@mars-sea/dsh-commandcode-provider 0.13.1 → 0.13.3

package/lib/index.d.ts

@@ -4,7 +4,7 @@ import { CredentialRef } from "@deepseek-ai/dsh-credentials";
 import { TypertRemoteService, TypertSchema } from "@deepseek-ai/dsh-typert-protocol";
 import { WebRuntime, WebSearchProvider, WebSearchRequest, WebSearchResult } from "@deepseek-ai/dsh-web";
 import { Context } from "@deepseek-ai/cordis";
-import { AttachmentStore } from "@deepseek-ai/dsh-attachment";
+import { AttachmentStore, ImageAttachmentRef, ImageRequestTarget } from "@deepseek-ai/dsh-attachment";
 import { CommandDefinition } from "@deepseek-ai/dsh-commands";
 //#region src/provider-errors.d.ts
 /**
@@ -61,7 +61,7 @@ type CommandCodeProtocol = 'cli' | 'openai' | 'messages';
 declare const DEFAULT_MAX_OUTPUT_TOKENS = 131072;
 //#endregion
 //#region src/adapter.d.ts
-declare const COMMAND_CODE_CLI_VERSION = "1.82.1";
+declare const COMMAND_CODE_CLI_VERSION = "1.82.2";
 declare const DEFAULT_API_BASE = "https://api.commandcode.ai";
 /**
  * The output budget this bundle asks for, and the ceiling it will never exceed.
@@ -284,6 +284,8 @@ interface CommandCodeUsage {
   totalTokensOut: number;
   totalCredits: number;
   periodBasis: string;
+  /** 官方汇总返回的统计起点；缺失时不把查询参数当作已确认的统计范围。 */
+  periodStart?: number;
 }
 /** Credit/limit state from `/alpha/billing/credits`. */
 interface CommandCodeCredits {
@@ -338,6 +340,8 @@ interface CommandCodePlan {
   monthlyCredits: number | null;
   /** Billing period end in millis; 0 when the endpoint did not report one. */
   currentPeriodEnd: number;
+  /** 官方订阅账期起点，毫秒；旧宿主或未报告该字段时缺省。 */
+  currentPeriodStart?: number;
 }
 /**
  * Why every account endpoint failed at once (the report then carries no data
@@ -443,6 +447,25 @@ declare class CommandCodeAdapter<C extends CommandCodeConnectionOptions = Comman
    * requirement — the meter throws when the counts differ.
    */
   imageRequestPricing(_provider: string, model: string): LlmImageRequestPricing | undefined;
+  /**
+   * The dimensions and byte target this route actually puts on the wire.
+   *
+   * The base default answers with the NORMALIZED original (`image.width`,
+   * `image.height`, `image.bytes`), which is right only for an adapter that
+   * does not resize. This route always re-encodes through
+   * `readImageRequest(ref, requestImageTarget(ref))` — long edge 1568, encoded
+   * bytes capped at 1 MiB — so the default would over-report both dimensions
+   * and bytes for every image, most visibly for Retina screenshots.
+   *
+   * `imageRequestPricing` above already prices at this same target; exposing it
+   * here keeps the harness's own readers (the CUA screenshot path calls
+   * `ctx.llm.imageRequestTarget`) consistent with what the gateway receives.
+   * @param _provider - registered provider route; one route per adapter here.
+   * @param _model - exact model id; the target is model-independent on this route.
+   * @param image - normalized attachment whose request projection is resolved.
+   * @returns the same target `readImageRequest` encodes to, before base64 expansion.
+   */
+  imageRequestTarget(_provider: string, _model: string, image: ImageAttachmentRef): ImageRequestTarget;
   listModels(provider: string, opts?: {
     unfiltered?: boolean;
   }): Promise<readonly LlmModelInfo[]>;

package/lib/index.js

@@ -431,7 +431,7 @@ const KNOWN_EFFORTS = {
 		"xhigh",
 		"max"
 	],
-	"inclusionai/ling-3.1-flash:free": [
+	"inclusionai/ling-3.1-flash": [
 		"low",
 		"medium",
 		"high"
@@ -711,22 +711,20 @@ function requiresMessagesEndpoint(modelId) {
 * **Generated — do not hand-edit.** `scripts/sync-output-limits.mjs` rewrites
 * this literal from models.dev; run it after a catalog sync, and use its
 * `--check` in CI. Command Code resells these models through their VENDOR's own
-* API, so the vendor's published ceiling is the value recorded: 77 of the 89
-* catalog ids as of 2026-10-14. Each row carries a comment naming the entry it
+* API, so the vendor's published ceiling is the value recorded: 76 of the 89
+* catalog ids as of 2026-10-15. Each row carries a comment naming the entry it
 * came from.
