@mars-sea/dsh-commandcode-provider 0.12.10 → 0.13.1

package/lib/index.js

@@ -1,8 +1,8 @@
+import { CONTEXT_WINDOW_EXCEEDED_CODE, IMAGE_OFFLOAD_REQUIRED_CODE, LlmAdapter, LlmError, ProviderRequestId, ReasoningEffortId, ToolCallId, assertUsableApiKey, assistantStreamFirstTokenTime, attributionHeaders, errorChain, isContextWindowExceededError, offloadedImageText, projectOffloadedImages, requiredImageOffload, resolveRetryPolicy, textOnlyImageText } from "@deepseek-ai/dsh-llm";
 import { homedir, tmpdir } from "node:os";
 import { dirname, join } from "node:path";
 import z from "@deepseek-ai/schemastery";
 import { MAX_TIMER_DELAY_MS } from "@deepseek-ai/dsh-timeout";
-import { CONTEXT_WINDOW_EXCEEDED_CODE, IMAGE_OFFLOAD_REQUIRED_CODE, LlmAdapter, LlmError, ProviderRequestId, ReasoningEffortId, ToolCallId, assertUsableApiKey, attributionHeaders, errorChain, isContextWindowExceededError, offloadedImageText, projectOffloadedImages, requiredImageOffload, resolveRetryPolicy, textOnlyImageText } from "@deepseek-ai/dsh-llm";
 import { credentialRef } from "@deepseek-ai/dsh-credentials";
 import { launchEnvironmentOf } from "@deepseek-ai/dsh-launch-environment";
 import { appendFileSync, existsSync, mkdirSync, readFileSync } from "node:fs";
@@ -103,7 +103,7 @@ const PLAN_ORDER = Object.fromEntries(PLAN_TIER_ORDER.map((tier, index) => [tier
 * subscription-plan labels, deals, and hourly (peak/off-peak) pricing.
 *
 * Everything here is synced from official sources — the command-code CLI
-* bundle's model table (`dist/cli.mjs`, re-verified at command-code@1.79.2) and
+* bundle's model table (`dist/cli.mjs`, re-verified at command-code@1.82.1) and
 * the official plan/pricing/model docs; see the dsh-commandcode-upstream skill
 * for the extraction procedures. Keeping the snapshot in its own module
 * confines those frequent sync diffs here: src/adapter.ts holds only the stable
@@ -226,6 +226,11 @@ const KNOWN_EFFORTS = {
 		"high",
 		"max"
 	],
+	"google/gemini-4-argon": [
+		"low",
+		"medium",
+		"high"
+	],
 	"google/gemini-3.1-flash-lite": [
 		"low",
 		"medium",
@@ -380,6 +385,12 @@ const KNOWN_EFFORTS = {
 		"high",
 		"xhigh"
 	],
+	"meta/muse-wonder-1.0": [
+		"low",
+		"medium",
+		"high",
+		"xhigh"
+	],
 	"gpt-6-astra": [
 		"low",
 		"medium",
@@ -483,6 +494,7 @@ const KNOWN_IMAGE_MODELS = /* @__PURE__ */ new Set([
 	"deepseek/deepseek-v4-flash-vision-exp",
 	"deepseek/deepseek-v4.1-flash",
 	"deepseek/deepseek-v4.1-flash-fast",
+	"google/gemini-4-argon",
 	"google/gemini-3.1-flash-lite",
 	"google/gemini-3.5-flash",
 	"google/gemini-3.5-flash-lite",
@@ -505,6 +517,7 @@ const KNOWN_IMAGE_MODELS = /* @__PURE__ */ new Set([
 	"meta/muse-spark-1.2-contributor",
 	"meta/muse-spark-1.3",
 	"meta/muse-spark-1.3-contributor",
+	"meta/muse-wonder-1.0",
 	"mistral/mistral-large-4",
 	"moonshotai/Kimi-K2.5",
 	"moonshotai/Kimi-K2.6",
@@ -528,7 +541,7 @@ const KNOWN_IMAGE_MODELS = /* @__PURE__ */ new Set([
 ]);
 /**
 * Models WITHOUT a zero-data-retention upstream, per the official CLI's own
-* registry (`command-code@1.79.2` `dist/cli.mjs`): `modelSupportsZdr(id)` is
+* registry (`command-code@1.82.1` `dist/cli.mjs`): `modelSupportsZdr(id)` is
 * exactly `!nonZdrSet.has(canonicalize(id))`, and `knownModelSupportsZdr`
 * carries the same membership in the sibling route table — the UNION of both
 * is this set. Reading only the sibling route table dropped
@@ -581,6 +594,7 @@ const KNOWN_NON_ZDR_MODELS = /* @__PURE__ */ new Set([
 	"meta/muse-spark-1.2-contributor",
 	"meta/muse-spark-1.3",
 	"meta/muse-spark-1.3-contributor",
+	"meta/muse-wonder-1.0",
 	"minimax/minimax-m3-free",
 	"poolside/laguna-s-2.1-free",
 	"sakana/fugu-ultra",
@@ -697,18 +711,22 @@ 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: 76 of the 86
-* catalog ids as of 2026-09-30. Each row carries a comment naming the entry it
+* 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
 * came from.
 *
-* Two of those 76 rest on more than a vendor lookup. `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) is hand-recorded in the script's
-* `MANUAL_CEILINGS`: models.dev has no `ling-3.1` entry yet, so it takes the Ling
-* family's uniform ceiling. Both are still correctable — a stricter gateway costs
-* one refused request, and `streamRequest()` remembers what the endpoint said.
