@fanchaozz/provider-manager 0.2.0 → 0.2.2

package/commands.ts

@@ -259,7 +259,8 @@ async function testCommand(ctx: ExtensionCommandContext, arg: string): Promise<v
 	}
 
 	const result = await testModel({ ctx: ctx as any, provider, model, mode: "full" });
-	ctx.ui.notify(formatTestResult(result), result.ok ? "info" : "warning");
+	// 同步 dashboard：测试结果统一 info（showStatus 可覆盖），失败语义靠文本 ✗ fail 前缀表达
+	ctx.ui.notify(formatTestResult(result), "info");
 }
 
 /** 批量测某 provider 全部 model；无参数时取第一个 provider。 */
@@ -295,7 +296,8 @@ async function testAllCommand(ctx: ExtensionCommandContext, providerId: string |
 		mode: "full",
 		concurrency: 3,
 	});
-	for (const r of results) {
-		ctx.ui.notify(formatTestResult(r), r.ok ? "info" : "warning");
-	}
+	// 批量结果拼成一条 notify：逐条 notify 会被 showStatus 原地覆盖，只残留最后一条
+	const okCount = results.filter((r) => r.ok).length;
+	const summary = results.map((r) => formatTestResult(r)).join("\n\n") + `\n${provider}: ${okCount}/${results.length} ok`;
+	ctx.ui.notify(summary, "info");
 }

package/forms.ts

@@ -36,6 +36,7 @@ export const DEFAULT_MODEL_CONFIG: {
     contextWindow: number;
     maxTokens: number;
     thinkingLevelMap: ModelConfig["thinkingLevelMap"];
+    compat: { supportsDeveloperRole: boolean };
 } = {
     reasoning: true,
     input: ["text", "image"],
@@ -50,6 +51,8 @@ export const DEFAULT_MODEL_CONFIG: {
         xhigh: null,
         max: null,
     },
+    // Zhipu GLM 等 OpenAI-compat 网关拒收 role:"developer"（会返 422）。默认 false → pi 用 system role。
+    compat: { supportsDeveloperRole: false },
 };
 
 // ============================================================================
@@ -88,13 +91,19 @@ export function ensureDefaultConfigFile(): string | null {
 }
 
 function isValidDefaultModelConfig(v: any): v is typeof DEFAULT_MODEL_CONFIG {
+	const compatOk = !v.compat
+		|| (typeof v.compat === "object" && !Array.isArray(v.compat) && (
+			v.compat.supportsDeveloperRole === undefined
+			|| typeof v.compat.supportsDeveloperRole === "boolean"
+		));
 	return (
 		v && typeof v === "object" &&
 		typeof v.reasoning === "boolean" &&
 		Array.isArray(v.input) && v.input.every((x: any) => x === "text" || x === "image") && v.input.length > 0 &&
 		typeof v.contextWindow === "number" && v.contextWindow > 0 && Number.isFinite(v.contextWindow) &&
 		typeof v.maxTokens === "number" && v.maxTokens > 0 && Number.isFinite(v.maxTokens) &&
-		v.thinkingLevelMap && typeof v.thinkingLevelMap === "object" && !Array.isArray(v.thinkingLevelMap)
+		v.thinkingLevelMap && typeof v.thinkingLevelMap === "object" && !Array.isArray(v.thinkingLevelMap) &&
+		compatOk
 	);
 }
 
@@ -109,7 +118,11 @@ export function loadDefaultModelConfig(): typeof DEFAULT_MODEL_CONFIG {
 		const raw = readFileSync(p, "utf8");
 		const parsed = JSON.parse(raw);
 		const cfg = parsed?.defaultModel;
-		if (isValidDefaultModelConfig(cfg)) return cfg;
+		if (isValidDefaultModelConfig(cfg)) {
+			// 补全缺失的 compat（老 config 没有这个字段时默认为 false）
+			if (!cfg.compat) cfg.compat = { supportsDeveloperRole: false };
+			return cfg;
+		}
 	} catch {
 		// 回退到代码默认
 	}
@@ -363,6 +376,8 @@ export async function editModelFlow(
         { key: "contextWindow", label: "contextWindow", type: "number", validate: (v) => typeof v === "number" && v >= 0 ? null : "must be non-negative" },
         { key: "maxTokens", label: "maxTokens", type: "number", validate: (v) => typeof v === "number" && v >= 0 ? null : "must be non-negative" },
         { key: "thinkingLevelMap", label: "thinkingLevelMap", type: "levelmap", hint: "(empty = remove)" },
+        // Zhipu GLM 等 OpenAI-compat 网关不接受 role:"developer" (会返 422)。默认 no 用 system role。
+        { key: "supportsDeveloperRole", label: "supportsDeveloperRole (compat)", type: "select", options: ["no", "yes"], hint: "Zhipu GLM 等需 no (用 system role)" },
     ];
     const initial: Record<string, unknown> = {
         name: cur.name ?? "",
@@ -371,6 +386,7 @@ export async function editModelFlow(
         contextWindow: cur.contextWindow ?? 0,
         maxTokens: cur.maxTokens ?? 0,
         thinkingLevelMap: cur.thinkingLevelMap ?? null,
+        supportsDeveloperRole: (cur.compat as any)?.supportsDeveloperRole === true ? "yes" : "no",
     };
     const result = await runFormEditor(ctx, `Edit model "${providerId}/${modelId}"`, fields, initial);
     if (!result.saved) { onDone?.(); return; }
@@ -383,6 +399,7 @@ export async function editModelFlow(
         contextWindow: (v.contextWindow as number) || undefined,
         maxTokens: (v.maxTokens as number) || undefined,
         thinkingLevelMap: (v.thinkingLevelMap as Record<string, unknown> | null) ?? undefined,
+        compat: { ...(cur.compat ?? {}), supportsDeveloperRole: v.supportsDeveloperRole === "yes" },
     };
     const newModels = (prov.models ?? []).map((m) => (m.id === modelId ? next : m));
     const newProv: ProviderConfig = { ...prov, models: newModels };

