@goodandready/dsh-opencode-provider 0.5.18 → 0.5.20

package/CHANGELOG.md

@@ -1,5 +1,20 @@
 # Changelog
 
+## 0.5.20
+
+- **CI/reproducibility**: regenerated `package-lock.json` with the normal npm resolver; clean `npm ci --ignore-scripts` now succeeds and `npm run check` passes.
+- **Streaming correctness**: SSE parsing now normalizes CRLF/CR across chunk boundaries; all four transports reject streams that end without their protocol terminal event. Pre-aborted caller signals are rejected before provider dispatch.
+- **Concurrent settings**: settings saves carry the draft revision and use the settings compare-and-swap; revision conflicts return 409 instead of overwriting another page's changes.
+- **UI safety**: removing an account clears unsaved per-row key drafts; clearing a credential asks for confirmation; update checks distinguish registry failure from “up to date”; service status never reports healthy before a successful check.
+- **Images**: request budget counts repeated image occurrences and base64 overhead; offloaded images no longer require the attachment service.
+- **Docs/tests**: added project `AGENTS.md`; removed the unsafe local engine-matrix script pending the approved DSH test contour (`#55`).
+
+## 0.5.19
+
+- **Per-model soft cooldown**: quota/free-tier/rate-limit rejections now cool only that key+model (`quota` 5m, `rate-limit` 60s); other models on the same account remain usable. Invalid credentials still cool the account. If every key is cooling, the request degrades to the normal candidate list instead of failing before dispatch.
+- **Service status**: `/state` exposes the usage check time, latency and error, plus catalog source (`live`/`snapshot`), count and fetch time. Settings shows a compact Service status card.
+- 105 tests; engine matrix/install matrix remain green.
+
 ## 0.5.18
 
 - **Compatibility range**: `engines.dsh`, DSH peer ranges and `dsh.compatibility.dsh` now use `>=0.2.1-rc.2 <0.2.2-0`; `dshReleases` keeps the tested rc.2/rc.3/rc.4 records. Future `0.2.1` RCs are admitted automatically; the next minor remains excluded.

package/README.md

@@ -130,12 +130,12 @@ npm pack
 Install the **packed tarball** into a profile:
 
 ```bash
-dsh plugin --profile web add /absolute/path/to/goodandready-dsh-opencode-provider-0.5.18.tgz
+dsh plugin --profile web add /absolute/path/to/goodandready-dsh-opencode-provider-0.5.20.tgz
 # DSH Desktop:
-dsh plugin --profile desktop add /absolute/path/to/goodandready-dsh-opencode-provider-0.5.18.tgz
+dsh plugin --profile desktop add /absolute/path/to/goodandready-dsh-opencode-provider-0.5.20.tgz
 
 # Or install the published package:
-dsh plugin --profile web add @goodandready/dsh-opencode-provider@0.5.18
+dsh plugin --profile web add @goodandready/dsh-opencode-provider@0.5.20
 ```
 
 Restart dsh (or DSH Desktop), then pick **OpenCode (plugin)** or **OpenCode Go (plugin)** in the model picker.

package/README.ru.md

@@ -129,12 +129,12 @@ npm pack
 Установка из **упакованного tarball**:
 
 ```bash
-dsh plugin --profile web add /absolute/path/to/goodandready-dsh-opencode-provider-0.5.18.tgz
+dsh plugin --profile web add /absolute/path/to/goodandready-dsh-opencode-provider-0.5.20.tgz
 # DSH Desktop:
-dsh plugin --profile desktop add /absolute/path/to/goodandready-dsh-opencode-provider-0.5.18.tgz
+dsh plugin --profile desktop add /absolute/path/to/goodandready-dsh-opencode-provider-0.5.20.tgz
 
 # Либо установка опубликованного пакета из npm:
-dsh plugin --profile web add @goodandready/dsh-opencode-provider@0.5.18
+dsh plugin --profile web add @goodandready/dsh-opencode-provider@0.5.20
 ```
 
 После установки перезапустите dsh (или DSH Desktop) и выберите **OpenCode (plugin)** или **OpenCode Go (plugin)** в списке моделей.

package/README.zh.md

@@ -124,12 +124,12 @@ npm pack
 通过**打包文件 (.tgz)**安装：
 
 ```bash
-dsh plugin --profile web add /absolute/path/to/goodandready-dsh-opencode-provider-0.5.18.tgz
+dsh plugin --profile web add /absolute/path/to/goodandready-dsh-opencode-provider-0.5.20.tgz
 # DSH 桌面端：
-dsh plugin --profile desktop add /absolute/path/to/goodandready-dsh-opencode-provider-0.5.18.tgz
+dsh plugin --profile desktop add /absolute/path/to/goodandready-dsh-opencode-provider-0.5.20.tgz
 
 # 或从 npm 安装已发布版本：
-dsh plugin --profile web add @goodandready/dsh-opencode-provider@0.5.18
+dsh plugin --profile web add @goodandready/dsh-opencode-provider@0.5.20
 ```
 
 安装完成后重启 dsh 或 DSH 桌面端，在模型选择器中选择 **OpenCode (plugin)** 或 **OpenCode Go (plugin)** 即可。