-*
-* The 10 unlisted models are ones no vendor publishes and the aggregators
+* 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
 * 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
@@ -724,6 +742,7 @@ const MODEL_OUTPUT_TOKEN_LIMITS = /* @__PURE__ */ new Map([
 	["claude-fable-5", 128e3],
 	["claude-fable-5-1", 128e3],
 	["claude-haiku-4-5-20251001", 64e3],
+	["claude-haiku-5-5", 128e3],
 	["claude-opus-4-7", 128e3],
 	["claude-opus-4-8", 128e3],
 	["claude-opus-5", 128e3],
@@ -740,6 +759,7 @@ const MODEL_OUTPUT_TOKEN_LIMITS = /* @__PURE__ */ new Map([
 	["google/gemini-3.6-flash", 65536],
 	["google/gemini-3.7-flash", 65536],
 	["google/gemini-3.8-flash", 65536],
+	["google/gemini-4-argon", 32768],
 	["gpt-5.3-codex", 128e3],
 	["gpt-5.4", 128e3],
 	["gpt-5.4-mini", 128e3],
@@ -759,6 +779,7 @@ const MODEL_OUTPUT_TOKEN_LIMITS = /* @__PURE__ */ new Map([
 	["meta/muse-spark-1.2-contributor", 131072],
 	["meta/muse-spark-1.3", 131072],
 	["meta/muse-spark-1.3-contributor", 131072],
+	["meta/muse-wonder-1.0", 131072],
 	["MiniMaxAI/MiniMax-M2.5", 131072],
 	["MiniMaxAI/MiniMax-M2.7", 131072],
 	["MiniMaxAI/MiniMax-M3", 512e3],
@@ -777,8 +798,6 @@ const MODEL_OUTPUT_TOKEN_LIMITS = /* @__PURE__ */ new Map([
 	["Qwen/Qwen3.8-Max", 131072],
 	["Qwen/Qwen3.8-Omni-Flash", 131072],
 	["sakana/fugu-ultra", 1e6],
-	["stealth/pixel-canary", 131072],
-	["stealth/space-bunny-alpha", 524288],
 	["stepfun/Step-3.5-Flash", 256e3],
 	["stepfun/Step-3.7-Flash", 256e3],
 	["stepfun/Step-5-Preview", 65536],
@@ -804,11 +823,13 @@ const MODEL_OUTPUT_TOKEN_LIMITS = /* @__PURE__ */ new Map([
 * `/docs/plans/max` and `/docs/resources/pricing-limits`). Each plan's model
 * list is a superset of the one below it: Go ⊂ GOAT ⊂ Pro ⊂ Provider/Max.
 * Models absent from every plan list (Claude Opus/Fable, Fugu Ultra) are
-* Provider-tier. Re-verified at command-code@1.79.2 (2026-10-09): 87 catalog
-* ids at 54/64/78/86 cumulative, a strict superset chain. Mistral Large 4
+* Provider-tier. Re-verified at command-code@1.82.1 (2026-10-14): 89 catalog
+* ids at 54/64/79/87/89 cumulative, a strict superset chain. Mistral Large 4
 * (1.75.0) was the only addition at the 1.77.0 check; the 1.78.0/1.79.0 pair
 * added Claude Haiku 5.5 at Pro and the free Glyph Cluster preview at Go.
-* 1.79.2 lowers Haiku 5.5's minimum plan to GOAT; every other tier is unchanged.
+* 1.79.2 lowers Haiku 5.5's minimum plan to GOAT; 1.80.0 adds Muse Wonder at
+* Max; 1.82.0 adds Gemini 4 Argon at Pro and nothing else moves. Every other
+* tier is unchanged.
 * The only removals since 1.49.0 are the two retired stealth previews (Pixel
 * Canary at 1.73.1, Space Bunny Alpha at 1.74.3). Per-entry tags below name the
 * releases that added or moved each row.
@@ -889,6 +910,7 @@ const KNOWN_PLANS = {
 	"claude-haiku-4-5-20251001": "pro",
 	"claude-sonnet-4-6": "pro",
 	"claude-sonnet-5": "pro",
+	"google/gemini-4-argon": "pro",
 	"google/gemini-3.1-flash-lite": "pro",
 	"google/gemini-3.5-flash": "pro",
 	"google/gemini-3.5-flash-lite": "pro",
@@ -908,7 +930,8 @@ const KNOWN_PLANS = {
 	"claude-opus-5": "provider",
 	"gpt-6-astra": "provider",
 	"sakana/fugu-ultra": "provider",
-	"gpt-6.1-sol": "max"
+	"gpt-6.1-sol": "max",
+	"meta/muse-wonder-1.0": "max"
 };
 /**
 * Whether a model is free (requests cost no credits), per the pricing page's
@@ -1497,8 +1520,8 @@ const MODEL_PRICE_ROWS = [
 		],
 		allowance: {
 			go: 10,
-			goat: 60,
-			pro: 70
+			goat: 100,
+			pro: 100
 		}
 	},
 	{
@@ -1611,6 +1634,19 @@ const MODEL_PRICE_ROWS = [
 			pro: 50
 		}
 	},
+	{
+		id: "gemini-4-argon",
+		rates: [
+			2,
+			10,
+			.1
+		],
+		allowance: {
+			go: 6,
+			goat: 20,
+			pro: 20
+		}
+	},
 	{
 		id: "glm-5",
 		rates: [
@@ -2318,6 +2354,19 @@ const MODEL_PRICE_ROWS = [
 			pro: 30
 		}
 	},
+	{
+		id: "muse-wonder-1.0",
+		rates: [
+			4,
+			20,
+			.2
+		],
+		allowance: {
+			go: 6,
+			goat: 20,
+			pro: 20
+		}
+	},
 	{
 		id: "nemotron-3-ultra",
 		rates: [
@@ -2764,10 +2813,10 @@ function recordOrEmpty(value) {
 }
 //#endregion
 //#region src/cost-projection.ts
-const record$2 = (v) => isRecord(v) ? v : void 0;
+const record$3 = (v) => isRecord(v) ? v : void 0;
 const finite = (v) => typeof v === "number" && Number.isFinite(v) && v >= 0;
-function tokensOf(value) {
-	const u = record$2(value);
+function tokensOf$1(value) {
+	const u = record$3(value);
 	if (!u || !finite(u.inputTokens) || !finite(u.outputTokens)) return void 0;
 	if (u.cacheReadTokens !== void 0 && !finite(u.cacheReadTokens) || u.cacheWriteTokens !== void 0 && !finite(u.cacheWriteTokens)) return void 0;
 	return {
@@ -2780,27 +2829,27 @@ function tokensOf(value) {
 /** Current v2 settlements store the last usage in their stream; v1 stores chunk events. */
 function usageOf(type, data) {
 	if (type === "assistant/chunk") {
-		const chunk = record$2(data.chunk);
-		return chunk?.type === "usage" ? tokensOf(chunk.usage) : void 0;
