@hasna/switcher 0.1.1 → 0.1.3

package/dist/sdk.js

@@ -1,10 +1,56 @@
 // src/domain.ts
+import { z as z2 } from "zod";
+
+// src/model-policy-schema.ts
 import { z } from "zod";
-var harnessSchema = z.enum(["claude", "codex", "grok", "opencode2", "pi"]);
-var protocolSchema = z.enum(["anthropic-messages", "openai-responses", "openai-chat"]);
-var idSchema = z.string().regex(/^[a-zA-Z0-9][a-zA-Z0-9._-]{0,79}$/);
-var label = z.string().min(1).max(200);
-var envRef = z.string().regex(/^SWITCHER_PROVIDER_[A-Z0-9_]+$/);
+var policyModelIdSchema = z.string().min(1).max(300).regex(/^[^\u0000-\u001f\u007f]+$/);
+var modelPolicyRoleSchema = z.enum(["subagent", "fast", "planning", "review", "summary", "compaction", "weak", "editor"]);
+var boundedModelList = z.array(policyModelIdSchema).max(500);
+var aliasSchema = z.string().regex(/^[A-Za-z0-9._/-]{1,120}$/).refine((v) => !["__proto__", "prototype", "constructor"].includes(v));
+var boundedModelMap = z.record(aliasSchema, policyModelIdSchema).superRefine((value, ctx) => {
+  if (Object.keys(value).length > 200)
+    ctx.addIssue({ code: "custom", message: "Model policy maps may contain at most 200 entries." });
+});
+var fallbacksSchema = z.record(policyModelIdSchema, z.array(policyModelIdSchema).max(20)).superRefine((value, ctx) => {
+  if (Object.keys(value).length > 200)
+    ctx.addIssue({ code: "custom", message: "Model policy fallback maps may contain at most 200 entries." });
+});
+var modelPolicySchema = z.object({
+  version: z.literal(1).default(1),
+  roles: z.object({
+    subagent: policyModelIdSchema.optional(),
+    fast: policyModelIdSchema.optional(),
+    planning: policyModelIdSchema.optional(),
+    review: policyModelIdSchema.optional(),
+    summary: policyModelIdSchema.optional(),
+    compaction: policyModelIdSchema.optional(),
+    weak: policyModelIdSchema.optional(),
+    editor: policyModelIdSchema.optional()
+  }).strict().optional(),
+  allowedModels: boundedModelList.optional(),
+  aliases: boundedModelMap.optional(),
+  fallbacks: fallbacksSchema.optional()
+}).strict();
+var routingDecisionSchema = z.enum(["allow", "alias", "reject", "fallback"]);
+var routingEventRoleSchema = z.enum(["main", ...modelPolicyRoleSchema.options]);
+var routingEventSchema = z.object({
+  at: z.string().datetime({ offset: true }),
+  requestId: z.string().regex(/^[A-Za-z0-9][A-Za-z0-9._:-]{0,127}$/),
+  requestedModel: policyModelIdSchema,
+  resolvedModel: policyModelIdSchema.optional(),
+  reportedModel: policyModelIdSchema.optional(),
+  decision: routingDecisionSchema,
+  role: routingEventRoleSchema.optional(),
+  reason: z.string().regex(/^[a-z][a-z0-9_]{0,63}$/).optional(),
+  upstreamStatus: z.number().int().min(100).max(599).optional()
+}).strict();
+var routingEventsSchema = z.array(routingEventSchema).max(1000);
+// src/domain.ts
+var harnessSchema = z2.enum(["claude", "codex", "grok", "opencode", "opencode2", "pi", "omp", "dsh", "cline", "hermes", "prime-agent", "gemini", "aider", "kilo"]);
+var protocolSchema = z2.enum(["anthropic-messages", "openai-responses", "openai-chat", "gemini-generate-content"]);
+var idSchema = z2.string().regex(/^[a-zA-Z0-9][a-zA-Z0-9._-]{0,79}$/);
+var label = z2.string().min(1).max(200);
+var envRef = z2.string().regex(/^SWITCHER_PROVIDER_[A-Z0-9_]+$/);
 function endpoint(value) {
   let url;
   try {
@@ -17,73 +63,79 @@ function endpoint(value) {
     throw new Fault(400, "invalid_url", "URL must use HTTPS, contain no credentials/query/fragment, or use HTTP on loopback.");
   return url.href.replace(/\/+$/, "");
 }
-var urlSchema = z.string().max(2000).superRefine((v, ctx) => {