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@fanchaozz/provider-manager",
-  "version": "0.2.0",
+  "version": "0.2.2",
   "description": "A pi extension that manages custom providers and models in ~/.pi/agent/models.json via a TUI dashboard, /providers slash command, and remote model sync.",
   "type": "module",
   "main": "index.ts",
@@ -17,7 +17,9 @@
     "ai"
   ],
   "pi": {
-    "extensions": ["./"]
+    "extensions": [
+      "./"
+    ]
   },
   "license": "MIT",
   "author": "fanchaozz <fanchaozz@users.noreply.github.com>",

package/test.ts

@@ -1,354 +1,354 @@
-/**
- * test.ts — model 可用性测试
- *
- * 3 档检查，每档独立返回 ok：
- *   - auth       modelRegistry.getProviderAuthStatus(provider) → configured? source?
- *   - reachable  GET `${baseUrl}/models`（或 Google 端点）期望 2xx
- *   - generated  modelRegistry.complete(...) 一次最小生成（maxTokens 硬上限 16）
- *
- * mode:
- *   - "quick" → auth + reachable
- *   - "full"  → generated（auth+reachable 也跑）
- *   - "both"  → 等价 "full"
- *
- * 安全：
- *   - 必传 signal，超时 10s（可改）
- *   - maxTokens 硬上限 16（即便 caller 传 1000 也截断）
- *   - 不并发跨 provider
- *   - 失败不重试
- *
- * 内存缓存：Map<"${provider}/${modelId}", TestResult>（会话级，不持久化）
- */
-
-import type { ExtensionContext } from "@earendil-works/pi-coding-agent";
-import { readModelsJson, type ModelsJson } from "./store.ts";
-
-// ============================================================================
-// Types
-// ============================================================================
-
-export type TestMode = "quick" | "full" | "both";
-
-export type CheckResult = {
-	ok: boolean;
-	[key: string]: unknown;
-};
-
-export type TestResult = {
-	provider: string;
-	model: string;
-	mode: TestMode;
-	ok: boolean;
-	latencyMs: number;
-	checks: {
-		auth: { ok: boolean; source?: string; label?: string; error?: string };
-		reachable: { ok: boolean; status?: number; url?: string; error?: string };
-		generated?: {
-			ok: boolean;
-			stopReason?: string;
-			content?: string;       // 截断到 50 字符
-			usage?: { input: number; output: number; cacheRead: number; cacheWrite: number; cost: number };
-			error?: string;
-		};
-	};
-	testedAt: number;
-};
-
-// ============================================================================
-// Cache (session-scoped, not persisted)
-// ============================================================================
-
-const cache = new Map<string, TestResult>();
-const cacheKey = (provider: string, model: string) => `${provider}/${model}`;
-
-export function getCached(provider: string, model: string): TestResult | undefined {
-	return cache.get(cacheKey(provider, model));
-}
-
-export function clearCache(): void {
-	cache.clear();
-}
-
-function storeCached(result: TestResult): void {
-	cache.set(cacheKey(result.provider, result.model), result);
-}
-
-// ============================================================================
-// Auth check
-// ============================================================================
-
-/** 把新旧两种 AuthStatus 形态（{ok, source} 和 {configured, source}）都接受 */
-function isAuthOk(status: any): boolean {
-	if (!status) return false;
-	if (typeof status.ok === "boolean") return status.ok;
-	if (typeof status.configured === "boolean") return status.configured;
-	return false;
-}
-
-function authSource(status: any): string | undefined {
-	return status?.source ?? status?.label;
-}
-
-function authLabel(status: any): string | undefined {
-	return status?.label;
-}
-
-function checkAuth(ctx: ExtensionContext, provider: string): TestResult["checks"]["auth"] {
-	let status: any;
-	try {
-		status = (ctx.modelRegistry as any).getProviderAuthStatus(provider);
-	} catch (err) {
-		return { ok: false, error: `auth check threw: ${err instanceof Error ? err.message : String(err)}` };
-	}
-	if (isAuthOk(status)) {
-		return { ok: true, source: authSource(status), label: authLabel(status) };
-	}
-	const src = authSource(status) ?? "unknown";
-	return { ok: false, source: src, label: authLabel(status), error: `No key configured (source: ${src})` };
-}
-
-// ============================================================================
-// Reachable check
-// ============================================================================
-
-/** 从 models.json 读 baseUrl/api（仅自定义 provider） */