+		const chunk = record$3(data.chunk);
+		return chunk?.type === "usage" ? tokensOf$1(chunk.usage) : void 0;
 	}
 	if (type !== "assistant/message" && type !== "assistant/attempt") return void 0;
-	const direct = tokensOf(data.usage);
+	const direct = tokensOf$1(data.usage);
 	if (direct !== void 0) return direct;
 	if (!Array.isArray(data.stream)) return void 0;
 	for (let i = data.stream.length - 1; i >= 0; i--) {
-		const item = record$2(data.stream[i]);
-		const chunk = item?.type === "chunk" ? record$2(item.chunk) : void 0;
-		if (chunk?.type === "usage") return tokensOf(chunk.usage);
+		const item = record$3(data.stream[i]);
+		const chunk = item?.type === "chunk" ? record$3(item.chunk) : void 0;
+		if (chunk?.type === "usage") return tokensOf$1(chunk.usage);
 	}
 }
 function parseState(value) {
-	const s = record$2(value), facts = record$2(s?.facts);
+	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$2(v), t = record$2(g?.tokens);
+		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 sel = record$2(s.selection), last = record$2(s.last);
+	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;
 }
@@ -2821,10 +2870,10 @@ function createCostProjection(table = modelPriceTable()) {
 			}
 		}),
 		apply(state, event) {
-			const data = record$2(event.data);
+			const data = record$3(event.data);
 			if (!data) return state;
 			if (event.type === "request/header") {
-				const config = record$2(record$2(data.header)?.config);
+				const config = record$3(record$3(data.header)?.config);
 				return {
 					...state,
 					selection: config && typeof config.provider === "string" && typeof config.model === "string" ? {
@@ -2906,6 +2955,502 @@ function installCostProjection(ctx) {
 	});
 }
 //#endregion
+//#region src/analytics.ts
+/** 在宿主内折叠原始历史；每次请求保留计价时间和上下文，禁止使用已合并的会话成本桶。 */
+const record$2 = (v) => isRecord(v) ? v : void 0;
+const number = (v) => typeof v === "number" && Number.isFinite(v) && v >= 0;
+function tokensOf(v) {
+	const u = record$2(v);
+	if (!u || !number(u.inputTokens) || !number(u.outputTokens)) return void 0;
+	if (u.cacheReadTokens !== void 0 && !number(u.cacheReadTokens) || u.cacheWriteTokens !== void 0 && !number(u.cacheWriteTokens)) return void 0;
+	return {
+		uncachedInputTokens: u.inputTokens,
+		outputTokens: u.outputTokens,
+		cacheReadTokens: u.cacheReadTokens ?? 0,
+		cacheWriteTokens: u.cacheWriteTokens ?? 0
+	};
+}
+/** 同一次尝试的用量覆盖前值；重试开始才新建请求，继承前缀只恢复模型选择。 */
+function foldRequests(session, log) {
+	const facts = [];
+	let provider = "", model = "", current, own = false, started = false, listed = false, turn = -1, step = -1;
+	const begin = (event, data, retry) => {
+		turn = Number(data.turn);
+		step = Number(data.step);
+		own = event.seq >= log.inheritedEventCount;
+		started = true;
+		current = {
+			session,
+			model,
+			at: event.time,
+			retry,
+			outcome: "unknown"
+		};
+		listed = own && provider === "commandcode";
+		if (listed) facts.push(current);
+	};
+	for (const event of log.events) {
+		const d = record$2(event.data);
+		if (!d) continue;
+		if (event.type === "request/header") {
+			const config = record$2(record$2(d.header)?.config);
+			provider = typeof config?.provider === "string" ? config.provider : "";
+			model = typeof config?.model === "string" ? config.model : "";
+			if (current && current.outcome === "unknown") {
+				current.model = model;
+				if (own && provider === "commandcode" && !listed) {
+					facts.push(current);
+					listed = true;
+				}
+				if (provider !== "commandcode" && listed) {
+					facts.pop();
+					listed = false;
+				}
+			}
+			continue;
+		}
+		if (event.type === "step/start" || event.type === "llm/retry-started") {
+			begin(event, d, event.type === "llm/retry-started");
+			continue;
+		}
+		if (!current && [
+			"assistant/message",
+			"assistant/attempt",
+			"assistant/chunk"
+		].includes(event.type)) {
+			begin(event, d, false);
+			started = false;
+		}
+		if (!current) continue;
+		if (event.type === "turn/end" && d.turn === turn) {
+			const reason = record$2(d.reason)?.kind;
+			if (current.outcome === "unknown") {
+				if (reason === "error") current.outcome = "failed";
+				else if (reason === "aborted") current.outcome = "cancelled";
+			}
+			current = void 0;
+			continue;
+		}
+		if (d.turn !== turn || d.step !== step) continue;
+		if (event.type === "llm/retry") {
+			current.outcome = "failed";
+			continue;
+		}
+		if (![
+			"assistant/message",
+			"assistant/attempt",
+			"assistant/chunk"
+		].includes(event.type)) continue;
+		const chunk = record$2(d.chunk);
+		let tokens = tokensOf(d.usage);
+		if (event.type === "assistant/chunk" && chunk?.type === "usage") tokens = tokensOf(chunk.usage);
+		if (tokens === void 0 && Array.isArray(d.stream)) for (let i = d.stream.length - 1; i >= 0; i--) {
+			const item = record$2(d.stream[i]), c = item?.type === "chunk" ? record$2(item.chunk) : void 0;
+			if (c?.type === "usage") {
+				tokens = tokensOf(c.usage);