 *
-* Two of those 77 rest on more than a vendor lookup, both hand-recorded in the
-* script's `MANUAL_CEILINGS`. `gpt-6.1-sol` (128000) comes from OpenAI's own row
-* and is corroborated by every other provider carrying it.
-* `inclusionai/ling-3.1-flash:free` (32768) has no models.dev row at all, so it
-* takes the Ling family's uniform ceiling; `inclusionai/ling-3.0-flash-sante:free`
-* (32768) lost both its vendor row and its unanimous third-party vote (vercel
-* files 32000), and is pinned so a regeneration cannot drop a live model's
-* ceiling back to the default. All are still correctable — a stricter gateway
-* costs one refused request, and `streamRequest()` remembers what the endpoint
-* said.
-*
-* The 12 unlisted models are ones no vendor publishes and the aggregators
+* One of those 76 rests on more than a vendor lookup, hand-recorded in the
+* script's `MANUAL_CEILINGS`: `inclusionai/ling-3.0-flash-sante:free` (32768)
+* lost both its vendor row and its unanimous third-party vote (vercel files
+* 32000), and is pinned so a regeneration cannot drop a live model's ceiling
+* back to the default. (`gpt-6.1-sol` (128000) needs no pin — OpenAI's own row
+* carries it and every other provider carrying the id agrees.) All are still
+* correctable — a stricter gateway costs one refused request, and
+* `streamRequest()` remembers what the endpoint said.
+*
+* The 13 unlisted models are ones no vendor publishes and the aggregators
 * disagree about. They are NOT a gap: a refusal states the ceiling in full, and
 * `streamRequest()` reads it off, retries once and remembers it for the model
 * (issue #71). A vendor figure the gateway turns out to be stricter than costs
@@ -772,7 +770,6 @@ const MODEL_OUTPUT_TOKEN_LIMITS = /* @__PURE__ */ new Map([
 	["gpt-6-sol", 128e3],
 	["gpt-6.1-sol", 128e3],
 	["inclusionai/ling-3.0-flash-sante:free", 32768],
-	["inclusionai/ling-3.1-flash:free", 32768],
 	["meituan/LongCat-2.0", 131072],
 	["meta/muse-spark-1.1", 131072],
 	["meta/muse-spark-1.2", 131072],
@@ -866,7 +863,7 @@ const KNOWN_PLANS = {
 	"gpt-6-luna": "go",
 	"meituan/LongCat-2.0": "go",
 	"inclusionai/ling-3.0-flash-sante:free": "go",
-	"inclusionai/ling-3.1-flash:free": "go",
+	"inclusionai/ling-3.1-flash": "go",
 	"meta/muse-spark-1.2-contributor": "go",
 	"meta/muse-spark-1.3-contributor": "go",
 	"mistral/mistral-large-4": "go",
@@ -1076,10 +1073,6 @@ const KNOWN_DEALS = {
 		label: "FREE",
 		free: true
 	},
-	"inclusionai/ling-3.1-flash:free": {
-		label: "FREE",
-		free: true
-	},
 	"stealth/glyph-cluster:free": {
 		label: "FREE",
 		free: true
@@ -1721,8 +1714,8 @@ const MODEL_PRICE_ROWS = [
 		],
 		allowance: {
 			go: 10,
-			goat: 60,
-			pro: 70
+			goat: 100,
+			pro: 100
 		}
 	},
 	{
@@ -2144,6 +2137,19 @@ const MODEL_PRICE_ROWS = [
 			pro: 30
 		}
 	},
+	{
+		id: "ling-3.1-flash",
+		rates: [
+			.3,
+			.9,
+			.06
+		],
+		allowance: {
+			go: 6,
+			goat: 20,
+			pro: 30
+		}
+	},
 	{
 		id: "longcat-2.0",
 		rates: [
@@ -2554,6 +2560,20 @@ const MODEL_PRICE_ROWS = [
 			pro: 30
 		}
 	},
+	{
+		id: "qwen-3.8-max-prime",
+		rates: [
+			4,
+			12,
+			.5,
+			5
+		],
+		allowance: {
+			go: 6,
+			goat: 20,
+			pro: 30
+		}
+	},
 	{
 		id: "qwen-3.8-omni-flash",
 		rates: [
@@ -2826,6 +2846,12 @@ function tokensOf$1(value) {
 		cacheWriteTokens: u.cacheWriteTokens ?? 0
 	};
 }
+/** The four buckets of an already-parsed record, or undefined when one is missing. */
+function costTokensOf(value) {
+	const t = record$3(value);
+	if (!t || !COST_TOKEN_KEYS.every((k) => finite(t[k]))) return void 0;
+	return t;
+}