package/lib/client.js

@@ -6,7 +6,7 @@ window.__ModuleLoader__.load({
 		Object.defineProperty(exports, Symbol.toStringTag, { value: "Module" });
 		let react = require("react");
 		let react_jsx_runtime = require("react/jsx-runtime");
-		//#region src/client/locales.ts
+		// src/client/locales.ts
 		const NS = "settings.opencode";
 		const dictionaries = {
 			en: {
@@ -25,11 +25,23 @@ window.__ModuleLoader__.load({
 				keyStatusUnchecked: "Key unchecked",
 				checkKey: "Check key",
 				keyStatusHint: "A key check reads the usage endpoint; it does not send a model request.",
+				serviceStatus: "Service status",
+				serviceOk: "Reachable",
+				serviceDegraded: "Degraded",
+				usageEndpoint: "Usage endpoint",
+				catalogStatus: "Model catalog",
+				catalogLive: "live",
+				catalogSnapshot: "snapshot",
+				catalogAge: "updated {age}",
+				checkedAt: "checked {time}",
+				latencyValue: "{ms} ms",
+				notChecked: "not checked",
 				keySource: "Source",
 				keyWritable: "Writable",
 				pasteKey: "Paste a key",
 				saveKey: "Save key",
 				clearKey: "Clear key",
+				clearKeyConfirm: "Delete the stored credential for {ref}? This cannot be undone.",
 				importAuth: "Import from OpenCode",
 				imported: "Imported the OpenCode key",
 				testConnection: "Test connection",
@@ -218,11 +230,23 @@ window.__ModuleLoader__.load({
 				keyStatusUnchecked: "未检查",
 				checkKey: "检查密钥",
 				keyStatusHint: "检查仅读取用量端点，不会发送模型请求。",
+				serviceStatus: "服务状态",
+				serviceOk: "可访问",
+				serviceDegraded: "异常",
+				usageEndpoint: "用量端点",
+				catalogStatus: "模型目录",
+				catalogLive: "实时",
+				catalogSnapshot: "快照",
+				catalogAge: "更新于 {age}",
+				checkedAt: "检查于 {time}",
+				latencyValue: "{ms} ms",
+				notChecked: "未检查",
 				keySource: "来源",
 				keyWritable: "可写",
 				pasteKey: "粘贴密钥",
 				saveKey: "保存密钥",
 				clearKey: "清除密钥",
+				clearKeyConfirm: "删除 {ref} 的已保存凭据？此操作无法撤销。",
 				importAuth: "从 OpenCode 导入",
 				imported: "已导入 OpenCode 密钥",
 				testConnection: "测试连接",
@@ -397,8 +421,7 @@ window.__ModuleLoader__.load({
 			}
 		};
 		const GO_PROVIDER = "opencode-go-direct";
-		//#endregion
-		//#region src/client/api.ts
+		// src/client/api.ts
 		async function call(path, init) {
 			const response = await fetch(path, {
 				credentials: "same-origin",
@@ -434,10 +457,13 @@ window.__ModuleLoader__.load({
 				body: JSON.stringify({ label })
 			});
 		}
-		function saveConfig(patch) {
+		function saveConfig(patch, revision) {
 			return call("/dsh-opencode/config", {
 				method: "POST",
-				body: JSON.stringify({ patch })
+				body: JSON.stringify(revision === void 0 ? { patch } : {
+					patch,
+					revision
+				})
 			});
 		}
 		function fetchUpdate() {
@@ -486,11 +512,9 @@ window.__ModuleLoader__.load({
 				})
 			});
 		}
-		//#endregion
-		//#region src/client/events.ts
+		// src/client/events.ts
 		const CONFIG_CHANGED_EVENT = "dsh-opencode-config-changed";
-		//#endregion
-		//#region src/client/heatmap.tsx
+		// src/client/heatmap.tsx
 		const DAYS = [
 			"Mon",
 			"Tue",
@@ -699,8 +723,7 @@ window.__ModuleLoader__.load({
 				]
 			});
 		}
-		//#endregion
-		//#region src/client/usage-analytics.tsx
+		// src/client/usage-analytics.tsx
 		function compact(value) {
 			if (!Number.isFinite(value) || value <= 0) return "0";
 			if (value >= 1e6) return `${(value / 1e6).toFixed(1)}M`;
@@ -909,8 +932,7 @@ window.__ModuleLoader__.load({
 				tab === "sessions" ? sessionRows(analytics, t) : null
 			] });
 		}
-		//#endregion
-		//#region src/client/usage-bars.tsx
+		// src/client/usage-bars.tsx
 		function percentClass(percent) {
 			if (percent >= 90) return "oc-bar-fill oc-bar-fill-bad";
 			if (percent >= 75) return "oc-bar-fill oc-bar-fill-warn";
@@ -979,8 +1001,7 @@ window.__ModuleLoader__.load({
 				]
 			});
 		}
-		//#endregion
-		//#region src/quota-runway.ts
+		// src/quota-runway.ts
 		const HOUR = 36e5;
 		const DAY = 24 * HOUR;
 		const DURATIONS = {
@@ -1046,8 +1067,7 @@ window.__ModuleLoader__.load({
 			};
 			return known === 0 ? { kind: "unknown" } : { kind: "safe" };
 		}
-		//#endregion
-		//#region src/client/quota-card.tsx
+		// src/client/quota-card.tsx
 		const PANEL_EVENT = "dsh-opencode-panel-state";
 		function broadcastPanel(open) {
 			if (typeof window === "undefined") return;
@@ -1589,11 +1609,11 @@ window.__ModuleLoader__.load({
 				]
 			});
 		}
-		//#endregion
-		//#region src/client/settings-draft.ts
+		// src/client/settings-draft.ts
 		function draftOf(state) {
 			const allIds = state.models.map((model) => model.id);
 			return {
+				revision: state.revision ?? 0,
 				apiKeyEnv: state.config.apiKeyEnv,
 				apiBase: state.config.apiBase,
 				goApiBase: state.config.goApiBase,
@@ -1679,8 +1699,7 @@ window.__ModuleLoader__.load({
 		function describeLimitsFull(limits) {
 			return `${planLine("Go", limits.go)} · ${planLine("Go Plus", limits.goPlus)}`;
 		}
-		//#endregion
-		//#region src/client/settings-widgets.tsx
+		// src/client/settings-widgets.tsx
 		function Field(props) {
 			return /* @__PURE__ */ (0, react_jsx_runtime.jsxs)("label", {
 				className: "oc-field",
@@ -1743,8 +1762,7 @@ window.__ModuleLoader__.load({
 			if (account.coolingUntil <= 0) return label;
 			return `${label} · ${untilText(account.coolingUntil)}`;
 		}
-		//#endregion
-		//#region src/client/update-card.tsx
+		// src/client/update-card.tsx
 		function PluginUpdateCard(props) {
 			const { t } = props;
 			const [status, setStatus] = (0, react.useState)(null);
@@ -1781,6 +1799,9 @@ window.__ModuleLoader__.load({
 						children: [t("pluginUpdate"), status === null ? null : status.updateAvailable ? /* @__PURE__ */ (0, react_jsx_runtime.jsx)("span", {
 							className: "oc-badge oc-badge-warn",
 							children: t("updateAvailable")
+						}) : status.latestCheckFailed || status.latestVersion === void 0 ? /* @__PURE__ */ (0, react_jsx_runtime.jsx)("span", {
+							className: "oc-badge oc-badge-warn",
+							children: t("updateCheckFailed")
 						}) : /* @__PURE__ */ (0, react_jsx_runtime.jsx)("span", {
 							className: "oc-badge oc-badge-ok",
 							children: t("upToDate")
@@ -1827,8 +1848,7 @@ window.__ModuleLoader__.load({
 				]
 			});
 		}
-		//#endregion
-		//#region src/client/auth-ref.ts
+		// src/client/auth-ref.ts
 		/**
 		* Collision-free credential ref for an OpenCode auth entry. Hex encoding is
 		* injective (unlike replacing punctuation with underscores) and uses only
@@ -1839,8 +1859,7 @@ window.__ModuleLoader__.load({
 			for (const byte of new TextEncoder().encode(entry)) hex += byte.toString(16).padStart(2, "0");
 			return `OPENCODE_AUTH_${hex}`;
 		}
-		//#endregion
-		//#region src/client/settings-page.tsx
+		// src/client/settings-page.tsx
 		function keyStatusLabel(status, t) {
 			if (status === "ok") return t("keyStatusOk");
 			if (status === "expired") return t("keyStatusExpired");
@@ -1854,6 +1873,13 @@ window.__ModuleLoader__.load({
 			if (status === "expired" || status === "invalid") return "oc-badge-bad";
 			return "oc-badge-warn";
 		}
+		function ageText(at) {
+			if (at === void 0) return "";
+			const minutes = Math.max(0, Math.round((Date.now() - at) / 6e4));
+			if (minutes < 1) return "now";
+			if (minutes < 60) return `${minutes}m`;
+			return `${Math.round(minutes / 60)}h`;
+		}
 		function OpenCodeSettingsPage(props) {
 			const { t } = props;
 			const [state, setState] = (0, react.useState)(null);
@@ -1899,7 +1925,7 @@ window.__ModuleLoader__.load({
 				setSaving(true);
 				setError(null);
 				try {
-					await saveConfig(patchOf(draft));
+					await saveConfig(patchOf(draft), draft.revision);
 					window.dispatchEvent(new CustomEvent(CONFIG_CHANGED_EVENT));
 					setNotice(t("saved"));
 					setState(null);
@@ -2236,7 +2262,9 @@ window.__ModuleLoader__.load({
 										type: "button",
 										className: "oc-button oc-button-danger",
 										onClick: () => {
-											clearKey(draft.apiKeyEnv.trim() === "" ? state.config.apiKeyEnv : draft.apiKeyEnv.trim()).then(() => {
+											const ref = draft.apiKeyEnv.trim() === "" ? state.config.apiKeyEnv : draft.apiKeyEnv.trim();
+											if (!window.confirm(t("clearKeyConfirm").replace("{ref}", ref))) return;
+											clearKey(ref).then(() => {
 												setNotice(t("saved"));
 												return load();
 											}).catch((caught) => setError(caught instanceof Error ? caught.message : t("failed")));
@@ -2300,6 +2328,62 @@ window.__ModuleLoader__.load({
 							})
 						]
 					}),
+					/* @__PURE__ */ (0, react_jsx_runtime.jsxs)("section", {
+						className: "oc-card",
+						children: [
+							/* @__PURE__ */ (0, react_jsx_runtime.jsxs)("h3", {
+								className: "oc-card-title",
+								children: [t("serviceStatus"), /* @__PURE__ */ (0, react_jsx_runtime.jsx)("span", {
+									className: `oc-badge ${state.service.usageCheckedAt === void 0 ? "oc-badge-warn" : state.service.usageError === void 0 ? "oc-badge-ok" : "oc-badge-bad"}`,
+									children: state.service.usageCheckedAt === void 0 ? t("notChecked") : state.service.usageError === void 0 ? t("serviceOk") : t("serviceDegraded")
+								})]
+							}),
+							/* @__PURE__ */ (0, react_jsx_runtime.jsxs)("div", {
+								className: "oc-row",
+								children: [
+									/* @__PURE__ */ (0, react_jsx_runtime.jsx)("span", {
+										className: "oc-hint",
+										children: t("usageEndpoint")
+									}),
+									/* @__PURE__ */ (0, react_jsx_runtime.jsx)("span", {
+										className: `oc-badge ${state.service.usageCheckedAt === void 0 ? "oc-badge-warn" : state.service.usageError === void 0 ? "oc-badge-ok" : "oc-badge-bad"}`,
+										children: state.service.usageCheckedAt === void 0 ? t("notChecked") : t("checkedAt").replace("{time}", new Date(state.service.usageCheckedAt).toLocaleTimeString())
+									}),
+									state.service.usageLatencyMs === void 0 ? null : /* @__PURE__ */ (0, react_jsx_runtime.jsx)("span", {
+										className: "oc-badge",
+										children: t("latencyValue").replace("{ms}", String(state.service.usageLatencyMs))
+									})
+								]
+							}),
+							state.service.usageError === void 0 ? null : /* @__PURE__ */ (0, react_jsx_runtime.jsx)("p", {
+								className: "oc-hint",
+								children: state.service.usageError
+							}),
+							/* @__PURE__ */ (0, react_jsx_runtime.jsxs)("div", {
+								className: "oc-row",
+								children: [
+									/* @__PURE__ */ (0, react_jsx_runtime.jsx)("span", {
+										className: "oc-hint",
+										children: t("catalogStatus")
+									}),
+									/* @__PURE__ */ (0, react_jsx_runtime.jsxs)("span", {
+										className: `oc-badge ${state.catalog.source === "live" ? "oc-badge-ok" : "oc-badge-warn"}`,
+										children: [
+											state.catalog.source === "live" ? t("catalogLive") : t("catalogSnapshot"),
+											" · ",
+											state.catalog.count,
+											" ",
+											t("modelCount")
+										]
+									}),
+									state.catalog.fetchedAt === void 0 ? null : /* @__PURE__ */ (0, react_jsx_runtime.jsx)("span", {
+										className: "oc-badge",
+										children: t("catalogAge").replace("{age}", ageText(state.catalog.fetchedAt))
+									})
+								]
+							})
+						]
+					}),
 					/* @__PURE__ */ (0, react_jsx_runtime.jsxs)("section", {
 						className: "oc-card",
 						children: [
@@ -2479,7 +2563,10 @@ window.__ModuleLoader__.load({
 												/* @__PURE__ */ (0, react_jsx_runtime.jsx)("button", {
 													type: "button",
 													className: "oc-button oc-button-danger",
-													onClick: () => update({ accounts: draft.accounts.filter((_, i) => i !== index) }),
+													onClick: () => {
+														setKeyDrafts({});
+														update({ accounts: draft.accounts.filter((_, i) => i !== index) });
+													},
 													children: t("removeAccount")
 												})
 											]
@@ -2921,8 +3008,7 @@ window.__ModuleLoader__.load({
 				]
 			});
 		}
-		//#endregion
-		//#region src/client/styles.ts
+		// src/client/styles.ts
 		const CSS_ID = "dsh-opencode-provider";
 		const CSS = `
 :where(.oc-page, .oc-panel, .oc-card, .oc-foot, .oc-railButton) {
@@ -3042,8 +3128,7 @@ window.__ModuleLoader__.load({
 .oc-table th, .oc-table td { text-align: left; padding: 5px 8px; border-bottom: 1px solid var(--oc-border-soft); vertical-align: top; }
 .oc-table th { font-weight: 600; opacity: 0.8; }
 `;
-		//#endregion
-		//#region src/client/index.tsx
+		// src/client/index.tsx
 		function injectCss() {
 			if (typeof document === "undefined") return () => {};
 			if (document.querySelector(`style[data-dsh-plugin="dsh-opencode-provider"]`) !== null) return () => {};
@@ -3089,7 +3174,6 @@ window.__ModuleLoader__.load({
 				}, QuotaCard));
 			});
 		}
-		//#endregion
 		exports.apply = apply;
 		exports.inject = inject;
 		return module.exports;