+				if (tokens) break;
+			}
+		}
+		if (tokens) current.tokens = tokens;
+		if (event.type === "assistant/message") current.outcome = d.interrupted === true ? "cancelled" : "completed";
+		if (started && Array.isArray(d.stream) && event.time >= current.at) {
+			current.latencyMs = event.time - current.at;
+			const first = assistantStreamFirstTokenTime(d.stream);
+			if (first !== void 0 && first >= current.at && first <= event.time) {
+				current.ttftMs = first - current.at;
+				if (current.tokens) current.decodeMs = event.time - first;
+			}
+		}
+	}
+	return facts;
+}
+const emptyMetrics = () => ({
+	sessions: 0,
+	requests: 0,
+	completed: 0,
+	failed: 0,
+	cancelled: 0,
+	unknown: 0,
+	retries: 0,
+	usageSamples: 0,
+	pricedRequests: 0,
+	cost: 0,
+	tokens: zeroCostTokens(),
+	latencySamples: 0,
+	latencyMs: 0,
+	ttftSamples: 0,
+	ttftMs: 0,
+	decodeSamples: 0,
+	decodeMs: 0,
+	decodeTokens: 0
+});
+function priceRequest(f, id, table, prices) {
+	if (!f.tokens) return void 0;
+	const price = prices.get(id);
+	if (!price) return void 0;
+	if (price.free === true) return 0;
+	const context = f.tokens.uncachedInputTokens + f.tokens.cacheReadTokens + f.tokens.cacheWriteTokens;
+	const r = requestRates(price, f.at, context, table);
+	if (f.tokens.cacheWriteTokens > 0 && r.cacheWriteCost === void 0) return void 0;
+	return (f.tokens.uncachedInputTokens * r.inputCost + f.tokens.outputTokens * r.outputCost + f.tokens.cacheReadTokens * r.cacheReadCost + f.tokens.cacheWriteTokens * (r.cacheWriteCost ?? 0)) / 1e6;
+}
+function aggregate(facts, table, prices) {
+	const m = emptyMetrics();
+	m.sessions = new Set(facts.map((f) => f.session)).size;
+	for (const f of facts) {
+		m.requests++;
+		m[f.outcome]++;
+		if (f.retry) m.retries++;
+		if (f.tokens) {
+			m.usageSamples++;
+			for (const k of COST_TOKEN_KEYS) m.tokens[k] += f.tokens[k];
+		}
+		const cost = priceRequest(f, f.model, table, prices);
+		if (cost !== void 0) {
+			m.pricedRequests++;
+			m.cost += cost;
+		}
+		if (f.latencyMs !== void 0) {
+			m.latencySamples++;
+			m.latencyMs += f.latencyMs;
+		}
+		if (f.ttftMs !== void 0) {
+			m.ttftSamples++;
+			m.ttftMs += f.ttftMs;
+		}
+		if (f.decodeMs !== void 0 && f.decodeMs > 0 && f.tokens) {
+			m.decodeSamples++;
+			m.decodeMs += f.decodeMs;
+			m.decodeTokens += f.tokens.outputTokens;
+		}
+	}
+	return m;
+}
+function buildAnalytics(facts, query, table, catalog, now = Date.now()) {
+	const selected = query.mode === "range" ? facts.filter((f) => f.at >= query.from && f.at < query.to) : facts;
+	const prices = /* @__PURE__ */ new Map();
+	for (const p of table.models) for (const id of [p.id, p.slug]) if (!prices.has(id)) prices.set(id, p);
+	const totals = aggregate(selected, table, prices);
+	const models = [...new Set(selected.map((f) => f.model))].map((id) => ({
+		id,
+		...aggregate(selected.filter((f) => f.model === id), table, prices)
+	})).sort((a, b) => b.cost - a.cost || a.id.localeCompare(b.id));
+	const scenarios = catalog.map((model) => {
+		let cost = 0, pricedRequests = 0, excludedRequests = 0, unknownRequests = 0;
+		for (const f of selected) {
+			if (!f.tokens) {
+				unknownRequests++;
+				continue;
+			}
+			const context = f.tokens.uncachedInputTokens + f.tokens.cacheReadTokens + f.tokens.cacheWriteTokens;
+			if (model.contextWindow !== null && context + f.tokens.outputTokens > model.contextWindow) {
+				excludedRequests++;
+				continue;
+			}
+			const priced = priceRequest(f, model.id, table, prices);
+			if (priced === void 0) unknownRequests++;
+			else {
+				cost += priced;
+				pricedRequests++;
+			}
+		}
+		return {
+			...model,
+			cost,
+			pricedRequests,
+			excludedRequests,
+			unknownRequests
+		};
+	});
+	const elapsed = query.mode === "range" ? Math.min(query.to, now) - query.from : 0;
+	const monthlyEstimate = elapsed >= 864e5 && totals.requests > 0 && totals.pricedRequests === totals.requests ? totals.cost * (2592e6 / elapsed) : null;
+	return {
+		generatedAt: now,
+		pricingKey: pricingKey(table),
+		query,
+		status: "ready",
+		scannedSessions: 0,
+		matchedSessions: totals.sessions,
+		unreadableSessions: 0,
+		skippedSessions: 0,
+		truncated: false,
+		catalogUnavailable: false,
+		totals,
+		models,
+		scenarios,
+		monthlyEstimate
+	};
+}
+//#endregion
+//#region src/wire-shared.ts
+/** The npm package identity every contribution and descriptor claims. */
+const REMOTE_PACKAGE = "@mars-sea/dsh-commandcode-provider";
+/** The Cordis service key the Gateway resolves every Command Code Remote from. */
+const REMOTE_SERVICE = "commandcodeUsage";
+/** The wire namespace all Command Code endpoints share. */
+const REMOTE_NAMESPACE = "commandcode";
+/**
+* Build the validator helpers one endpoint uses. `prefix` names the endpoint
+* in the rejection message (e.g. `commandcode/report result:`), so each wire
+* file keeps its own diagnostic phrasing while sharing the helper bodies.