 /** Current v2 settlements store the last usage in their stream; v1 stores chunk events. */
 function usageOf(type, data) {
 	if (type === "assistant/chunk") {
@@ -2846,16 +2872,51 @@ function parseState(value) {
 	const s = record$3(value), facts = record$3(s?.facts);
 	if (!s || !facts || !finite(facts.pricingKey) || !Array.isArray(facts.groups)) throw new TypeError("invalid commandCodeCost state");
 	const group = (v) => {
-		const g = record$3(v), t = record$3(g?.tokens);
-		return !!g && typeof g.provider === "string" && typeof g.model === "string" && (g.at === null || finite(g.at)) && finite(g.contextTokens) && !!t && COST_TOKEN_KEYS.every((k) => finite(t[k]));
+		const g = record$3(v);
+		return !!g && typeof g.provider === "string" && typeof g.model === "string" && (g.at === null || finite(g.at)) && finite(g.contextTokens) && costTokensOf(g.tokens) !== void 0;
 	};
 	const sel = record$3(s.selection), last = record$3(s.last);
-	if (!(s.selection === null || sel && typeof sel.provider === "string" && typeof sel.model === "string") || !(s.at === null || finite(s.at)) || !facts.groups.every(group) || !(s.last === null || last && finite(last.turn) && finite(last.step) && group(last.group))) throw new TypeError("invalid commandCodeCost state");
-	return value;
+	const summaryTokens = facts.summaryTokens === void 0 ? zeroCostTokens() : costTokensOf(facts.summaryTokens);
+	if (summaryTokens === void 0 || !(s.selection === null || sel && typeof sel.provider === "string" && typeof sel.model === "string") || !(s.at === null || finite(s.at)) || !facts.groups.every(group) || !(s.last === null || last && finite(last.turn) && finite(last.step) && group(last.group))) throw new TypeError("invalid commandCodeCost state");
+	return {
+		...value,
+		facts: {
+			...facts,
+			summaryTokens
+		}
+	};
 }
 /** Optional registry uses only synchronous parse/view; no new runtime module is required. */
 function createCostProjection(table = modelPriceTable()) {
 	const version = pricingKey(table);
+	/** Groups that still carry tokens; a retry that replaced a group can drive one to zero. */
+	const liveGroups = (groups) => groups.filter((g) => COST_TOKEN_KEYS.some((k) => g.tokens[k] > 0));
+	/**
+	* Fold one settled usage group into `groups`, first undoing the group a retry
+	* supersedes. Shared by turn/step settlements and summary calls, which have
+	* no turn/step of their own.
+	*/
+	const foldGroup = (groups, previous, group, tokens) => {
+		const byKey = /* @__PURE__ */ new Map();
+		groups.forEach((g, i) => {
+			const key = costGroupKey(g, table);
+			if (!byKey.has(key)) byKey.set(key, i);
+		});
+		if (previous) {
+			const old = byKey.get(costGroupKey(previous, table));
+			if (old !== void 0) for (const k of COST_TOKEN_KEYS) groups[old].tokens[k] -= previous.tokens[k];
+		}
+		const targetIdx = byKey.get(costGroupKey(group, table));
+		let target = targetIdx === void 0 ? void 0 : groups[targetIdx];
+		if (!target) {
+			target = {
+				...group,
+				tokens: zeroCostTokens()
+			};
+			groups.push(target);
+		}
+		for (const k of COST_TOKEN_KEYS) target.tokens[k] += tokens[k];
+	};
 	return {
 		key: "commandCodeCost",
 		stateVersion: version,
@@ -2866,7 +2927,8 @@ function createCostProjection(table = modelPriceTable()) {
 			last: null,
 			facts: {
 				pricingKey: version,
-				groups: []
+				groups: [],
+				summaryTokens: zeroCostTokens()
 			}
 		}),
 		apply(state, event) {
@@ -2887,6 +2949,32 @@ function createCostProjection(table = modelPriceTable()) {
 				at: finite(event.time) ? event.time : null,
 				last: null
 			};
+			if (event.type === "compaction/summary") {