-function readProviderConfig(json: ModelsJson, provider: string): { baseUrl?: string; apiKey?: string; api?: string } {
-	const p = json.providers[provider];
-	if (!p) return {};
-	const out: { baseUrl?: string; apiKey?: string; api?: string } = {};
-	if (typeof p.baseUrl === "string") out.baseUrl = p.baseUrl;
-	if (typeof p.apiKey === "string") out.apiKey = p.apiKey;
-	if (typeof p.api === "string") out.api = p.api;
-	return out;
-}
-
-function probeUrl(baseUrl: string, api: string | undefined, apiKey: string | undefined): { url: string; headers: Record<string, string> } {
-	const base = baseUrl.replace(/\/+$/, "");
-	if (api === "google-generative-ai") {
-		const key = apiKey ? `?key=${encodeURIComponent(apiKey)}` : "";
-		return { url: `${base}/models${key}`, headers: {} };
-	}
-	// OpenAI-compat / anthropic-messages 等统一走 GET /models
-	const headers: Record<string, string> = {};
-	if (apiKey) headers["Authorization"] = `Bearer ${apiKey}`;
-	return { url: `${base}/models`, headers };
-}
-
-async function checkReachable(json: ModelsJson, provider: string, timeoutMs: number, signal?: AbortSignal): Promise<TestResult["checks"]["reachable"]> {
-	const cfg = readProviderConfig(json, provider);
-	if (!cfg.baseUrl) {
-		return { ok: false, error: `provider "${provider}" has no baseUrl in models.json; reachable check skipped` };
-	}
-	const { url, headers } = probeUrl(cfg.baseUrl, cfg.api, cfg.apiKey);
-
-	const ctrl = new AbortController();
-	const timer = setTimeout(() => ctrl.abort(new Error("reachable timeout")), timeoutMs);
-	const onAbort = () => ctrl.abort(signal!.reason);
-	if (signal) signal.addEventListener("abort", onAbort);
-
-	try {
-		const res = await fetch(url, { method: "GET", headers, signal: ctrl.signal });
-		const ok = res.status >= 200 && res.status < 300;
-		// 不读 body，只要状态码
-		try { await res.body?.cancel(); } catch { /* ignore */ }
-		return ok
-			? { ok: true, status: res.status, url }
-			: { ok: false, status: res.status, url, error: `HTTP ${res.status} ${res.statusText}` };
-	} catch (err) {
-		return { ok: false, url, error: err instanceof Error ? err.message : String(err) };
-	} finally {
-		clearTimeout(timer);
-		if (signal) signal.removeEventListener("abort", onAbort);
-	}
-}
-
-// ============================================================================
-// Generated check
-// ============================================================================
-
-const DEFAULT_PROMPT = "Reply with the single word: ok";
-const DEFAULT_MAX_TOKENS = 4;
-const HARD_MAX_TOKENS_CAP = 16;
-const MAX_CONTENT_LEN = 50;
-
-async function checkGenerated(opts: {
-	ctx: ExtensionContext;
-	modelId: string;
-	provider: string;
-	prompt?: string;
-	maxTokens?: number;
-	timeoutMs: number;
-	signal?: AbortSignal;
-}): Promise<TestResult["checks"]["generated"]> {
-	const prompt = opts.prompt ?? DEFAULT_PROMPT;
-	const requested = opts.maxTokens ?? DEFAULT_MAX_TOKENS;
-	const maxTokens = Math.min(Math.max(1, requested), HARD_MAX_TOKENS_CAP);
-
-	// 找到 model 对象（modelRegistry 有 find(provider, id)）
-	const model = (opts.ctx.modelRegistry as any).find?.(opts.provider, opts.modelId);
-	if (!model) {
-		return { ok: false, error: `model ${opts.provider}/${opts.modelId} not found in modelRegistry` };
-	}
-
-	const ctrl = new AbortController();
-	const timer = setTimeout(() => ctrl.abort(new Error("generated timeout")), opts.timeoutMs);
-	const onAbort = () => ctrl.abort(opts.signal!.reason);
-	if (opts.signal) opts.signal.addEventListener("abort", onAbort);
-
-	const context = {
-		systemPrompt: "You are a test probe. Reply concisely.",
-		messages: [{ role: "user", content: prompt, timestamp: Date.now() }],
-	};
-
-	try {
-		const msg = await (opts.ctx.modelRegistry as any).complete(model, context, {
-			signal: ctrl.signal,
-			maxTokens,
-		});
-		const ok = msg?.stopReason === "stop" || msg?.stopReason === "length";
-		const text = (msg?.content ?? [])
-			.filter((c: any) => c?.type === "text")
-			.map((c: any) => c.text)
-			.join("");
-		const content = text.length > MAX_CONTENT_LEN ? text.slice(0, MAX_CONTENT_LEN) + "..." : text;
-		const usage = msg?.usage ? {
-			input: msg.usage.input ?? 0,
-			output: msg.usage.output ?? 0,
-			cacheRead: msg.usage.cacheRead ?? 0,