+*
+* The helpers return the reject call directly in the failure branch: since
+* `reject` is typed `never`, the ternary's union collapses to the success type
+* without relying on TypeScript's control-flow analysis of a never-returning
+* call (which only recognizes function declarations, not the destructured
+* arrow `reject` callers receive from this factory). Callers that need the
+* narrowing in an `if` must therefore assign inside the positive branch.
+*/
+function makeBoundaryValidator(prefix) {
+	const reject = (field) => {
+		throw new TypeError(`${prefix} invalid ${field}`);
+	};
+	const record = (value, field) => typeof value === "object" && value !== null && !Array.isArray(value) ? value : reject(field);
+	const stringField = (source, key, field) => typeof source[key] === "string" ? source[key] : reject(field);
+	const numberField = (source, key, field) => typeof source[key] === "number" && Number.isFinite(source[key]) ? source[key] : reject(field);
+	const booleanField = (source, key, field) => typeof source[key] === "boolean" ? source[key] : reject(field);
+	return {
+		reject,
+		record,
+		stringField,
+		numberField,
+		booleanField
+	};
+}
+/** Use the same strict-codec shape for invocation parameters and results. */
+function makeStrictCodec(typeSymbol, schema) {
+	return {
+		mode: "strict",
+		typeSymbol,
+		create: () => schema
+	};
+}
+/**
+* Build one strict invocation descriptor. Every Command Code Remote shares the
+* `commandcode` namespace, the `commandcodeUsage` service, and a strict result
+* — only the endpoint, method, result type symbol, and schema differ.
+*/
+function makeRemoteDescriptor(endpoint, method, typeSymbol, schema) {
+	return {
+		id: `${REMOTE_PACKAGE}#${endpoint}`,
+		service: REMOTE_SERVICE,
+		namespace: REMOTE_NAMESPACE,
+		method,
+		invocation: { kind: "direct" },
+		parameters: [],
+		result: makeStrictCodec(typeSymbol, schema)
+	};
+}
+//#endregion
+//#region src/analytics-wire.ts
+const g = makeBoundaryValidator("commandcode/analytics");
+const nonnegative = (s, k) => {
+	const n = g.numberField(s, k, k);
+	return n >= 0 ? n : g.reject(k);
+};
+const count = (s, k) => {
+	const n = nonnegative(s, k);
+	return Number.isSafeInteger(n) ? n : g.reject(k);
+};
+function parseAnalyticsQuery(value) {
+	const s = g.record(value, "query");
+	if (s.mode === "recent") {
+		const limit = count(s, "limit");
+		if (limit < 1 || limit > 100) return g.reject("limit");
+		return {
+			mode: "recent",
+			limit
+		};
+	}
+	if (s.mode !== "range") return g.reject("mode");
+	const from = count(s, "from"), to = count(s, "to");
+	if (to <= from || to - from > 316224e5 || to > 864e13) return g.reject("range");
+	return {
+		mode: "range",
+		from,
+		to
+	};
+}
+const counts$2 = [
+	"sessions",
+	"requests",
+	"completed",
+	"failed",
+	"cancelled",
+	"unknown",
+	"retries",
+	"usageSamples",
+	"pricedRequests",
+	"latencySamples",
+	"ttftSamples",
+	"decodeSamples"
+];
+const quantities = [
+	"cost",
+	"latencyMs",
+	"ttftMs",
+	"decodeMs",
+	"decodeTokens"
+];
+const tokenKeys = [
+	"uncachedInputTokens",
+	"outputTokens",
+	"cacheReadTokens",
+	"cacheWriteTokens"
+];
+function metrics(value) {
+	const s = g.record(value, "metrics"), result = {};
+	for (const k of counts$2) result[k] = count(s, k);
+	for (const k of quantities) result[k] = nonnegative(s, k);
+	const tokens = g.record(s.tokens, "tokens");
+	result.tokens = Object.fromEntries(tokenKeys.map((k) => [k, nonnegative(tokens, k)]));
+	if (result.completed + result.failed + result.cancelled + result.unknown !== result.requests || result.pricedRequests > result.usageSamples || result.usageSamples > result.requests || result.retries > result.requests || result.ttftSamples > result.requests || result.latencySamples > result.requests || result.decodeSamples > result.requests) return g.reject("coverage");
+	return result;
+}
+const nullable = (s, k) => s[k] === null ? null : nonnegative(s, k);
+function parseAnalyticsReport(value) {
+	const s = g.record(value, "report");
+	if (s.status !== "ready" && s.status !== "unavailable") return g.reject("status");
+	if (!Array.isArray(s.models) || !Array.isArray(s.scenarios) || s.models.length > 1e3 || s.scenarios.length > 1e3) return g.reject("models");
+	return {
+		generatedAt: count(s, "generatedAt"),
+		pricingKey: count(s, "pricingKey"),
+		query: parseAnalyticsQuery(s.query),
+		status: s.status,
+		scannedSessions: count(s, "scannedSessions"),
+		matchedSessions: count(s, "matchedSessions"),
+		unreadableSessions: count(s, "unreadableSessions"),
+		skippedSessions: count(s, "skippedSessions"),
+		truncated: g.booleanField(s, "truncated", "truncated"),
+		catalogUnavailable: g.booleanField(s, "catalogUnavailable", "catalogUnavailable"),
+		totals: metrics(s.totals),