+				const summary = tokensOf$1(data.usage);
+				if (summary === void 0) return state;
+				const group = {
+					provider: typeof data.provider === "string" ? data.provider : "",
+					model: typeof data.model === "string" ? data.model : "",
+					at: finite(event.time) ? event.time : null,
+					contextTokens: summary.uncachedInputTokens + summary.cacheReadTokens + summary.cacheWriteTokens,
+					tokens: summary
+				};
+				const groups = state.facts.groups.map((g) => ({
+					...g,
+					tokens: { ...g.tokens }
+				}));
+				foldGroup(groups, void 0, group, summary);
+				const summaryTokens = { ...state.facts.summaryTokens };
+				for (const k of COST_TOKEN_KEYS) summaryTokens[k] += summary[k];
+				return {
+					...state,
+					facts: {
+						pricingKey: version,
+						groups: liveGroups(groups),
+						summaryTokens
+					}
+				};
+			}
 			const tokens = usageOf(event.type, data);
 			if (tokens === void 0 || !finite(data.turn) || !finite(data.step)) return state;
 			const previous = state.last?.turn === data.turn && state.last.step === data.step ? state.last.group : void 0;
@@ -2902,26 +2990,7 @@ function createCostProjection(table = modelPriceTable()) {
 				...g,
 				tokens: { ...g.tokens }
 			}));
-			const byKey = /* @__PURE__ */ new Map();
-			groups.forEach((g, i) => {
-				const key = costGroupKey(g, table);
-				if (!byKey.has(key)) byKey.set(key, i);
-			});
-			const groupKey = costGroupKey(group, table);
-			if (previous) {
-				const old = byKey.get(costGroupKey(previous, table));
-				if (old !== void 0) for (const k of COST_TOKEN_KEYS) groups[old].tokens[k] -= previous.tokens[k];
-			}
-			const targetIdx = byKey.get(groupKey);
-			let target = targetIdx === void 0 ? void 0 : groups[targetIdx];
-			if (!target) {
-				target = {
-					...group,
-					tokens: zeroCostTokens()
-				};
-				groups.push(target);
-			}
-			for (const k of COST_TOKEN_KEYS) target.tokens[k] += tokens[k];
+			foldGroup(groups, previous, group, tokens);
 			return {
 				...state,
 				last: {
@@ -2931,7 +3000,8 @@ function createCostProjection(table = modelPriceTable()) {
 				},
 				facts: {
 					pricingKey: version,
-					groups: groups.filter((g) => COST_TOKEN_KEYS.some((k) => g.tokens[k] > 0))
+					groups: liveGroups(groups),
+					summaryTokens: state.facts.summaryTokens
 				}
 			};
 		},
@@ -3009,6 +3079,20 @@ function foldRequests(session, log) {
 			}
 			continue;
 		}
+		if (event.type === "compaction/summary") {
+			if (event.seq >= log.inheritedEventCount && d.provider === "commandcode") {
+				const tokens = tokensOf(d.usage);
+				facts.push({
+					session,
+					model: typeof d.model === "string" ? d.model : "",
+					at: event.time,
+					retry: false,
+					outcome: "completed",
+					...tokens === void 0 ? {} : { tokens }
+				});
+			}
+			continue;
+		}
 		if (event.type === "step/start" || event.type === "llm/retry-started") {
 			begin(event, d, event.type === "llm/retry-started");
 			continue;
@@ -4157,8 +4241,14 @@ function parseProviderError(errText) {
 * `CONTEXT_WINDOW_EXCEEDED`, because that failure has a recovery the harness
 * performs and a generic provider error does not. Every message built here is
 * bilingual — the harness renders it verbatim in its retry chrome.
+*
+* The optional `priorRejection` records a refusal from an earlier transport
+* attempt in this same call (see {@link isMessagesPermissionRejection}), so a
+* failed CLI fallback still shows what the Messages surface said first. It is
+* presentation only: classification, retry budget and failure code all keep
+* reading the current attempt.