-			cacheWrite: msg.usage.cacheWrite ?? 0,
-			cost: msg.usage.cost?.total ?? 0,
-		} : undefined;
-		return ok
-			? { ok: true, stopReason: msg.stopReason, content, usage }
-			: { ok: false, stopReason: msg?.stopReason, content, error: msg?.errorMessage ?? `stopReason=${msg?.stopReason}` };
-	} catch (err) {
-		return { ok: false, error: err instanceof Error ? err.message : String(err) };
-	} finally {
-		clearTimeout(timer);
-		if (opts.signal) opts.signal.removeEventListener("abort", onAbort);
-	}
-}
-
-// ============================================================================
-// Public API
-// ============================================================================
-
-export type TestModelOpts = {
-	ctx: ExtensionContext;
-	provider: string;
-	model: string;
-	mode?: TestMode;       // default "quick"
-	prompt?: string;
-	maxTokens?: number;
-	timeoutMs?: number;    // default 10000
-	signal?: AbortSignal;
-};
-
-/** 测单个 model；结果写入 cache */
-export async function testModel(opts: TestModelOpts): Promise<TestResult> {
-	const mode = opts.mode ?? "quick";
-	const timeoutMs = opts.timeoutMs ?? parseInt(process.env.PI_PROVIDER_TEST_TIMEOUT ?? "10000", 10);
-
-	const start = Date.now();
-	const json = await readModelsJson();
-
-	const checks: TestResult["checks"] = {
-		auth: checkAuth(opts.ctx, opts.provider),
-		reachable: { ok: false, error: "skipped" },
-	};
-
-	if (checks.auth.ok) {
-		checks.reachable = await checkReachable(json, opts.provider, timeoutMs, opts.signal);
-	} else {
-		checks.reachable = { ok: false, error: "skipped (auth failed)" };
-	}
-
-	if (mode === "full" || mode === "both") {
-		if (checks.auth.ok && checks.reachable.ok) {
-			checks.generated = await checkGenerated({
-				ctx: opts.ctx,
-				provider: opts.provider,
-				modelId: opts.model,
-				prompt: opts.prompt,
-				maxTokens: opts.maxTokens,
-				timeoutMs,
-				signal: opts.signal,
-			});
-		} else {
-			checks.generated = { ok: false, error: "skipped (auth or reachable failed)" };
-		}
-	}
-
-	const ok = checks.auth.ok && checks.reachable.ok && (checks.generated?.ok ?? true);
-	const result: TestResult = {
-		provider: opts.provider,
-		model: opts.model,
-		mode,
-		ok,
-		latencyMs: Date.now() - start,
-		checks,
-		testedAt: Date.now(),
-	};
-	storeCached(result);
-	return result;
-}
-
-export type TestProviderOpts = {
-	ctx: ExtensionContext;
-	provider: string;
-	modelIds: string[];
-	mode?: TestMode;
-	concurrency?: number;  // default 3
-	timeoutMs?: number;
-	signal?: AbortSignal;
-	onProgress?: (done: number, total: number, result: TestResult) => void;
-};
-
-/** 批量测 provider 下多个 model；并发度默认 3 */
-export async function testProvider(opts: TestProviderOpts): Promise<TestResult[]> {
-	const concurrency = Math.max(1, opts.concurrency ?? 3);
-	const total = opts.modelIds.length;
-	const out: TestResult[] = [];
-	let cursor = 0;
-	let done = 0;
-
-	async function worker(): Promise<void> {
-		while (cursor < total) {
-			const i = cursor++;
-			const modelId = opts.modelIds[i];
-			const result = await testModel({
-				ctx: opts.ctx,
-				provider: opts.provider,
-				model: modelId,
-				mode: opts.mode,
-				timeoutMs: opts.timeoutMs,
-				signal: opts.signal,
-			});
-			out[i] = result;
-			done++;
-			opts.onProgress?.(done, total, result);
-		}
-	}
-
-	const workers = Array.from({ length: Math.min(concurrency, total) }, () => worker());
-	await Promise.all(workers);
-	return out;
-}
-
-// ============================================================================
-// Display helpers
-// ============================================================================
-
-/** 把 TestResult 格式化成多行可读文本（detail panel / notify） */
-export function formatTestResult(r: TestResult): string {
-	const lines: string[] = [];
-	lines.push(`${r.provider}/${r.model}  ${r.ok ? "✓ ok" : "✗ fail"}  (${r.latencyMs}ms, mode=${r.mode})`);
-	lines.push(`  auth      ${r.checks.auth.ok ? "✓" : "✗"} ${r.checks.auth.source ?? ""} ${r.checks.auth.error ? "— " + r.checks.auth.error : ""}`);
-	lines.push(`  reachable ${r.checks.reachable.ok ? "✓" : "✗"} ${r.checks.reachable.status ?? ""} ${r.checks.reachable.error ? "— " + r.checks.reachable.error : ""}`);
-	if (r.checks.generated) {
-		const g = r.checks.generated;