+var urlSchema = z2.string().max(2000).superRefine((v, ctx) => {
   try {
     endpoint(v);
   } catch {
     ctx.addIssue({ code: "custom", message: "Invalid endpoint URL" });
   }
 }).transform(endpoint);
-var modelSchema = z.object({
-  id: z.string().min(1).max(300),
+var modelSchema = z2.object({
+  id: z2.string().min(1).max(300),
   name: label,
-  description: z.string().max(8000).optional(),
-  available: z.boolean().optional(),
-  contextWindow: z.number().int().positive().optional(),
-  maxOutputTokens: z.number().int().positive().optional(),
-  inputModalities: z.array(z.string().max(50)).max(20).optional(),
-  outputModalities: z.array(z.string().max(50)).max(20).optional(),
-  supportedParameters: z.array(z.string().max(100)).max(100).optional()
+  description: z2.string().max(8000).optional(),
+  available: z2.boolean().optional(),
+  contextWindow: z2.number().int().positive().optional(),
+  maxOutputTokens: z2.number().int().positive().optional(),
+  inputModalities: z2.array(z2.string().max(50)).max(20).optional(),
+  outputModalities: z2.array(z2.string().max(50)).max(20).optional(),
+  supportedParameters: z2.array(z2.string().max(100)).max(100).optional(),
+  supportedGenerationMethods: z2.array(z2.string().min(1).max(100)).max(100).optional()
 }).strict();
-var providerInputSchema = z.object({
+var providerInputSchema = z2.object({
   id: idSchema,
   name: label,
   baseUrl: urlSchema,
   protocol: protocolSchema,
   credentialEnv: envRef.optional(),
-  authStyle: z.enum(["bearer", "x-api-key"]).default("bearer"),
+  authStyle: z2.enum(["bearer", "x-api-key", "api-key"]).default("bearer"),
   catalogBaseUrl: urlSchema.optional(),
-  catalogFormat: z.enum(["openai", "ollama", "mistral", "together", "fireworks", "dashscope", "none"]).optional(),
-  catalogAuthStyle: z.enum(["bearer", "x-api-key", "none"]).optional(),
+  catalogFormat: z2.enum(["openai", "ollama", "mistral", "together", "fireworks", "dashscope", "gemini", "none"]).optional(),
+  catalogAuthStyle: z2.enum(["bearer", "x-api-key", "api-key", "none"]).optional(),
   catalogCredentialEnv: envRef.optional(),
-  catalogAccountId: z.string().regex(/^[A-Za-z0-9][A-Za-z0-9._-]{0,127}$/).optional(),
-  modelsPath: z.string().regex(/^[a-zA-Z0-9_/-]+$/).max(200).default("models"),
-  manualModels: z.array(modelSchema).max(1e4).default([])
+  catalogAccountId: z2.string().regex(/^[A-Za-z0-9][A-Za-z0-9._-]{0,127}$/).optional(),
+  modelsPath: z2.string().regex(/^[a-zA-Z0-9_/-]+$/).max(200).default("models"),
+  manualModels: z2.array(modelSchema).max(1e4).default([])
 }).strict().refine((p) => !p.modelsPath.split("/").includes("..") && !p.modelsPath.startsWith("/"), "modelsPath must be relative");
-var providerPresetSchema = z.object({
+var providerPresetSchema = z2.object({
   id: idSchema,
   name: label,
   credentialEnv: envRef.optional(),
-  credentialAliases: z.array(z.string().regex(/^[A-Z][A-Z0-9_]+$/)),
-  protocols: z.array(z.object({
+  credentialAliases: z2.array(z2.string().regex(/^[A-Z][A-Z0-9_]+$/)),
+  protocols: z2.array(z2.object({
     protocol: protocolSchema,
     baseUrl: urlSchema.optional(),
-    authStyle: z.enum(["bearer", "x-api-key"]),
+    authStyle: z2.enum(["bearer", "x-api-key", "api-key"]),
     catalogBaseUrl: urlSchema.optional(),
-    catalogFormat: z.enum(["openai", "ollama", "mistral", "together", "fireworks", "dashscope", "none"]),
-    catalogAuthStyle: z.enum(["bearer", "x-api-key", "none"]).optional(),
-    modelsPath: z.string(),
-    notes: z.array(z.string())
+    catalogFormat: z2.enum(["openai", "ollama", "mistral", "together", "fireworks", "dashscope", "gemini", "none"]),