 */
-function generateHttpError(status, errText, retryAfterMs, parsed = parseProviderError(errText)) {
+function generateHttpError(status, errText, retryAfterMs, parsed = parseProviderError(errText), priorRejection) {
 	const providerCode = parsed.nested === void 0 ? void 0 : parsed.code;
 	const providerDetail = parsed.nested === void 0 ? "" : [
 		parsed.code,
@@ -4175,9 +4265,14 @@ function generateHttpError(status, errText, retryAfterMs, parsed = parseProvider
 	if (status === 422 && isZdrNoProviders(providerCode, providerDetail, errText)) return bilingual("PROVIDER_HTTP_ERROR", `Command Code API error 422 (${detail}): zero data retention is enforced but no ZDR-capable upstream is available for this model — turn the ZDR setting off (or use another model)`, "Command Code API 返回 422：已开启零数据保留（ZDR），但该模型当前没有可用的 ZDR 上游——请关闭 ZDR 开关，或改用其他模型", { status: 422 });
 	if (status === 401) return bilingual("INVALID_CREDENTIAL", `Command Code API error 401 (${detail}): the API key is missing or invalid — check the key stored for COMMANDCODE_API_KEY (Models page) or the auth file`, "Command Code API 返回 401：API 密钥缺失或无效——请在设置页检查 COMMANDCODE_API_KEY 存储的密钥，或检查 auth 文件", { status: 401 });
 	if (status === 413) return bilingual("PROVIDER_HTTP_ERROR", "Command Code API error 413: the request body exceeds the provider's size limit (the cap is undocumented, measured at about 50 MB) — the session history is too large to send, usually because of accumulated image attachments. Start a new session, or drop the image-heavy part of this one, and retry", "Command Code API 返回 413：请求体超过服务端的体积上限（官方未公开，实测约 50 MB）——会话历史过大，通常是历史图片累积所致。请新建会话，或移除本会话中图片较多的部分后重试", { status: 413 });
-	const diagnosticCode = parsed.code ?? /^\s*(MODEL_NOT_IN_PLAN|INSUFFICIENT_CREDITS|USAGE_EXCEEDED|PREMIUM_CREDITS_EXHAUSTED)\b/i.exec(parsed.message ?? "")?.[1];
-	const diagnostic = diagnosticCode !== void 0 && /^[A-Za-z0-9_.:-]{1,80}$/.test(diagnosticCode) ? `（${diagnosticCode}）` : "";
-	return new LlmError(`Command Code 请求失败（HTTP ${status}）${diagnostic}：` + (status === 429 ? "请求受到用量或速率限制，请稍后重试并检查账户额度。" : status === 408 ? "服务端处理超时，请稍后重试。" : status >= 500 ? "上游服务暂时无法完成请求，请稍后重试。" : "服务端拒绝了本次请求，请检查模型、套餐和请求设置。"), httpErrorCode(status), {
+	const safeIdentifier = (value) => value !== void 0 && /^[A-Za-z0-9_.:-]{1,80}$/.test(value) ? value : void 0;
+	const diagnosticCode = safeIdentifier(parsed.code) ?? /^\s*(MODEL_NOT_IN_PLAN|INSUFFICIENT_CREDITS|USAGE_EXCEEDED|PREMIUM_CREDITS_EXHAUSTED)\b/i.exec(parsed.message ?? "")?.[1];
+	const diagnosticType = safeIdentifier(parsed.type);
+	const identifiers = [diagnosticCode === void 0 ? void 0 : `错误码：${diagnosticCode}`, diagnosticType === void 0 ? void 0 : `错误类型：${diagnosticType}`].filter((value) => value !== void 0);
+	const diagnostic = identifiers.length === 0 ? "" : `（${identifiers.join("；")}）`;
+	const priorIdentifier = priorRejection === void 0 ? void 0 : safeIdentifier(priorRejection);
+	const priorNote = priorIdentifier === void 0 ? "" : `此前 Messages 接口以 ${priorIdentifier} 拒绝，已回退 CLI 直连，仍未成功。`;
+	return new LlmError(`Command Code 请求失败（HTTP ${status}）${diagnostic}：` + (status === 429 ? "请求受到用量或速率限制，请稍后重试并检查账户额度。" : status === 408 ? "服务端处理超时，请稍后重试。" : status >= 500 ? "上游服务暂时无法完成请求，请稍后重试。" : "服务端拒绝了本次请求，请检查模型、套餐和请求设置。") + priorNote, httpErrorCode(status), {
 		status,
 		...retryAfterMs !== void 0 && retryAfterMs > 0 && retryAfterMs <= 9e5 ? { providerRetryAfterMs: retryAfterMs } : {}
 	});
@@ -4327,6 +4422,22 @@ function isUpgradeRequiredError(status, errText, parsed = parseProviderError(err
 	if (parsed.message?.toLowerCase().includes("go plan") && parsed.message.toLowerCase().includes("api access")) return true;
 	return false;
 }
+/**
+* True when the Messages surface (`/provider/v1/messages`) refused one request
+* with Anthropic's `permission_error` (issue #84).