package/lib/index.d.ts

@@ -3,7 +3,6 @@ import z from "@deepseek-ai/schemastery";
 import { Context } from "@deepseek-ai/cordis";
 import { AttachmentStore } from "@deepseek-ai/dsh-attachment";
 import { WebServer } from "@deepseek-ai/dsh-host-webserver";
-//#region src/config.d.ts
 type TransportId = 'openai-chat' | 'openai-responses' | 'anthropic' | 'google';
 interface AccountConfig {
   label?: string;
@@ -92,8 +91,6 @@ interface ResolvedConfig {
   showSidebarQuota: boolean;
   userAgent: string;
 }
-//#endregion
-//#region src/credentials.d.ts
 interface ResolvedKey {
   readonly key: string;
   readonly source: string;
@@ -115,8 +112,6 @@ declare function openCodeAuthPath(): string;
 declare function createKeyPool(ctx: Context, config: () => ResolvedConfig): KeyPool;
 /** Backwards-compatible single-key facade over the pool. */
 declare function createKeyResolver(ctx: Context, config: () => ResolvedConfig): KeyResolver;
-//#endregion
-//#region src/request-log.d.ts
 interface RequestOutcome {
   ok: boolean;
   latencyMs: number;
@@ -153,8 +148,6 @@ interface RequestHeatmap {
   models: ModelRequestHeatmap[];
   accounts: AccountRequestHeatmap[];
 }
-//#endregion
-//#region src/adapter.d.ts
 type RoutingReason = 'rule' | 'sticky' | 'pinned' | 'default' | 'rotation' | 'quota-fallback';
 interface RoutingTraceEntry {
   readonly at: number;
@@ -190,6 +183,8 @@ declare class OpenCodeAdapter extends LlmAdapter {
   private stickyWrites;
   /** key fingerprint → cooldown deadline and the rejection that caused it. */
   private readonly cooldowns;
+  /** key + model → short soft rejection; other models stay usable. */
+  private readonly modelCooldowns;
   private readonly sticky;
   private readonly routing;
   private warnedTools;
@@ -215,6 +210,11 @@ declare class OpenCodeAdapter extends LlmAdapter {
   private send;
   private keyId;
   private coolingDown;
+  /** Current model-scoped soft cooldown for one key, if any. */
+  modelHealth(key: string, model: string): {
+    coolingUntil: number;
+    reason: string;
+  } | undefined;
   /** Current cooldown for one resolved key, if any. */
   accountHealth(key: string): {
     coolingUntil: number;
@@ -225,11 +225,12 @@ declare class OpenCodeAdapter extends LlmAdapter {
   private loadCooldowns;
   private pruneCooldowns;
   private persistCooldowns;
+  private modelCoolingDown;
+  private markRejection;
+  private markModelCooldown;
   private markCooldown;
   private isTransient;
 }
-//#endregion
-//#region src/usage.d.ts
 interface UsageWindow {
   readonly percentUsed: number;
   readonly resetsAt?: string;
@@ -247,6 +248,7 @@ interface UsageResult {
   readonly report?: UsageReport;
   readonly error?: string;
   readonly keyStatus?: UsageKeyStatus;
+  readonly latencyMs?: number;
   readonly checkedAt: number;
 }
 interface UsageWindowPayload {
@@ -269,8 +271,6 @@ interface UsageReader {
 }
 /** Cached reader for the (undocumented) Go usage endpoint. Fail-open by design. */
 declare function createUsageReader(config: () => ResolvedConfig, resolveKey: KeyResolver, fetchImpl?: typeof fetch): UsageReader;
-//#endregion
-//#region src/commands.d.ts
 interface CommandDeps {
   readonly config: () => ResolvedConfig;
   readonly pool: KeyPool;
@@ -278,8 +278,6 @@ interface CommandDeps {
   readonly adapter: OpenCodeAdapter;
 }
 declare function applyCommands(ctx: Context, deps: CommandDeps): void;
-//#endregion
-//#region src/analytics.d.ts
 interface ModelUsageRow {
   provider: string;
   model: string;
@@ -355,8 +353,6 @@ interface AnalyticsReport {
   days: DayUsageRow[];
   sessions: SessionUsageRow[];
 }
-//#endregion
-//#region src/routes.d.ts
 interface RouteDeps {
   readonly config: () => ResolvedConfig;
   readonly adapter: OpenCodeAdapter;
@@ -368,8 +364,6 @@ interface RouteDeps {
 }
 /** Register the Host HTTP surface consumed by the browser settings page. */
 declare function applyRoutes(ctx: Context, web: WebServer, deps: RouteDeps): void;
-//#endregion
-//#region src/session-id.d.ts
 /**
  * A stable, opaque session id for `x-opencode-session`. The header must stay
  * identical across every turn and retry of one conversation; deriving it from
@@ -378,16 +372,12 @@ declare function applyRoutes(ctx: Context, web: WebServer, deps: RouteDeps): voi
  * routing key, so the `dsh_` prefix is safe.
  */
 declare function openCodeSessionId(options: GenerateOptions): string;
-//#endregion
-//#region src/errors.d.ts
 /**
  * Translate an OpenCode HTTP failure into the harness failure taxonomy.
  * OpenCode reports several distinct conditions as 400/401, so the body type
  * and message decide the code — never the status alone.
  */
 declare function classifyResponseError(status: number, text: string, headers: Headers): LlmError;
-//#endregion
-//#region src/models.generated.d.ts
 type OpenCodeTransport = 'openai-chat' | 'openai-responses' | 'anthropic' | 'google';
 interface OpenCodeModelMeta {
   readonly id: string;
@@ -404,8 +394,6 @@ interface OpenCodeModelMeta {
     readonly cacheRead: number;
   };
 }
-//#endregion
-//#region src/catalog.d.ts
 interface EffectiveModel extends OpenCodeModelMeta {
   readonly transport: OpenCodeTransport;
   readonly contextWindow: number;
@@ -413,8 +401,6 @@ interface EffectiveModel extends OpenCodeModelMeta {
 declare function effectiveModel(id: string, config: ResolvedConfig): EffectiveModel;
 declare function describeModel(meta: EffectiveModel, provider?: string): string;
 declare function clearCatalogCache(): void;
-//#endregion
-//#region src/index.d.ts
 declare const name = "llm-opencode";
 declare const inject: string[];
 /**
@@ -427,6 +413,5 @@ declare const inject: string[];
  * next request without a restart.
  */
 declare function apply(ctx: Context, config: Config): void;
-//#endregion
 export { type CommandDeps, Config, type Config as OpenCodeConfig, OpenCodeAdapter, type OpenCodeAdapterDeps, type ResolvedConfig, type ResolvedKey, type RouteDeps, type TransportId, type UsageReport, type UsageResult, type UsageWindow, apply, applyCommands, applyRoutes, classifyResponseError, clearCatalogCache, createKeyPool, createKeyResolver, createUsageReader, describeModel, effectiveModel, inject, name, normalizeUsage, openCodeAuthPath, openCodeSessionId };
 //# sourceMappingURL=index.d.ts.map