-		lines.push(`  generated ${g.ok ? "✓" : "✗"} ${g.stopReason ?? ""} ${g.content ? `— "${g.content}"` : ""} ${g.error ? "— " + g.error : ""}`);
-		if (g.usage) lines.push(`            usage: in=${g.usage.input} out=${g.usage.output} cost=$${g.usage.cost.toFixed(6)}`);
-	}
-	return lines.join("\n");
-}
+/**
+ * test.ts — model 可用性测试
+ *
+ * 3 档检查，每档独立返回 ok：
+ *   - auth       modelRegistry.getProviderAuthStatus(provider) → configured? source?
+ *   - reachable  GET `${baseUrl}/models`（或 Google 端点）期望 2xx
+ *   - generated  modelRegistry.complete(...) 一次最小生成（maxTokens 硬上限 16）
+ *
+ * mode:
+ *   - "quick" → auth + reachable
+ *   - "full"  → generated（auth+reachable 也跑）
+ *   - "both"  → 等价 "full"
+ *
+ * 安全：
+ *   - 必传 signal，超时 10s（可改）
+ *   - maxTokens 硬上限 16（即便 caller 传 1000 也截断）
+ *   - 不并发跨 provider
+ *   - 失败不重试
+ *
+ * 内存缓存：Map<"${provider}/${modelId}", TestResult>（会话级，不持久化）
+ */
+
+import type { ExtensionContext } from "@earendil-works/pi-coding-agent";
+import { readModelsJson, type ModelsJson } from "./store.ts";
+
+// ============================================================================
+// Types
+// ============================================================================
+
+export type TestMode = "quick" | "full" | "both";
+
+export type CheckResult = {
+	ok: boolean;
+	[key: string]: unknown;
+};
+
+export type TestResult = {
+	provider: string;
+	model: string;
+	mode: TestMode;
+	ok: boolean;
+	latencyMs: number;
+	checks: {
+		auth: { ok: boolean; source?: string; label?: string; error?: string };
+		reachable: { ok: boolean; status?: number; url?: string; error?: string };
+		generated?: {
+			ok: boolean;
+			stopReason?: string;
+			content?: string;       // 截断到 50 字符
+			usage?: { input: number; output: number; cacheRead: number; cacheWrite: number; cost: number };
+			error?: string;
+		};
+	};
+	testedAt: number;
+};
+
+// ============================================================================
+// Cache (session-scoped, not persisted)
+// ============================================================================
+
+const cache = new Map<string, TestResult>();
+const cacheKey = (provider: string, model: string) => `${provider}/${model}`;
+
+export function getCached(provider: string, model: string): TestResult | undefined {
+	return cache.get(cacheKey(provider, model));
+}
+
+export function clearCache(): void {
+	cache.clear();
+}
+
+function storeCached(result: TestResult): void {
+	cache.set(cacheKey(result.provider, result.model), result);
+}
+
+// ============================================================================
+// Auth check
+// ============================================================================
+
+/** 把新旧两种 AuthStatus 形态（{ok, source} 和 {configured, source}）都接受 */
+function isAuthOk(status: any): boolean {
+	if (!status) return false;
+	if (typeof status.ok === "boolean") return status.ok;
+	if (typeof status.configured === "boolean") return status.configured;
+	return false;
+}
+
+function authSource(status: any): string | undefined {
+	return status?.source ?? status?.label;
+}
+
+function authLabel(status: any): string | undefined {
+	return status?.label;
+}
+
+function checkAuth(ctx: ExtensionContext, provider: string): TestResult["checks"]["auth"] {
+	let status: any;
+	try {
+		status = (ctx.modelRegistry as any).getProviderAuthStatus(provider);
+	} catch (err) {
+		return { ok: false, error: `auth check threw: ${err instanceof Error ? err.message : String(err)}` };
+	}
+	if (isAuthOk(status)) {
+		return { ok: true, source: authSource(status), label: authLabel(status) };
+	}
+	const src = authSource(status) ?? "unknown";
+	return { ok: false, source: src, label: authLabel(status), error: `No key configured (source: ${src})` };
+}
+
+// ============================================================================
+// Reachable check
+// ============================================================================
+
+/** 从 models.json 读 baseUrl/api（仅自定义 provider） */
+function readProviderConfig(json: ModelsJson, provider: string): { baseUrl?: string; apiKey?: string; api?: string } {
+	const p = json.providers[provider];
+	if (!p) return {};
+	const out: { baseUrl?: string; apiKey?: string; api?: string } = {};