+*
+* 职责：把「反代 Claude 链路对该账号不放行」这一种拒绝，从普通权限失败里
+* 单独认出来，交给适配器决定是否改用 CLI 直连（`/alpha/generate`）。
+*
+* 为什么边界如此窄：只认 HTTP 403 加上结构化的 `code`/`type` 等于
+* `permission_error`，不扫原始正文。正文可能回显输入或返回整页英文错误，
+* 扫文本会把无关失败也变成一次传输切换；而真正「模型不在套餐内」的拒绝
+* （`MODEL_NOT_IN_PLAN` 等其它标识）继续按原样上报，不被这条规则吞掉。
+*/
+function isMessagesPermissionRejection(status, parsed) {
+	if (status !== 403) return false;
+	return (parsed.code ?? parsed.type)?.toLowerCase() === "permission_error";
+}
 //#endregion
 //#region src/stream-trace.ts
 /**
@@ -5902,7 +6013,7 @@ var GatewayFacts = class {
 * per-request options thunk and an API-key resolver from the plugin entry
 * (src/index.ts), so a settings change reaches the very next request.
 */
-const COMMAND_CODE_CLI_VERSION = "1.82.1";
+const COMMAND_CODE_CLI_VERSION = "1.82.2";
 const DEFAULT_API_BASE = "https://api.commandcode.ai";
 /**
 * The output budget this bundle asks for, and the ceiling it will never exceed.
@@ -6057,7 +6168,7 @@ function requestFingerprint(protocol, body) {
 	};
 }
 /** Parse a billing-period timestamp (ISO string or millis) into millis; 0 when absent/invalid. */
-function periodEndValue(value) {
+function periodTimestampValue(value) {
 	const asNumber = numberValue(value);
 	if (asNumber !== void 0) return asNumber;
 	const asString = stringValue(value);
@@ -6906,7 +7017,7 @@ const USAGE_ENDPOINT_COUNT = 4;
 */
 function parseUsageTotals(usage) {
 	if (usage === void 0) return void 0;
-	return {
+	const result = {
 		totalCount: numberValue(usage.totalCount) ?? 0,
 		totalCost: numberValue(usage.totalCost) ?? 0,
 		successRate: numberValue(usage.successRate) ?? 0,
@@ -6915,8 +7026,11 @@ function parseUsageTotals(usage) {
 		totalTokensIn: numberValue(usage.totalTokensIn) ?? 0,
 		totalTokensOut: numberValue(usage.totalTokensOut) ?? 0,
 		totalCredits: numberValue(usage.totalCredits) ?? 0,
-		periodBasis: stringValue(usage.periodBasis) ?? "billing-period"
+		periodBasis: stringValue(usage.periodBasis) ?? "unreported"
 	};
+	const start = periodTimestampValue(usage.periodStart);
+	if (start > 0 && Number.isFinite(new Date(start).getTime())) result.periodStart = start;
+	return result;
 }
 /**
 * Parse one window block into a window limit, or undefined when the endpoint
@@ -6962,9 +7076,8 @@ function parseCreditLimits(credits) {
 	return parsed;
 }
 /**
-* Parse the account identity from whoami plus the plan identity from the
-* subscriptions/credits payloads. Returns the org id for the subscriptions
-* query alongside, so getUsage fetches whoami first and the rest in parallel.
+* Read the account identity and organization scope from whoami. Every billing
+* and usage request must use this same organization, so getUsage reads it first.
 */
 function parseAccountIdentity(whoami) {
 	const whoamiData = whoami !== void 0 && isRecord(whoami.user) ? whoami.user : void 0;
@@ -7533,6 +7646,27 @@ var CommandCodeAdapter = class CommandCodeAdapter extends LlmAdapter {
 		const imageCapable = KNOWN_IMAGE_MODELS.has(model);
 		return { priceImages: (images) => images.map((image) => priceRequestImage(image, model, imageCapable)) };
 	}
+	/**
+	* The dimensions and byte target this route actually puts on the wire.
+	*
+	* The base default answers with the NORMALIZED original (`image.width`,
+	* `image.height`, `image.bytes`), which is right only for an adapter that
+	* does not resize. This route always re-encodes through
+	* `readImageRequest(ref, requestImageTarget(ref))` — long edge 1568, encoded
+	* bytes capped at 1 MiB — so the default would over-report both dimensions
+	* and bytes for every image, most visibly for Retina screenshots.
+	*
+	* `imageRequestPricing` above already prices at this same target; exposing it
+	* here keeps the harness's own readers (the CUA screenshot path calls
+	* `ctx.llm.imageRequestTarget`) consistent with what the gateway receives.
+	* @param _provider - registered provider route; one route per adapter here.
+	* @param _model - exact model id; the target is model-independent on this route.
+	* @param image - normalized attachment whose request projection is resolved.
+	* @returns the same target `readImageRequest` encodes to, before base64 expansion.