package/lib/index.js

@@ -10,7 +10,7 @@ import { launchEnvironmentOf } from "@deepseek-ai/dsh-launch-environment";
 import { spawn } from "node:child_process";
 import { fileURLToPath } from "node:url";
 import { isIP } from "node:net";
-//#region src/config.ts
+// src/config.ts
 const NS = "llm-opencode";
 const DEFAULT_API_KEY_ENV = "OPENCODE_API_KEY";
 const DEFAULT_API_BASE = "https://opencode.ai/inference";
@@ -18,7 +18,7 @@ const DEFAULT_GO_API_BASE = "https://opencode.ai/inference/go";
 const DEFAULT_GO_ZEN_BASE = "https://opencode.ai/zen/go/v1";
 const DEFAULT_USAGE_URL = "https://opencode.ai/zen/go/v1/usage";
 const DEFAULT_AUTH_ENTRY = "opencode-go";
-const DEFAULT_USER_AGENT = "dsh-opencode-provider/0.5.18";
+const DEFAULT_USER_AGENT = "dsh-opencode-provider/0.5.20";
 const DEFAULT_CONTEXT_WINDOW = 262144;
 const DEFAULT_MAX_TOKENS = 32768;
 const DEFAULT_GENERATE_MAX_TOKENS = 32768;
@@ -151,8 +151,7 @@ function resolveConfig(config) {
 		userAgent: typeof config.userAgent === "string" && config.userAgent !== "" ? config.userAgent : DEFAULT_USER_AGENT
 	};
 }
-//#endregion
-//#region src/models.generated.ts
+// src/models.generated.ts
 const GENERATED_MODELS = {
 	"beam-free": {
 		"id": "beam-free",
@@ -791,8 +790,7 @@ const GENERATED_MODELS = {
 		}
 	}
 };
-//#endregion
-//#region src/provider-ids.ts
+// src/provider-ids.ts
 /**
 * Provider ids owned by this plugin. The built-in llm-pi-ai catalog already
 * declares `opencode` and `opencode-go`; the core registry rejects duplicate
@@ -801,8 +799,7 @@ const GENERATED_MODELS = {
 const CONSOLE_PROVIDER = "opencode-direct";
 const GO_PROVIDER = "opencode-go-direct";
 const PLUGIN_PROVIDER_SET = /* @__PURE__ */ new Set([CONSOLE_PROVIDER, GO_PROVIDER]);
-//#endregion
-//#region src/routing.ts
+// src/routing.ts
 function isGoProvider(provider) {
 	return provider === "opencode-go-direct" || provider === "opencode-go";
 }
@@ -826,8 +823,7 @@ function endpointFor(provider, transport, model, config) {
 		case "google": return `${root}/google/v1beta/models/${encodeURIComponent(model)}:streamGenerateContent?alt=sse`;
 	}
 }
-//#endregion
-//#region src/subscription-limits.ts
+// src/subscription-limits.ts
 const UNLIMITED = {
 	monthlyUsd: "unlimited",
 	fiveHour: "unlimited",
@@ -974,8 +970,7 @@ function formatRequests(value) {
 function describeLimitsShort(limits) {
 	return `Go ${formatRequests(limits.go.fiveHour)}/5h · ${formatRequests(limits.go.weekly)}/wk`;
 }
-//#endregion
-//#region src/catalog.ts
+// src/catalog.ts
 const CATALOG_TTL_MS = 6e5;
 const CATALOG_TIMEOUT_MS = 15e3;
 const catalogCache = /* @__PURE__ */ new Map();
@@ -1090,11 +1085,21 @@ function describeModel(meta, provider) {
 	}
 	return parts.join(" · ");
 }
+function catalogStatus(provider, config, count) {
+	const cached = catalogCache.get(`${provider}|${modelsUrlFor(provider, config)}`);
+	return cached === void 0 ? {
+		source: "snapshot",
+		count
+	} : {
+		source: "live",
+		count,
+		fetchedAt: cached.at
+	};
+}
 function clearCatalogCache() {
 	catalogCache.clear();
 }
-//#endregion
-//#region src/account-rules.ts
+// src/account-rules.ts
 function glob(pattern) {
 	const escaped = pattern.replace(/[.+^${}()|[\]\\]/g, "\\$&").replace(/\*/g, ".*").replace(/\?/g, ".");
 	return new RegExp(`^${escaped}$`, "i");
@@ -1117,8 +1122,7 @@ function orderKeysForModel(keys, model, rules) {
 	if (preferred === void 0) return keys;
 	return [preferred, ...keys.filter((entry) => entry !== preferred)];
 }
-//#endregion
-//#region src/errors.ts
+// src/errors.ts
 function parseErrorBody(text) {
 	try {
 		const body = JSON.parse(text);
@@ -1179,8 +1183,7 @@ function classifyStreamError(value, status = 200) {
 	if (value.error === void 0) return void 0;
 	return classifyResponseError(status, JSON.stringify(value), new Headers());
 }
-//#endregion
-//#region src/images.ts
+// src/images.ts
 const MAX_LONG_EDGE = 1568;
 const MAX_BYTES = 1048576;
 /** Total base64 image budget for one request; leaves room under the gateway cap. */
@@ -1191,7 +1194,7 @@ const MAX_REQUEST_IMAGES = 50;
 * dropped first; the transports render them as an "image omitted" placeholder
 * instead of failing the whole request.
 */
-function selectRequestImages(refs) {
+function selectRequestImages(refs, occurrences) {
 	const unique = [];
 	const seen = /* @__PURE__ */ new Set();
 	for (const ref of refs) {
@@ -1199,11 +1202,18 @@ function selectRequestImages(refs) {
 		seen.add(ref.attachmentId);
 		unique.push(ref);
 	}
+	const countOf = (ref) => Math.max(1, occurrences?.get(ref.attachmentId) ?? 1);
+	const encoded = (bytes) => Math.ceil(Math.max(0, bytes) * 4 / 3);
 	let start = 0;
-	let bytes = unique.reduce((sum, ref) => sum + Math.max(0, ref.bytes), 0);
-	while (start < unique.length && (unique.length - start > MAX_REQUEST_IMAGES || bytes > MAX_REQUEST_IMAGE_BYTES)) {
+	let totalImages = unique.reduce((sum, ref) => sum + countOf(ref), 0);
+	let bytes = unique.reduce((sum, ref) => sum + encoded(ref.bytes) * countOf(ref), 0);
+	while (start < unique.length && (totalImages > MAX_REQUEST_IMAGES || bytes > MAX_REQUEST_IMAGE_BYTES)) {
 		const dropped = unique[start];
-		if (dropped !== void 0) bytes -= Math.max(0, dropped.bytes);
+		if (dropped !== void 0) {
+			const count = countOf(dropped);
+			totalImages -= count;
+			bytes -= encoded(dropped.bytes) * count;
+		}
 		start += 1;
 	}
 	const kept = unique.slice(start);
@@ -1226,21 +1236,23 @@ function targetOf(ref) {
 	};
 }
 function hasImages(options) {
-	return options.messages.some((message) => message.content.some((block) => block.type === "image"));
+	return options.messages.some((message) => message.content.some((block) => block.type === "image" && block.offloaded !== true));
 }
 /** Decode the distinct request images that fit the request-wide budget. */
 async function resolveImages(options, store, signal) {
 	const refs = [];
+	const counts = /* @__PURE__ */ new Map();
 	const seen = /* @__PURE__ */ new Set();
 	for (const message of options.messages) for (const block of message.content) {
 		if (block.type !== "image" || block.offloaded === true) continue;
 		const ref = block.attachment;
+		counts.set(ref.attachmentId, (counts.get(ref.attachmentId) ?? 0) + 1);
 		if (seen.has(ref.attachmentId)) continue;
 		seen.add(ref.attachmentId);
 		refs.push(ref);
 	}
 	const images = /* @__PURE__ */ new Map();
-	for (const ref of selectRequestImages(refs).kept) {
+	for (const ref of selectRequestImages(refs, counts).kept) {
 		const request = await store.readImageRequest(ref, targetOf(ref), signal);
 		const base64 = Buffer.from(request.data).toString("base64");
 		images.set(ref.attachmentId, {
@@ -1256,8 +1268,7 @@ async function resolveImages(options, store, signal) {