+	if (typeof p.baseUrl === "string") out.baseUrl = p.baseUrl;
+	if (typeof p.apiKey === "string") out.apiKey = p.apiKey;
+	if (typeof p.api === "string") out.api = p.api;
+	return out;
+}
+
+function probeUrl(baseUrl: string, api: string | undefined, apiKey: string | undefined): { url: string; headers: Record<string, string> } {
+	const base = baseUrl.replace(/\/+$/, "");
+	if (api === "google-generative-ai") {
+		const key = apiKey ? `?key=${encodeURIComponent(apiKey)}` : "";
+		return { url: `${base}/models${key}`, headers: {} };
+	}
+	// OpenAI-compat / anthropic-messages 等统一走 GET /models
+	const headers: Record<string, string> = {};
+	if (apiKey) headers["Authorization"] = `Bearer ${apiKey}`;
+	return { url: `${base}/models`, headers };
+}
+
+async function checkReachable(json: ModelsJson, provider: string, timeoutMs: number, signal?: AbortSignal): Promise<TestResult["checks"]["reachable"]> {
+	const cfg = readProviderConfig(json, provider);
+	if (!cfg.baseUrl) {
+		return { ok: false, error: `provider "${provider}" has no baseUrl in models.json; reachable check skipped` };
+	}
+	const { url, headers } = probeUrl(cfg.baseUrl, cfg.api, cfg.apiKey);
+
+	const ctrl = new AbortController();
+	const timer = setTimeout(() => ctrl.abort(new Error("reachable timeout")), timeoutMs);
+	const onAbort = () => ctrl.abort(signal!.reason);
+	if (signal) signal.addEventListener("abort", onAbort);
+
+	try {
+		const res = await fetch(url, { method: "GET", headers, signal: ctrl.signal });
+		const ok = res.status >= 200 && res.status < 300;
+		// 不读 body，只要状态码
+		try { await res.body?.cancel(); } catch { /* ignore */ }
+		return ok
+			? { ok: true, status: res.status, url }
+			: { ok: false, status: res.status, url, error: `HTTP ${res.status} ${res.statusText}` };
+	} catch (err) {
+		return { ok: false, url, error: err instanceof Error ? err.message : String(err) };
+	} finally {
+		clearTimeout(timer);
+		if (signal) signal.removeEventListener("abort", onAbort);
+	}
+}
+
+// ============================================================================
+// Generated check
+// ============================================================================
+
+const DEFAULT_PROMPT = "Reply with the single word: ok";
+const DEFAULT_MAX_TOKENS = 4;
+const HARD_MAX_TOKENS_CAP = 16;
+const MAX_CONTENT_LEN = 50;
+
+async function checkGenerated(opts: {
+	ctx: ExtensionContext;
+	modelId: string;
+	provider: string;
+	prompt?: string;
+	maxTokens?: number;
+	timeoutMs: number;
+	signal?: AbortSignal;
+}): Promise<TestResult["checks"]["generated"]> {
+	const prompt = opts.prompt ?? DEFAULT_PROMPT;
+	const requested = opts.maxTokens ?? DEFAULT_MAX_TOKENS;
+	const maxTokens = Math.min(Math.max(1, requested), HARD_MAX_TOKENS_CAP);
+
+	// 找到 model 对象（modelRegistry 有 find(provider, id)）
+	const model = (opts.ctx.modelRegistry as any).find?.(opts.provider, opts.modelId);
+	if (!model) {
+		return { ok: false, error: `model ${opts.provider}/${opts.modelId} not found in modelRegistry` };
+	}
+
+	const ctrl = new AbortController();
+	const timer = setTimeout(() => ctrl.abort(new Error("generated timeout")), opts.timeoutMs);
+	const onAbort = () => ctrl.abort(opts.signal!.reason);
+	if (opts.signal) opts.signal.addEventListener("abort", onAbort);
+
+	const context = {
+		systemPrompt: "You are a test probe. Reply concisely.",
+		messages: [{ role: "user", content: prompt, timestamp: Date.now() }],
+	};
+
+	try {
+		const msg = await (opts.ctx.modelRegistry as any).complete(model, context, {
+			signal: ctrl.signal,
+			maxTokens,
+		});
+		const ok = msg?.stopReason === "stop" || msg?.stopReason === "length";
+		const text = (msg?.content ?? [])
+			.filter((c: any) => c?.type === "text")
+			.map((c: any) => c.text)
+			.join("");
+		const content = text.length > MAX_CONTENT_LEN ? text.slice(0, MAX_CONTENT_LEN) + "..." : text;
+		const usage = msg?.usage ? {
+			input: msg.usage.input ?? 0,
+			output: msg.usage.output ?? 0,
+			cacheRead: msg.usage.cacheRead ?? 0,
+			cacheWrite: msg.usage.cacheWrite ?? 0,
+			cost: msg.usage.cost?.total ?? 0,
+		} : undefined;
+		return ok
+			? { ok: true, stopReason: msg.stopReason, content, usage }
+			: { ok: false, stopReason: msg?.stopReason, content, error: msg?.errorMessage ?? `stopReason=${msg?.stopReason}` };