+		monthlyEstimate: nullable(s, "monthlyEstimate"),
+		models: s.models.map((value) => {
+			const m = g.record(value, "model");
+			return {
+				...metrics(m),
+				id: g.stringField(m, "id", "id")
+			};
+		}),
+		scenarios: s.scenarios.map((value) => {
+			const m = g.record(value, "scenario");
+			if (!Array.isArray(m.efforts) || !m.efforts.every((v) => typeof v === "string")) return g.reject("efforts");
+			return {
+				id: g.stringField(m, "id", "id"),
+				name: g.stringField(m, "name", "name"),
+				contextWindow: nullable(m, "contextWindow"),
+				images: g.booleanField(m, "images", "images"),
+				efforts: m.efforts,
+				cost: nonnegative(m, "cost"),
+				pricedRequests: count(m, "pricedRequests"),
+				excludedRequests: count(m, "excludedRequests"),
+				unknownRequests: count(m, "unknownRequests")
+			};
+		})
+	};
+}
+const ANALYTICS_DESCRIPTOR = {
+	...makeRemoteDescriptor("commandcode/analytics", "analytics", `${REMOTE_PACKAGE}#AnalyticsReport`, { parse: parseAnalyticsReport }),
+	parameters: [{
+		name: "query",
+		wire: "query",
+		source: "json",
+		codec: makeStrictCodec(`${REMOTE_PACKAGE}#AnalyticsQuery`, { parse: parseAnalyticsQuery })
+	}]
+};
+/** 逐会话失败隔离；上限和遗漏计数会返回界面，部分结果不能伪装成完整历史。 */
+async function readAnalyticsHistory(history, input, catalog, signal) {
+	const query = parseAnalyticsQuery(input), table = modelPriceTable();
+	if (!history) return {
+		...buildAnalytics([], query, table, []),
+		status: "unavailable"
+	};
+	const records = await history.listSessions(signal);
+	const facts = [];
+	let scannedSessions = 0, unreadableSessions = 0, skippedSessions = 0, events = 0, truncated = records.length > 500;
+	for (const entry of records.slice(0, 500)) {
+		if (signal?.aborted) throw new Error("模型分析已取消");
+		if (events >= 2e5) {
+			truncated = true;
+			break;
+		}
+		scannedSessions++;
+		try {
+			const log = await history.readSession(entry.header.id);
+			if (!Number.isSafeInteger(log.inheritedEventCount) || log.inheritedEventCount < 0 || log.inheritedEventCount > log.events.length) {
+				skippedSessions++;
+				continue;
+			}
+			events += log.events.length;
+			if (events > 2e5) {
+				skippedSessions++;
+				truncated = true;
+				break;
+			}
+			facts.push(...foldRequests(entry.header.id, log));
+		} catch {
+			unreadableSessions++;
+		}
+	}
+	let selected = facts;
+	if (query.mode === "recent") {
+		const latest = /* @__PURE__ */ new Map();
+		for (const f of facts) latest.set(f.session, Math.max(latest.get(f.session) ?? 0, f.at));
+		const ids = new Set([...latest].sort((a, b) => b[1] - a[1] || a[0].localeCompare(b[0])).slice(0, query.limit).map(([id]) => id));
+		selected = facts.filter((f) => ids.has(f.session));
+	}
+	let models = [], catalogUnavailable = false;
+	if (signal?.aborted) throw new Error("模型分析已取消");
+	try {
+		models = await catalog();
+	} catch {
+		catalogUnavailable = true;
+	}
+	const report = buildAnalytics(selected, query, table, models);
+	return {
+		...report,
+		scannedSessions,
+		unreadableSessions,
+		skippedSessions,
+		truncated,
+		catalogUnavailable,
+		monthlyEstimate: truncated || unreadableSessions > 0 || skippedSessions > 0 ? null : report.monthlyEstimate
+	};
+}
+/** 用反射和可选 fiber 对接，保持未挂载历史服务的无界面/TUI 配置可用。 */
+function createAnalyticsReader(ctx, adapter) {
+	let history, busy = false;
+	ctx.inject(["sessionQuery"], (c) => {
+		const service = c.get("sessionQuery");
+		if (typeof service?.listSessions !== "function" || typeof service.readSession !== "function") return;
+		history = service;
+		c.effect(() => () => {
+			if (history === service) history = void 0;
+		}, "commandcode analytics history");
+	});
+	return async (query, signal) => {
+		if (busy) throw new Error("模型分析正在读取历史，请稍后刷新");
+		busy = true;
+		try {
+			return await readAnalyticsHistory(history, query, async () => {
+				const available = await adapter.listModels("commandcode");
+				return Promise.all(available.map(async (m) => {
+					const resolved = await adapter.resolveModel("commandcode", m.id);
+					return {
+						id: m.id,
+						name: m.name.replace(/ \(CC\)$/, ""),
+						contextWindow: resolved.context?.contextWindow ?? null,
+						images: resolved.inputModalities?.includes("image") ?? false,
+						efforts: [...KNOWN_EFFORTS[m.id] ?? []]
+					};
+				}));
+			}, signal);
+		} finally {
+			busy = false;
+		}
+	};
+}
+//#endregion
 //#region src/response-encoding.ts
 /**
 * Ask Command Code for plain response bodies. A DSH host using an incompatible
@@ -5357,7 +5902,7 @@ 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 = "1.79.2";
+const COMMAND_CODE_CLI_VERSION = "1.82.1";
 const DEFAULT_API_BASE = "https://api.commandcode.ai";
 /**
 * The output budget this bundle asks for, and the ceiling it will never exceed.