+	*/
+	imageRequestTarget(_provider, _model, image) {
+		return requestImageTarget(image);
+	}
 	async listModels(provider, opts) {
 		const connection = { ...this.deps.options() };
 		const catalog = await this.gatewayFacts.refresh(connection);
@@ -7879,16 +8013,22 @@ var CommandCodeAdapter = class CommandCodeAdapter extends LlmAdapter {
 		const report = { failures };
 		const { account, orgId } = parseAccountIdentity(await getJson("/alpha/whoami"));
 		if (account !== void 0) report.account = account;
-		const [usage, credits, subscription] = await Promise.all([
-			getJson("/alpha/usage/summary"),
-			getJson("/alpha/billing/credits"),
-			getJson(orgId === void 0 ? "/alpha/billing/subscriptions" : `/alpha/billing/subscriptions?orgId=${encodeURIComponent(orgId)}`)
-		]);
-		const totals = parseUsageTotals(usage);
+		const scopedPath = (path, since) => {
+			const params = new URLSearchParams();
+			if (orgId !== void 0) params.set("orgId", orgId);
+			if (since !== void 0) params.set("since", since);
+			const query = params.toString();
+			return query ? `${path}?${query}` : path;
+		};
+		const [credits, subscription] = await Promise.all([getJson(scopedPath("/alpha/billing/credits")), getJson(scopedPath("/alpha/billing/subscriptions"))]);
+		const subData = subscription !== void 0 && isRecord(subscription.data) ? subscription.data : void 0;
+		const rawStart = subData?.currentPeriodStart;
+		const periodStart = periodTimestampValue(rawStart);
+		const validStart = periodStart > 0 && Number.isFinite(new Date(periodStart).getTime());
+		const totals = parseUsageTotals(await getJson(scopedPath("/alpha/usage/summary", validStart ? String(rawStart) : void 0)));
 		if (totals !== void 0) report.usage = totals;
 		const limits = parseCreditLimits(credits);
 		if (limits !== void 0) report.credits = limits;
-		const subData = subscription !== void 0 && isRecord(subscription.data) ? subscription.data : void 0;
 		const creditsData = credits !== void 0 && isRecord(credits.credits) ? credits.credits : void 0;
 		const planId = stringValue(subData?.planId) ?? stringValue(creditsData?.planId);
 		if (subData !== void 0 || planId !== void 0) {
@@ -7898,8 +8038,9 @@ var CommandCodeAdapter = class CommandCodeAdapter extends LlmAdapter {
 				name: info?.name ?? planId ?? "",
 				status: stringValue(subData?.status) ?? "",
 				monthlyCredits: info?.monthlyCredits ?? null,
-				currentPeriodEnd: periodEndValue(subData?.currentPeriodEnd)
+				currentPeriodEnd: periodTimestampValue(subData?.currentPeriodEnd)
 			};
+			if (validStart) report.plan.currentPeriodStart = periodStart;
 		}
 		const blocked = classifyTotalFailure(failures, failed);
 		if (blocked !== void 0) report.blocked = blocked;
@@ -8053,6 +8194,7 @@ var CommandCodeAdapter = class CommandCodeAdapter extends LlmAdapter {
 		const reasoningEffort = effort === "off" && declared === void 0 ? void 0 : declared;
 		const systemText = [options.system ?? "", ...options.messages.filter((m) => m.role === "system").map((m) => m.content.map(blockText).filter(Boolean).join("\n"))].filter(Boolean).join("\n\n");
 		let protocol = this.resolveProtocol(connection, apiKey, options.model, modelEntry);
+		let messagesPermissionRejection;
 		let requestOptions = withSurfaceOffload(options, this.surfaceOffload, REQUEST_IMAGE_BUDGETS[0], connection.offloadSeenImagesForCache === true && protocol === "cli" ? options.model : void 0);
 		const buildBody = async (target) => {
 			const finishBody = timing.phase("body");
@@ -8143,8 +8285,10 @@ var CommandCodeAdapter = class CommandCodeAdapter extends LlmAdapter {
 					break;
 				}
 				const providerError = parseProviderError(attempt.errText);
-				if (protocol !== "cli" && isUpgradeRequiredError(attempt.status, attempt.errText, providerError)) {
+				const messagesPermissionDenied = protocol === "messages" && isMessagesPermissionRejection(attempt.status, providerError);
+				if (protocol !== "cli" && (messagesPermissionDenied || isUpgradeRequiredError(attempt.status, attempt.errText, providerError))) {
 					const wasMessages = protocol === "messages";
+					if (messagesPermissionDenied) messagesPermissionRejection = providerError.code ?? providerError.type;