+	} catch (err) {
+		return { ok: false, error: err instanceof Error ? err.message : String(err) };
+	} finally {
+		clearTimeout(timer);
+		if (opts.signal) opts.signal.removeEventListener("abort", onAbort);
+	}
+}
+
+// ============================================================================
+// Public API
+// ============================================================================
+
+export type TestModelOpts = {
+	ctx: ExtensionContext;
+	provider: string;
+	model: string;
+	mode?: TestMode;       // default "quick"
+	prompt?: string;
+	maxTokens?: number;
+	timeoutMs?: number;    // default 10000
+	signal?: AbortSignal;
+};
+
+/** 测单个 model；结果写入 cache */
+export async function testModel(opts: TestModelOpts): Promise<TestResult> {
+	const mode = opts.mode ?? "quick";
+	const timeoutMs = opts.timeoutMs ?? parseInt(process.env.PI_PROVIDER_TEST_TIMEOUT ?? "10000", 10);
+
+	const start = Date.now();
+	const json = await readModelsJson();
+
+	const checks: TestResult["checks"] = {
+		auth: checkAuth(opts.ctx, opts.provider),
+		reachable: { ok: false, error: "skipped" },
+	};
+
+	if (checks.auth.ok) {
+		checks.reachable = await checkReachable(json, opts.provider, timeoutMs, opts.signal);
+	} else {
+		checks.reachable = { ok: false, error: "skipped (auth failed)" };
+	}
+
+	if (mode === "full" || mode === "both") {
+		if (checks.auth.ok && checks.reachable.ok) {
+			checks.generated = await checkGenerated({
+				ctx: opts.ctx,
+				provider: opts.provider,
+				modelId: opts.model,
+				prompt: opts.prompt,
+				maxTokens: opts.maxTokens,
+				timeoutMs,
+				signal: opts.signal,
+			});
+		} else {
+			checks.generated = { ok: false, error: "skipped (auth or reachable failed)" };
+		}
+	}
+
+	const ok = checks.auth.ok && checks.reachable.ok && (checks.generated?.ok ?? true);
+	const result: TestResult = {
+		provider: opts.provider,
+		model: opts.model,
+		mode,
+		ok,
+		latencyMs: Date.now() - start,
+		checks,
+		testedAt: Date.now(),
+	};
+	storeCached(result);
+	return result;
+}
+
+export type TestProviderOpts = {
+	ctx: ExtensionContext;
+	provider: string;
+	modelIds: string[];
+	mode?: TestMode;
+	concurrency?: number;  // default 3
+	timeoutMs?: number;
+	signal?: AbortSignal;
+	onProgress?: (done: number, total: number, result: TestResult) => void;
+};
+
+/** 批量测 provider 下多个 model；并发度默认 3 */
+export async function testProvider(opts: TestProviderOpts): Promise<TestResult[]> {
+	const concurrency = Math.max(1, opts.concurrency ?? 3);
+	const total = opts.modelIds.length;
+	const out: TestResult[] = [];
+	let cursor = 0;
+	let done = 0;
+
+	async function worker(): Promise<void> {
+		while (cursor < total) {
+			const i = cursor++;
+			const modelId = opts.modelIds[i];
+			const result = await testModel({
+				ctx: opts.ctx,
+				provider: opts.provider,
+				model: modelId,
+				mode: opts.mode,
+				timeoutMs: opts.timeoutMs,
+				signal: opts.signal,
+			});
+			out[i] = result;
+			done++;
+			opts.onProgress?.(done, total, result);
+		}
+	}
+
+	const workers = Array.from({ length: Math.min(concurrency, total) }, () => worker());
+	await Promise.all(workers);
+	return out;
+}
+
+// ============================================================================
+// Display helpers
+// ============================================================================
+
+/** 把 TestResult 格式化成多行可读文本（detail panel / notify） */
+export function formatTestResult(r: TestResult): string {
+	const lines: string[] = [];
+	lines.push(`${r.provider}/${r.model}  ${r.ok ? "✓ ok" : "✗ fail"}  (${r.latencyMs}ms, mode=${r.mode})`);
+	lines.push(`  auth      ${r.checks.auth.ok ? "✓" : "✗"} ${r.checks.auth.source ?? ""} ${r.checks.auth.error ? "— " + r.checks.auth.error : ""}`);
+	lines.push(`  reachable ${r.checks.reachable.ok ? "✓" : "✗"} ${r.checks.reachable.status ?? ""} ${r.checks.reachable.error ? "— " + r.checks.reachable.error : ""}`);
+	if (r.checks.generated) {
+		const g = r.checks.generated;
+		lines.push(`  generated ${g.ok ? "✓" : "✗"} ${g.stopReason ?? ""} ${g.content ? `— "${g.content}"` : ""} ${g.error ? "— " + g.error : ""}`);
+		if (g.usage) lines.push(`            usage: in=${g.usage.input} out=${g.usage.output} cost=$${g.usage.cost.toFixed(6)}`);
+	}
+	return lines.join("\n");
+}