 function offloadPlaceholder(ref) {
 	return `[image omitted: ${ref.name ?? ref.attachmentId}]`;
 }
-//#endregion
-//#region src/files.ts
+// src/files.ts
 /**
 * Model-facing handle for a file when the adapter receives a raw FileBlock.
 * The harness normally projects files to text with an execution-world path
@@ -1267,8 +1278,7 @@ function filePlaceholder(ref) {
 	const digest = String(ref.attachmentId).slice(7, 15);
 	return `[File ${JSON.stringify(ref.name)} (${ref.bytes} bytes, sha256:${digest}) was uploaded, but the current execution environment cannot access a readable path. Report that limitation if its contents are needed; do not claim to have read it.]`;
 }
-//#endregion
-//#region src/tool-schema.ts
+// src/tool-schema.ts
 /**
 * Normalize a tool's parameters schema to the object root the gateway
 * validates. Ported behavior from the Command Code provider: pass an object
@@ -1374,8 +1384,7 @@ function normalizeToolParameters(parameters, depth = 0) {
 		additionalProperties: true
 	};
 }
-//#endregion
-//#region src/messages.ts
+// src/messages.ts
 function textOnly$3(content) {
 	const parts = [];
 	for (const block of content) if (block.type === "text") parts.push(block.text);
@@ -1523,8 +1532,7 @@ function buildChatBody(options, model, config, images) {
 		...options.stop === void 0 || options.stop.length === 0 ? {} : { stop: options.stop }
 	};
 }
-//#endregion
-//#region src/request-log.ts
+// src/request-log.ts
 const DAY_MS$1 = 864e5;
 const MAX_LOG_BYTES = 2097152;
 const KEEP_LINES = 5e3;
@@ -1707,8 +1715,7 @@ async function readRequestHeatmap(days, now = Date.now()) {
 		}).sort((a, b) => b.total - a.total || a.account.localeCompare(b.account))
 	};
 }
-//#endregion
-//#region src/session-id.ts
+// src/session-id.ts
 function sha256(value) {
 	return createHash("sha256").update(value).digest("hex");
 }
@@ -1762,8 +1769,7 @@ function limitTools(tools, max = 128) {
 	const rest = out.filter((tool) => !CORE_TOOLS.has(tool.name));
 	return [...core, ...rest].slice(0, max);
 }
-//#endregion
-//#region src/transports/sse.ts
+// src/transports/sse.ts
 async function readWithIdleTimeout(reader, idleTimeoutMs, signal) {
 	if (signal?.aborted === true) throw signal.reason;
 	return await new Promise((resolve, reject) => {
@@ -1787,6 +1793,12 @@ async function readWithIdleTimeout(reader, idleTimeoutMs, signal) {
 		reader.read().then((value) => finish(() => resolve(value)), (error) => finish(() => reject(error)));
 	});
 }
+function appendDecoded(buffer, decoded) {
+	let text = buffer + decoded;
+	const trailingCR = text.endsWith("\r");
+	if (trailingCR) text = text.slice(0, -1);
+	return text.replace(/\r\n?/g, "\n") + (trailingCR ? "\r" : "");
+}
 /** Yield the `data:` payload of every SSE event, honoring cancellation and idle timeouts. */
 async function* sseData(body, idleTimeoutMs, signal) {
 	const reader = body.getReader();
@@ -1796,7 +1808,7 @@ async function* sseData(body, idleTimeoutMs, signal) {
 		for (;;) {
 			const result = await readWithIdleTimeout(reader, idleTimeoutMs, signal);
 			if (result.done) break;
-			buffer += decoder.decode(result.value, { stream: true }).replace(/\r\n/g, "\n");
+			buffer = appendDecoded(buffer, decoder.decode(result.value, { stream: true }));
 			for (;;) {
 				const boundary = buffer.indexOf("\n\n");
 				if (boundary < 0) break;
@@ -1824,7 +1836,7 @@ async function* sseEvents(body, idleTimeoutMs, signal) {
 		for (;;) {
 			const result = await readWithIdleTimeout(reader, idleTimeoutMs, signal);
 			if (result.done) break;
-			buffer += decoder.decode(result.value, { stream: true }).replace(/\r\n/g, "\n");
+			buffer = appendDecoded(buffer, decoder.decode(result.value, { stream: true }));
 			for (;;) {
 				const boundary = buffer.indexOf("\n\n");
 				if (boundary < 0) break;
@@ -1854,8 +1866,7 @@ async function* sseEvents(body, idleTimeoutMs, signal) {
 		reader.releaseLock();
 	}
 }
-//#endregion
-//#region src/transports/anthropic.ts
+// src/transports/anthropic.ts
 const EFFORT_LEVELS$1 = /* @__PURE__ */ new Set([
 	"low",
 	"medium",
@@ -2040,6 +2051,7 @@ async function* parseAnthropicStream(response, idleTimeoutMs, signal) {
 	let nextIndex = 0;
 	let sawContent = false;
 	let sawToolUse = false;
+	let sawMessageStop = false;
 	let stopReason;
 	let inputUsage;
 	let outputTokens;
@@ -2208,8 +2220,12 @@ async function* parseAnthropicStream(response, idleTimeoutMs, signal) {
 			if (typeof event.usage?.output_tokens === "number") outputTokens = event.usage.output_tokens;
 			continue;
 		}
-		if (event.type === "message_stop") break;
+		if (event.type === "message_stop") {
+			sawMessageStop = true;
+			break;
+		}
 	}
+	if (sawContent && !sawMessageStop) throw new LlmError("Anthropic stream ended without message_stop", "TRANSPORT");
 	const toolBlocks = [...blocks.values()].filter((block) => block.kind === "tool-call");
 	if (toolBlocks.length > 0) {
 		for (const block of toolBlocks) yield {
@@ -2239,8 +2255,7 @@ async function* parseAnthropicStream(response, idleTimeoutMs, signal) {
 		} } : {}
 	};
 }
-//#endregion
-//#region src/transports/google.ts
+// src/transports/google.ts
 function textOnly$1(content) {
 	const parts = [];
 	for (const block of content) if (block.type === "text") parts.push(block.text);
@@ -2483,6 +2498,7 @@ async function* parseGoogleStream(response, idleTimeoutMs, signal) {
 			}
 		}
 	}
+	if (sawContent && finishReason === void 0) throw new LlmError("Google stream ended without a finish reason", "TRANSPORT");
 	if (textIndex !== void 0) yield {
 		type: "block-end",
 		index: textIndex,
@@ -2514,8 +2530,7 @@ async function* parseGoogleStream(response, idleTimeoutMs, signal) {
 		} } : {}
 	};
 }
-//#endregion
-//#region src/transports/openai-chat.ts
+// src/transports/openai-chat.ts
 function mapUsage$1(usage) {
 	const prompt = usage.prompt_tokens ?? 0;
 	const cached = usage.prompt_tokens_details?.cached_tokens ?? 0;
@@ -2554,8 +2569,12 @@ async function* parseChatStream(response, idleTimeoutMs, signal) {
 	let finishReason;
 	let usage;
 	let sawContent = false;
+	let sawTerminal = false;
 	for await (const data of sseData(response.body, idleTimeoutMs, signal)) {
-		if (data === "[DONE]") break;
+		if (data === "[DONE]") {
+			sawTerminal = true;
+			break;
+		}
 		let chunk;
 		try {
 			chunk = JSON.parse(data);
@@ -2638,9 +2657,13 @@ async function* parseChatStream(response, idleTimeoutMs, signal) {
 					argumentsDelta
 				};
 			}
-			if (choice.finish_reason !== void 0 && choice.finish_reason !== null) finishReason = choice.finish_reason;
+			if (choice.finish_reason !== void 0 && choice.finish_reason !== null) {
+				finishReason = choice.finish_reason;
+				sawTerminal = true;
+			}
 		}
 	}