@@ -6433,6 +6978,70 @@ function parseAccountIdentity(whoami) {
 		orgId: orgData === void 0 ? void 0 : stringValue(orgData.id)
 	};
 }
+const JSON_READ_ERROR_NAMES = /* @__PURE__ */ new Set([
+	"Error",
+	"TypeError",
+	"SyntaxError",
+	"AbortError",
+	"TimeoutError"
+]);
+const JSON_READ_ERROR_CODES = /* @__PURE__ */ new Set([
+	"ECONNRESET",
+	"ETIMEDOUT",
+	"EPIPE",
+	"ERR_STREAM_PREMATURE_CLOSE",
+	"ERR_INVALID_STATE",
+	"UND_ERR_SOCKET",
+	"UND_ERR_BODY_TIMEOUT",
+	"UND_ERR_ABORTED",
+	"Z_DATA_ERROR",
+	"Z_BUF_ERROR"
+]);
+const JSON_CONTENT_TYPES = /* @__PURE__ */ new Set([
+	"application/json",
+	"application/problem+json",
+	"text/html",
+	"text/plain",
+	"application/octet-stream"
+]);
+const JSON_CONTENT_ENCODINGS = /* @__PURE__ */ new Set([
+	"identity",
+	"gzip",
+	"x-gzip",
+	"deflate",
+	"br",
+	"zstd"
+]);
+/** 原异常消息可能包含正文、账号或地址，只保留白名单标识，并有界读取 cause 链。 */
+function jsonReadErrorIdentifiers(error) {
+	const identifiers = /* @__PURE__ */ new Set();
+	const seen = /* @__PURE__ */ new Set();
+	let current = error;
+	for (let depth = 0; depth < 4 && isRecord(current) && !seen.has(current); depth++) {
+		seen.add(current);
+		if (typeof current.name === "string" && JSON_READ_ERROR_NAMES.has(current.name)) identifiers.add(current.name);
+		if (typeof current.code === "string" && JSON_READ_ERROR_CODES.has(current.code)) identifiers.add(current.code);
+		current = current.cause;
+	}
+	return [...identifiers];
+}
+/** 响应头只作为线索：不据此拒绝可解析正文，也不把未经筛选的头值放入账号报告。 */
+function jsonResponseMetadata(response) {
+	const metadata = [];
+	const contentType = response.headers.get("content-type")?.split(";")[0]?.trim().toLowerCase();
+	if (contentType) metadata.push(`内容类型：${JSON_CONTENT_TYPES.has(contentType) ? contentType : "其他"}`);
+	const encoding = response.headers.get("content-encoding")?.trim().toLowerCase();
+	if (encoding) metadata.push(`内容编码：${JSON_CONTENT_ENCODINGS.has(encoding) ? encoding : "其他"}`);
+	return metadata;
+}
+function jsonResponseFailureDetail(response, failure) {
+	let detail;
+	if (failure.stage === "read") detail = `${failure.timedOut ? "读取响应正文超时" : "响应正文读取中断"}${failure.identifiers.length === 0 ? "" : `（${failure.identifiers.join(" / ")}）`}`;
+	else if (failure.stage === "parse") detail = "响应正文不是有效的 JSON（结构化数据解析失败；SyntaxError）";
+	else detail = `响应 JSON 顶层应为对象，实际为${failure.shape}`;
+	const metadata = jsonResponseMetadata(response);
+	return `${detail}${metadata.length === 0 ? "" : `；${metadata.join("；")}`}`;
+}
 function classifyTotalFailure(failures, failed) {
 	if (failures.length !== USAGE_ENDPOINT_COUNT) return void 0;
 	if (failed.every((item) => item.kind === "http" && item.status === 401)) return "invalid-key";
@@ -7006,34 +7615,59 @@ var CommandCodeAdapter = class CommandCodeAdapter extends LlmAdapter {
 	/**
 	* Fetch one JSON GET and return its HTTP status with a parsed record. The
 	* catalog, billing probe, and usage report each apply their own failure
-	* policy. Non-2xx and invalid JSON bodies come back without a record; only
-	* a fetch failure propagates to the caller.
+	* policy. Non-2xx and unusable bodies come back without a record; only
+	* a failure before receiving the response propagates to the caller.
+	* 正文读取、JSON 解析与顶层结构分别诊断，不能把已收到响应后的断流当成格式错误。
 	*
 	* `timeoutMs` is the caller's budget, unless it provides its own signal.
 	* The usage report passes the connection's request budget rather than the
 	* catalog/picker's shorter default.
 	*/
 	async fetchJson(url, headers, timeoutMs = MODELS_TIMEOUT_MS, signal) {
+		const requestSignal = signal ?? AbortSignal.timeout(timeoutMs);
 		const response = await this.fetchImpl(url, {
 			headers,
-			signal: signal ?? AbortSignal.timeout(timeoutMs)
+			signal: requestSignal
 		});
 		if (!response.ok) return { status: response.status };
+		let text;
+		try {
+			text = await response.text();
+		} catch (error) {
+			const identifiers = jsonReadErrorIdentifiers(error);
+			if (requestSignal.aborted) {
+				for (const identifier of jsonReadErrorIdentifiers(requestSignal.reason)) if (!identifiers.includes(identifier)) identifiers.push(identifier);
+			}
+			const timedOut = identifiers.includes("TimeoutError") || identifiers.includes("ETIMEDOUT") || identifiers.includes("UND_ERR_BODY_TIMEOUT");
+			return {
+				status: response.status,
+				responseFailure: jsonResponseFailureDetail(response, {
+					stage: "read",
+					identifiers,
+					timedOut
+				})
+			};
+		}
 		let parsed;
 		try {
-			parsed = await response.json();
+			parsed = JSON.parse(text);
 		} catch {
 			return {
 				status: response.status,
-				invalidResponse: true
+				responseFailure: jsonResponseFailureDetail(response, { stage: "parse" })
 			};
 		}
-		return isRecord(parsed) ? {
+		if (isRecord(parsed)) return {
 			status: response.status,
 			record: parsed
-		} : {
+		};
+		const shape = parsed === null ? "空值" : Array.isArray(parsed) ? "数组" : typeof parsed === "string" ? "字符串" : typeof parsed === "number" ? "数字" : "布尔值";
+		return {
 			status: response.status,
-			invalidResponse: true
+			responseFailure: jsonResponseFailureDetail(response, {
+				stage: "shape",
+				shape
+			})
 		};
 	}
 	/**
@@ -7223,10 +7857,10 @@ var CommandCodeAdapter = class CommandCodeAdapter extends LlmAdapter {
 		const failed = [];
 		const getJson = async (path) => {
 			try {
-				const { status, record, invalidResponse } = await this.fetchJson(`${base}${path}`, headers, connection.requestTimeoutMs);
+				const { status, record, responseFailure } = await this.fetchJson(`${base}${path}`, headers, connection.requestTimeoutMs);
 				if (record === void 0) {