package/ui.ts

@@ -35,6 +35,7 @@ type ModelRow = {
 	// 详情面板需要从 raw ModelConfig 透传
 	thinkingLevelMap?: Partial<Record<"off" | "minimal" | "low" | "medium" | "high" | "xhigh" | "max", string | null>>;
 	cost?: { input: number; output: number; cacheRead: number; cacheWrite: number };
+	compat?: Record<string, unknown>;
 };
 
 type ProviderRow = {
@@ -209,6 +210,7 @@ function buildProviders(ctx: ExtensionCommandContext, json: ModelsJson): { provi
 				// 详情面板需要这些字段
 				thinkingLevelMap: m.thinkingLevelMap,
 				cost: m.cost,
+				compat: m.compat,
 			})),
 		};
 	});
@@ -414,9 +416,10 @@ class Dashboard {
 			let okCount = 0;
 			for (const r of results) {
 				if (r.ok) okCount++;
-				ctx.ui.notify(formatTestResult(r), r.ok ? "info" : "warning");
 			}
-			ctx.ui.notify(`${provider.id}: ${okCount}/${results.length} ok`, okCount === results.length ? "success" : "warning");
+			// 批量结果拼成一条 notify：逐条 notify 会被 showStatus 原地覆盖，只残留汇总行
+			const summary = results.map((r) => formatTestResult(r)).join("\n\n") + `\n${provider.id}: ${okCount}/${results.length} ok`;
+			ctx.ui.notify(summary, "info");
 		} else {
 			// t: 测当前 pane 的 model（provider pane 测第一个 model；model pane 测当前 model）
 			let modelId: string | undefined;
@@ -428,7 +431,8 @@ class Dashboard {
 			if (!modelId) { ctx.ui.notify(`${provider.id} 无 model`, "warning"); return; }
 			ctx.ui.notify(`testing ${provider.id}/${modelId}...`, "info");
 			const r = await testModel({ ctx, provider: provider.id, model: modelId, mode: "full" });
-			ctx.ui.notify(formatTestResult(r), r.ok ? "success" : "warning");
+			// 同上：统一 info 避免滞留
+			ctx.ui.notify(formatTestResult(r), "info");
 		}
 		this.invalidate();
 	}
@@ -660,6 +664,16 @@ class Dashboard {
 				if (cost.cacheRead) lines.push(`    cache read:   $${cost.cacheRead}/M`);
 				if (cost.cacheWrite) lines.push(`    cache write:  $${cost.cacheWrite}/M`);
 			}
+			// compat：Zhipu GLM 等 OpenAI-compat 网关拒收 role:"developer"（会返 422）。为 false 时 pi 用 system role。
+			const compat = m.compat;
+			if (compat && typeof compat === "object") {
+				lines.push("");
+				lines.push(th.fg("muted", "  Compat"));
+				if (typeof (compat as any).supportsDeveloperRole === "boolean") {
+					const sdr = (compat as any).supportsDeveloperRole;
+					lines.push(`    supportsDeveloperRole: ${sdr ? th.fg("success", "yes") : th.fg("warning", "no")}`);
+				}
+			}
 		}
 		return lines.map((l) => truncateToWidth(l, width));
 	}