 @@ function createServer(options) {
 		const model = body && typeof body === "object" && "model" in body ? body.model : void 0;
 		if (typeof model === "string") setRequestModelMapping(request, {
 			originalModel: model,
-			mappedModel: model
+			steps: []
 		});
 	}).onAfterResponse(({ request, requestStart, set }) => {
 		const elapsed = formatElapsed(requestStart);