-					failures.push(`${path}: HTTP ${status}${invalidResponse ? " returned an unreadable or invalid JSON body" : ""}`);
-					failed.push(invalidResponse ? {
+					failures.push(`${path}: HTTP ${status}${responseFailure ? `；${responseFailure}` : ""}`);
+					failed.push(responseFailure ? {
 						kind: "invalid-response",
 						status
 					} : {
@@ -7589,7 +8223,7 @@ const commandcodeCommand = {
 		unconfigured: "  (未配置 API 密钥)",
 		blockedInvalidKey: "⛔ API 密钥无效或已过期 — 服务端拒绝了全部请求（401），请检查该账户的密钥配置",
 		blockedServiceUnavailable: "⚠️ Command Code 服务暂时不可用（5xx），稍后重试",
-		blockedInvalidResponse: "⚠️ Command Code 返回了无法读取的响应 — 请检查宿主的 HTTP 代理或响应解压配置",
+		blockedInvalidResponse: "Command Code 响应正文读取或解析失败，请根据下方详情排查并重试",
 		blockedNetwork: "⚠️ 无法连接 Command Code 服务 — 请检查网络或 API 地址",
 		planLine: "  📦 套餐    {name}{status}{period}",
 		planPeriodSuffix: " · 账期截止 {date}",
@@ -7620,7 +8254,7 @@ const commandcodeCommand = {
 		unconfigured: "  (no API key configured)",
 		blockedInvalidKey: "⛔ API key invalid or expired — the server rejected every request (401); check the key configured for this account",
 		blockedServiceUnavailable: "⚠️ Command Code service temporarily unavailable (5xx); try again later",
-		blockedInvalidResponse: "⚠️ Command Code returned an unreadable response — check the host HTTP proxy or response decoding",
+		blockedInvalidResponse: "Reading or parsing the Command Code response failed; check the details below and retry",
 		blockedNetwork: "⚠️ could not reach the Command Code service — check your network or the API base setting",
 		planLine: "  📦 Plan     {name}{status}{period}",
 		planPeriodSuffix: " · period ends {date}",
@@ -7788,66 +8422,6 @@ function applyCommands(ctx, deps) {
 	ctx.commands.register(commandDefinition(deps));
 }
 //#endregion
-//#region src/wire-shared.ts
-/** The npm package identity every contribution and descriptor claims. */
-const REMOTE_PACKAGE = "@mars-sea/dsh-commandcode-provider";
-/** The Cordis service key the Gateway resolves every Command Code Remote from. */
-const REMOTE_SERVICE = "commandcodeUsage";
-/** The wire namespace all Command Code endpoints share. */
-const REMOTE_NAMESPACE = "commandcode";
-/**
-* Build the validator helpers one endpoint uses. `prefix` names the endpoint
-* in the rejection message (e.g. `commandcode/report result:`), so each wire
-* file keeps its own diagnostic phrasing while sharing the helper bodies.
-*
-* The helpers return the reject call directly in the failure branch: since
-* `reject` is typed `never`, the ternary's union collapses to the success type
-* without relying on TypeScript's control-flow analysis of a never-returning
-* call (which only recognizes function declarations, not the destructured
-* arrow `reject` callers receive from this factory). Callers that need the
-* narrowing in an `if` must therefore assign inside the positive branch.
-*/
-function makeBoundaryValidator(prefix) {
-	const reject = (field) => {
-		throw new TypeError(`${prefix} invalid ${field}`);
-	};
-	const record = (value, field) => typeof value === "object" && value !== null && !Array.isArray(value) ? value : reject(field);
-	const stringField = (source, key, field) => typeof source[key] === "string" ? source[key] : reject(field);
-	const numberField = (source, key, field) => typeof source[key] === "number" && Number.isFinite(source[key]) ? source[key] : reject(field);
-	const booleanField = (source, key, field) => typeof source[key] === "boolean" ? source[key] : reject(field);
-	return {
-		reject,
-		record,
-		stringField,
-		numberField,
-		booleanField
-	};
-}
-/** Use the same strict-codec shape for invocation parameters and results. */
-function makeStrictCodec(typeSymbol, schema) {
-	return {
-		mode: "strict",
-		typeSymbol,
-		create: () => schema
-	};
-}
-/**
-* Build one strict invocation descriptor. Every Command Code Remote shares the
-* `commandcode` namespace, the `commandcodeUsage` service, and a strict result
-* — only the endpoint, method, result type symbol, and schema differ.
-*/
-function makeRemoteDescriptor(endpoint, method, typeSymbol, schema) {
-	return {
-		id: `${REMOTE_PACKAGE}#${endpoint}`,
-		service: REMOTE_SERVICE,
-		namespace: REMOTE_NAMESPACE,
-		method,
-		invocation: { kind: "direct" },
-		parameters: [],
-		result: makeStrictCodec(typeSymbol, schema)
-	};
-}
-//#endregion
 //#region src/usage-wire.ts
 /** The npm package identity both contribution registrations claim. */
 const USAGE_REMOTE_PACKAGE = REMOTE_PACKAGE;
@@ -8304,6 +8878,10 @@ var CommandCodeUsageService = class extends TypertRemoteService {
 	async prices() {
 		return (this.deps.prices ?? modelPriceTable)();
 	}
+	async analytics(query) {
+		if (!this.deps.analytics) throw new Error("宿主不支持模型分析，请更新插件");
+		return this.deps.analytics(query, this.ctx.invocation?.signal);
+	}
 	/**
 	* Start (or rejoin) a browser-login attempt and return its fresh status —
 	* `waiting` carrying the Studio URL. Rejects when the flow cannot start
@@ -8393,6 +8971,7 @@ function applyUsageRemote(ctx, deps) {
 				...USAGE_HOST_CONTRIBUTION.invocations,
 				MODELS_DESCRIPTOR,
 				PRICES_DESCRIPTOR,
+				ANALYTICS_DESCRIPTOR,
 				...LOGIN_HOST_CONTRIBUTION.invocations,
 				...ENROLLMENT_DESCRIPTORS
 			]
@@ -10338,7 +10917,8 @@ function apply(ctx, config) {
 		reports: usageReports,
 		login: loginFlow,
 		enrollment,
-		listModels: catalogForEditors
+		listModels: catalogForEditors,
+		analytics: createAnalyticsReader(ctx, adapter)
 	});
 	const searchSelection = commandCodeSearchSelection();
 	const applySearchSelection = (enabled) => {