 					protocol = "cli";
 					if (!wasMessages) this.rememberProtocol(connection.apiBase, apiKey, "cli");
 					requestOptions = withSurfaceOffload(options, this.surfaceOffload, REQUEST_IMAGE_BUDGETS[0], connection.offloadSeenImagesForCache === true ? options.model : void 0);
@@ -8164,6 +8308,7 @@ var CommandCodeAdapter = class CommandCodeAdapter extends LlmAdapter {
 					});
 					if (next !== void 0 && !tried.has(next)) {
 						apiKey = next;
+						messagesPermissionRejection = void 0;
 						const nextProtocol = this.resolveProtocol(connection, apiKey, options.model, modelEntry);
 						if (nextProtocol !== protocol) {
 							protocol = nextProtocol;
@@ -8183,7 +8328,7 @@ var CommandCodeAdapter = class CommandCodeAdapter extends LlmAdapter {
 						...wait > 0 && wait <= 9e5 ? { providerRetryAfterMs: wait } : {}
 					});
 				}
-				throw generateHttpError(attempt.status, attempt.errText, attempt.retryAfterMs, providerError);
+				throw generateHttpError(attempt.status, attempt.errText, attempt.retryAfterMs, providerError, messagesPermissionRejection);
 			}
 		} catch (error) {
 			trace.record("connect-error", {
@@ -8365,7 +8510,7 @@ function renderReport(report, locale, title) {
 	}
 	if (report.usage) {
 		const u = report.usage;
-		lines.push(commandCopy(locale, "usageHeader"), commandCopy(locale, "requestsLine").replace("{n}", String(u.completedCount)).replace("{f}", String(u.failedCount)).replace("{r}", successRateText(u.successRate)), commandCopy(locale, "costLine").replace("{money}", money(u.totalCost)).replace("{credits}", moneyShort(u.totalCredits)), commandCopy(locale, "tokensLine").replace("{in}", tokensCompact(u.totalTokensIn)).replace("{out}", tokensCompact(u.totalTokensOut)), "");
+		lines.push(commandCopy(locale, "usageHeader"), commandCopy(locale, "requestsLine").replace("{n}", String(u.totalCount)).replace("{f}", String(u.failedCount)).replace("{r}", successRateText(u.successRate)), commandCopy(locale, "costLine").replace("{money}", money(u.totalCost)).replace("{credits}", moneyShort(u.totalCredits)), commandCopy(locale, "tokensLine").replace("{in}", tokensCompact(u.totalTokensIn)).replace("{out}", tokensCompact(u.totalTokensOut)), "");
 	}
 	if (report.credits) {
 		const c = report.credits;
@@ -8486,6 +8631,7 @@ function parseUsageReport(value) {
 			totalCredits: numberField(usage, "totalCredits", "usage.totalCredits"),
 			periodBasis: stringField$1(usage, "periodBasis", "usage.periodBasis")
 		};
+		if (usage.periodStart !== void 0) report.usage.periodStart = numberField(usage, "periodStart", "usage.periodStart");
 	}
 	if (source.credits !== void 0) {
 		const credits = record$1(source.credits, "credits");
@@ -8513,6 +8659,7 @@ function parseUsageReport(value) {
 			monthlyCredits: monthly,
 			currentPeriodEnd: numberField(plan, "currentPeriodEnd", "plan.currentPeriodEnd")
 		};
+		if (plan.currentPeriodStart !== void 0) report.plan.currentPeriodStart = numberField(plan, "currentPeriodStart", "plan.currentPeriodStart");
 	}
 	return report;
 }
@@ -10756,7 +10903,7 @@ function apply(ctx, config) {
 		const resolved = await scopedPool(connection).resolveKey(model === void 0 ? {} : { model });
 		if (resolved !== void 0) return assertUsableApiKey(resolved.key, "llm-commandcode", resolved.slot.ref ?? `${resolved.slot.label} (config.apiKey)`);
 		const ref = connection.apiKeyEnv;
-		throw new LlmError(`llm-commandcode: no API key for provider route "${PROVIDER}"; store ${ref} through the credentials service (the web Models page writes it), export it in the launching environment, set config.apiKey, or run \`command-code login\` to write ~/.commandcode/auth.json`, "MISSING_CREDENTIAL");
+		throw new LlmError(`llm-commandcode: no API key for provider route "${PROVIDER}"; store ${ref} through the credentials service (the web Models page writes it), export it in the launching environment, set config.apiKey, or run \`command-code login\` to write ~/.commandcode/auth.json. The headless profile mounts no credentials service, so a key saved only in the store never reaches it: export the variable or put it in a .env file for those runs.`, "MISSING_CREDENTIAL");
 	};
 	const adapter = new CommandCodeAdapter({
 		options,