+	if (sawContent && !sawTerminal) throw new LlmError("OpenCode stream ended without a terminal finish event", "TRANSPORT");
 	if (textIndex !== void 0) yield {
 		type: "block-end",
 		index: textIndex,
@@ -2771,8 +2794,7 @@ async function* parseCompletion(chunk) {
 		reason: finishReasonOf$1(choice?.finish_reason, (message?.tool_calls?.length ?? 0) > 0)
 	};
 }
-//#endregion
-//#region src/transports/responses.ts
+// src/transports/responses.ts
 const EFFORT_LEVELS = /* @__PURE__ */ new Set([
 	"minimal",
 	"low",
@@ -2927,6 +2949,7 @@ async function* parseResponsesStream(response, idleTimeoutMs, signal) {
 	const tools = /* @__PURE__ */ new Map();
 	let sawContent = false;
 	let hasTools = false;
+	let completed = false;
 	let usage;
 	let incomplete;
 	for await (const { event: eventName, data } of sseEvents(response.body, idleTimeoutMs, signal)) {
@@ -3040,11 +3063,13 @@ async function* parseResponsesStream(response, idleTimeoutMs, signal) {
 			continue;
 		}
 		if (type === "response.completed") {
+			completed = true;
 			usage = mapUsage(event.response?.usage);
 			incomplete = event.response?.incomplete_details?.reason;
 			continue;
 		}
 	}
+	if (sawContent && !completed) throw new LlmError("OpenAI Responses stream ended without response.completed", "TRANSPORT");
 	if (textIndex !== void 0) yield {
 		type: "block-end",
 		index: textIndex,
@@ -3081,8 +3106,7 @@ async function* parseResponsesStream(response, idleTimeoutMs, signal) {
 		reason: finishReasonOf(hasTools, incomplete)
 	};
 }
-//#endregion
-//#region src/adapter.ts
+// src/adapter.ts
 function defaultCooldownPath() {
 	const home = process.env.DSH_HOME?.trim() || join(homedir(), ".dsh");
 	return join(home, "storages", "dsh-opencode", "cooldowns.json");
@@ -3100,6 +3124,8 @@ var OpenCodeAdapter = class extends LlmAdapter {
 	stickyWrites = Promise.resolve();
 	/** key fingerprint → cooldown deadline and the rejection that caused it. */
 	cooldowns = /* @__PURE__ */ new Map();
+	/** key + model → short soft rejection; other models stay usable. */
+	modelCooldowns = /* @__PURE__ */ new Map();
 	sticky = /* @__PURE__ */ new Map();
 	routing = [];
 	warnedTools = false;
@@ -3202,7 +3228,10 @@ var OpenCodeAdapter = class extends LlmAdapter {
 				}
 			}
 		}
-		const initial = keys.find((entry) => fallback || !this.coolingDown(entry.key)) ?? keys[0];
+		const eligible = fallback ? keys : keys.filter((entry) => !this.coolingDown(entry.key) && !this.modelCoolingDown(entry.key, options.model));
+		const ordered = eligible.length > 0 ? eligible : keys;
+		const forced = !fallback && eligible.length === 0;
+		const initial = ordered[0];
 		const url = endpointFor(options.provider, model.transport, options.model, config);
 		let images;
 		if (hasImages(options)) {
@@ -3226,8 +3255,8 @@ var OpenCodeAdapter = class extends LlmAdapter {
 		let lastError;
 		let lastAccount = initial?.label ?? "default";
 		let loggedAny = false;
-		for (const [index, entry] of keys.entries()) {
-			if (!fallback && this.coolingDown(entry.key)) continue;
+		for (const [index, entry] of ordered.entries()) {
+			if (!fallback && !forced && (this.coolingDown(entry.key) || this.modelCoolingDown(entry.key, options.model))) continue;
 			lastAccount = entry.label ?? "default";
 			const attemptReason = index === 0 ? reason : "rotation";
 			const attemptStartedAt = Date.now();
@@ -3254,9 +3283,9 @@ var OpenCodeAdapter = class extends LlmAdapter {
 					const error = classifyResponseError(response.status, text, response.headers);
 					lastError = error;
 					logAttempt(false, error.code);
-					const canRotate = index < keys.length - 1;
+					const canRotate = index < ordered.length - 1;
 					if (error.code === "INVALID_CREDENTIAL" || error.code === "RATE_LIMIT" || error.code === "QUOTA") {
-						if (!fallback) this.markCooldown(entry.key, response.status, response.headers, error.code);
+						if (!fallback) this.markRejection(entry.key, options.model, response.status, response.headers, error.code);
 						responseFailure = true;
 						if (canRotate) continue;
 					}
@@ -3297,8 +3326,8 @@ var OpenCodeAdapter = class extends LlmAdapter {
 				}
 				lastError = classified;
 				const transient = this.isTransient(classified);
-				if (transient && !responseFailure && !fallback) this.markCooldown(entry.key, 0);
-				if (!(index < keys.length - 1 && transient)) {
+				if (transient && !responseFailure && !fallback) this.markRejection(entry.key, options.model, 0, void 0, classified.code);
+				if (!(index < ordered.length - 1 && transient)) {
 					if (!fallback && config.quotaFallbackModel !== "" && isGoProvider(options.provider) && classified.code === "QUOTA") break;
 					throw classified;
 				}
@@ -3416,6 +3445,8 @@ var OpenCodeAdapter = class extends LlmAdapter {
 			"user-agent": config.userAgent,
 			...transport === "anthropic" ? { "anthropic-version": "2023-06-01" } : {}
 		};
+		const callerAborted = () => signal?.aborted === true;
+		if (callerAborted()) throw signal?.reason instanceof Error ? signal.reason : new LlmError("OpenCode request was aborted before dispatch", "TRANSPORT");
 		const controller = new AbortController();
 		let timedOut = false;
 		const onAbort = () => controller.abort(signal?.reason);
@@ -3433,7 +3464,7 @@ var OpenCodeAdapter = class extends LlmAdapter {
 				signal: controller.signal
 			});
 		} catch (error) {
-			if (signal?.aborted === true) throw signal.reason ?? error;
+			if (callerAborted()) throw signal?.reason ?? error;
 			if (timedOut) throw new LlmError(`OpenCode did not return response headers within ${config.requestTimeoutMs}ms`, "TIMEOUT", { cause: error });
 			throw error;
 		} finally {
@@ -3453,6 +3484,20 @@ var OpenCodeAdapter = class extends LlmAdapter {
 		}
 		return true;
 	}
+	/** Current model-scoped soft cooldown for one key, if any. */
+	modelHealth(key, model) {
+		const id = `${this.keyId(key)}:${model}`;
+		const entry = this.modelCooldowns.get(id);
+		if (entry === void 0) return void 0;
+		if (entry.until <= Date.now()) {
+			this.modelCooldowns.delete(id);
+			return;
+		}
+		return {
+			coolingUntil: entry.until,
+			reason: entry.reason
+		};
+	}
 	/** Current cooldown for one resolved key, if any. */
 	accountHealth(key) {
 		const id = this.keyId(key);
@@ -3505,6 +3550,30 @@ var OpenCodeAdapter = class extends LlmAdapter {
 			await writeFile(this.cooldownPath, payload, "utf8");
 		}).catch(() => {});
 	}
+	modelCoolingDown(key, model) {
+		const id = `${this.keyId(key)}:${model}`;
+		const entry = this.modelCooldowns.get(id);
+		if (entry === void 0) return false;
+		if (entry.until <= Date.now()) {
+			this.modelCooldowns.delete(id);
+			return false;
+		}
+		return true;
+	}
+	markRejection(key, model, status, headers, code) {
+		if (code === "QUOTA" || code === "RATE_LIMIT") {
+			this.markModelCooldown(key, model, code);
+			return;
+		}
+		this.markCooldown(key, status, headers, code);
+	}
+	markModelCooldown(key, model, code) {
+		const delayMs = code === "QUOTA" ? 3e5 : 6e4;
+		this.modelCooldowns.set(`${this.keyId(key)}:${model}`, {
+			until: Date.now() + delayMs,
+			reason: code === "QUOTA" ? "quota" : "rate-limit"
+		});
+	}
 	markCooldown(key, status, headers, code) {
 		let delayMs = 3e4;
 		let reason = "transport";
@@ -3532,8 +3601,7 @@ var OpenCodeAdapter = class extends LlmAdapter {
 		return error.code === "RATE_LIMIT" || error.code === "QUOTA" || error.code === "SERVER" || error.code === "TIMEOUT" || error.code === "TRANSPORT" || error.code === "EMPTY_RESPONSE";
 	}
 };
-//#endregion
-//#region src/commands.ts
+// src/commands.ts
 function bar(percent, width = 10) {
 	const clamped = Math.max(0, Math.min(100, percent));
 	const filled = Math.round(clamped / 100 * width);
@@ -3719,8 +3787,7 @@ function applyCommands(ctx, deps) {
 		}
 	});
 }
-//#endregion
-//#region src/credentials.ts
+// src/credentials.ts
 /** OpenCode stores its own sign-in at this XDG data path. */
 function openCodeAuthPath() {
 	const base = process.env.XDG_DATA_HOME?.trim() || join(homedir(), ".local", "share");
@@ -3854,8 +3921,7 @@ function missingKey(config) {
 function createKeyResolver(ctx, config) {
 	return createKeyPool(ctx, config).resolve;
 }
-//#endregion
-//#region src/local-host.ts
+// src/local-host.ts
 const LOOPBACK = /* @__PURE__ */ new Set([
 	"localhost",
 	"localhost.",
@@ -3896,8 +3962,7 @@ function sameOrigin(authority, origin) {
 		return false;
 	}
 }
-//#endregion
-//#region src/plugin-updater.ts
+// src/plugin-updater.ts
 const UPDATE_HEADER = "x-dsh-plugin-update";
 const UPDATE_TIMEOUT_MS = 6e5;
 const VERSION_CACHE_MS = 3e5;
@@ -4193,8 +4258,7 @@ function registerPluginUpdater(ctx, options) {
 		}
 	});
 }
-//#endregion
-//#region src/analytics.ts
+// src/analytics.ts
 const DAY_MS = 864e5;
 const CACHE_TTL_MS = 6e4;
 const MAX_FILES = 5e3;
@@ -4465,8 +4529,7 @@ async function readModelAnalytics(days = 30) {
 	});
 	return report;
 }
-//#endregion
-//#region src/routes-public.ts
+// src/routes-public.ts
 function publicConfig(config) {
 	return {
 		apiKeyEnv: config.apiKeyEnv,
@@ -4499,8 +4562,7 @@ function publicConfig(config) {
 		userAgent: config.userAgent
 	};
 }
-//#endregion
-//#region src/routes-values.ts
+// src/routes-values.ts
 const TRANSPORTS = /* @__PURE__ */ new Set([
 	"openai-chat",
 	"openai-responses",
@@ -4560,8 +4622,7 @@ function cleanTransports(value) {
 	for (const [key, entry] of Object.entries(value)) if (typeof entry === "string" && TRANSPORTS.has(entry)) out[key] = entry;
 	return out;
 }
-//#endregion
-//#region src/routes.ts
+// src/routes.ts
 const MAX_BODY_BYTES = 262144;
 function json(res, status, value) {
 	res.statusCode = status;
@@ -4779,7 +4840,13 @@ async function buildState(ctx, deps) {
 		key: keyInfo,
 		accounts,
 		models,
+		catalog: catalogStatus(GO_PROVIDER, config, models.length),
 		usage: usage ?? null,
+		service: {
+			...usage?.checkedAt === void 0 ? {} : { usageCheckedAt: usage.checkedAt },
+			...usage?.latencyMs === void 0 ? {} : { usageLatencyMs: usage.latencyMs },
+			...usage?.error === void 0 ? {} : { usageError: usage.error }
+		},
 		routing: deps.adapter.routingTrace(),
 		auth: {
 			entry: config.openCodeAuthEntry,
@@ -4889,12 +4956,14 @@ function applyRoutes(ctx, web, deps) {
 				json(res, 400, { error: "no known fields in patch" });
 				return;
 			}
-			await settings.mutate(NS, ops, settingsRevision(ctx));
+			const revision = typeof body.revision === "number" && Number.isInteger(body.revision) ? body.revision : settingsRevision(ctx);
+			await settings.mutate(NS, ops, revision);
 			deps.usage.clear();
 			deps.refreshTui?.();
 			json(res, 200, { ok: true });
 		} catch (error) {
-			json(res, 500, { error: error instanceof Error ? error.message : "config write failed" });
+			const message = error instanceof Error ? error.message : "config write failed";
+			json(res, /revision|conflict|stale/i.test(message) ? 409 : 500, { error: message });
 		}
 	});
 	route("/dsh-opencode/key", async (req, res) => {
@@ -5062,8 +5131,7 @@ function applyRoutes(ctx, web, deps) {
 		for (const dispose of disposers.splice(0)) dispose();
 	}, "dsh-opencode-provider: HTTP routes");
 }
-//#endregion
-//#region src/tui-settings.ts
+// src/tui-settings.ts
 const TUI_ACCOUNT_AUTO = "auto";
 function setOrClear(text) {
 	const trimmed = text.trim();
@@ -5307,8 +5375,7 @@ function applyOpenCodeTuiSettings(ctx, deps) {
 	}, "dsh-opencode-provider: tui settings section");
 	return refresh;
 }
-//#endregion
-//#region src/usage.ts
+// src/usage.ts
 const SUCCESS_TTL_MS = 3e5;
 const FAILURE_TTL_MS = 45e3;
 const USAGE_TIMEOUT_MS = 2e4;
@@ -5360,7 +5427,9 @@ function createUsageReader(config, resolveKey, fetchImpl = fetch) {
 			if (Date.now() - cached.checkedAt < ttl) return cached;
 		}
 		const result = async () => {
+			const startedAt = Date.now();
 			const signal = AbortSignal.timeout(USAGE_TIMEOUT_MS);
+			const latency = () => Math.max(0, Date.now() - startedAt);
 			try {
 				let response;
 				let lastNetworkError;
@@ -5386,23 +5455,27 @@ function createUsageReader(config, resolveKey, fetchImpl = fetch) {
 					return {
 						error: "usage endpoint rejected the key (inference may still work)",
 						keyStatus: /expired/i.test(body) ? "expired" : "invalid",
+						latencyMs: latency(),
 						checkedAt: Date.now()
 					};
 				}
 				if (!response.ok) return {
 					error: `usage endpoint returned HTTP ${response.status}`,
 					keyStatus: "unavailable",
+					latencyMs: latency(),
 					checkedAt: Date.now()
 				};
 				return {
 					report: normalizeUsage(await response.json()),
 					keyStatus: "ok",
+					latencyMs: latency(),
 					checkedAt: Date.now()
 				};
 			} catch (error) {
 				return {
 					error: error instanceof Error ? error.message : "usage request failed",
 					keyStatus: "unavailable",
+					latencyMs: latency(),
 					checkedAt: Date.now()
 				};
 			}
@@ -5416,8 +5489,7 @@ function createUsageReader(config, resolveKey, fetchImpl = fetch) {
 	};
 	return read;
 }
-//#endregion
-//#region src/index.ts
+// src/index.ts
 const name = "llm-opencode";
 const inject = ["llm"];
 /**
@@ -5501,7 +5573,6 @@ function apply(ctx, config) {
 		}), "dsh-opencode-provider: plugin updater");
 	});
 }
-//#endregion
 export { Config, OpenCodeAdapter, apply, applyCommands, applyRoutes, classifyResponseError, clearCatalogCache, createKeyPool, createKeyResolver, createUsageReader, describeModel, effectiveModel, inject, name, normalizeUsage, openCodeAuthPath, openCodeSessionId };
 
 //# sourceMappingURL=index.js.map

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@goodandready/dsh-opencode-provider",
-  "version": "0.5.18",
+  "version": "0.5.20",
   "description": "OpenCode / OpenCode Go LLM provider plugin for DeepSeek Harness (dsh).",
   "license": "MIT",
   "type": "module",
@@ -69,10 +69,8 @@
     "build": "tsdown",
     "prepare": "tsdown 1>&2",
     "typecheck": "tsc --noEmit",
-    "test": "node --import tsx --test tests/core.test.ts tests/routes.test.ts tests/limits.test.ts tests/analytics.test.ts tests/adapter.test.ts tests/usage.test.ts tests/request-log.test.ts tests/account-rules.test.ts tests/commands.test.ts tests/tool-schema.test.ts tests/tui-settings.test.ts tests/plan-filter.test.ts tests/image-budget.test.ts tests/tools.test.ts tests/plugin-updater.test.ts tests/client-render.test.tsx",
+    "test": "node --import tsx --test tests/core.test.ts tests/routes.test.ts tests/limits.test.ts tests/analytics.test.ts tests/adapter.test.ts tests/usage.test.ts tests/request-log.test.ts tests/account-rules.test.ts tests/commands.test.ts tests/tool-schema.test.ts tests/tui-settings.test.ts tests/plan-filter.test.ts tests/image-budget.test.ts tests/tools.test.ts tests/plugin-updater.test.ts tests/client-render.test.tsx tests/sse.test.ts",
     "check": "npm run typecheck && npm test && npm run build",
-    "test:engine": "node scripts/test-engines.mjs",
-    "test:install": "node scripts/test-engines.mjs --install-only",
     "release": "npm run check && npm pack",
     "sync:models": "node scripts/sync-models.mjs"
   },