+    catalogAuthStyle: z2.enum(["bearer", "x-api-key", "api-key", "none"]).optional(),
+    modelsPath: z2.string(),
+    notes: z2.array(z2.string())
   }).strict()).min(1),
-  sources: z.array(z.string().url()),
-  verification: z.literal("documented")
+  sources: z2.array(z2.string().url()),
+  verification: z2.literal("documented")
 }).strict();
-var profileInputSchema = z.object({
+var profileInputSchema = z2.object({
   id: idSchema,
   name: label,
   providerId: idSchema,
   harness: harnessSchema,
-  model: z.string().min(1).max(300)
+  model: z2.string().min(1).max(300),
+  modelPolicy: modelPolicySchema.optional()
 }).strict();
-var runInputSchema = z.object({
+var runInputSchema = z2.object({
+  modelPolicyVersion: z2.literal(1),
   profileId: idSchema,
   harness: harnessSchema,
-  model: z.string().min(1).max(300),
-  planToken: z.string().regex(/^[a-f0-9]{64}$/)
+  model: z2.string().min(1).max(300),
+  modelPolicy: modelPolicySchema.optional(),
+  planToken: z2.string().regex(/^[a-f0-9]{64}$/)
 }).strict();
-var runUpdateSchema = z.object({
-  status: z.enum(["exited", "failed", "interrupted"]),
-  exitCode: z.number().int().min(0).max(255)
+var runUpdateSchema = z2.object({
+  status: z2.enum(["exited", "failed", "interrupted"]),
+  exitCode: z2.number().int().min(0).max(255),
+  routingEvents: routingEventsSchema.optional(),
+  routingEventsDropped: z2.number().int().min(0).max(1e6).optional()
 }).strict();
 
 class Fault extends Error {
@@ -102,7 +154,15 @@ function parse(schema, value) {
   return result.data;
 }
 function compatible(harness, protocol) {
-  return harness === "claude" ? protocol === "anthropic-messages" : harness === "codex" ? protocol === "openai-responses" : true;
+  if (harness === "claude")
+    return protocol === "anthropic-messages";
+  if (harness === "codex")
+    return protocol === "openai-responses";
+  if (harness === "gemini")
+    return protocol === "gemini-generate-content";
+  if (protocol === "gemini-generate-content")
+    return false;
+  return true;
 }
 
 // src/http.ts
@@ -168,7 +228,18 @@ var providerPresets = [
   ], ["https://api-docs.deepseek.com/guides/anthropic_api", "https://api-docs.deepseek.com/api/list-models"], "DEEPSEEK_API_KEY"),
   preset("openrouter", "OpenRouter", ["openai-chat", "openai-responses", "anthropic-messages"].map((protocol) => route(protocol, "https://openrouter.ai/api/v1", { catalogAuthStyle: "none" })), ["https://openrouter.ai/docs/api/api-reference/models/list-all-models-and-their-properties", "https://openrouter.ai/docs/guides/overview"], "OPENROUTER_API_KEY"),
   preset("anthropic", "Anthropic", [route("anthropic-messages", "https://api.anthropic.com/v1", { authStyle: "x-api-key" })], ["https://platform.claude.com/docs/en/api/overview", "https://platform.claude.com/docs/en/api/models/list"], "ANTHROPIC_API_KEY"),
+  preset("gemini", "Google Gemini", [route("gemini-generate-content", "https://generativelanguage.googleapis.com/v1beta", {
+    authStyle: "x-api-key",
+    catalogBaseUrl: "https://generativelanguage.googleapis.com/v1beta",
+    catalogFormat: "gemini",
+    catalogAuthStyle: "x-api-key",
+    notes: ["Gemini CLI uses the native generateContent wire with x-goog-api-key authentication; model IDs are returned as models/{id}."]
+  }), route("openai-chat", "https://generativelanguage.googleapis.com/v1beta/openai")], ["https://ai.google.dev/api", "https://ai.google.dev/api/models", "https://github.com/google-gemini/gemini-cli", "https://ai.google.dev/gemini-api/docs/openai"], "GEMINI_API_KEY"),
   preset("openai", "OpenAI", [route("openai-responses", "https://api.openai.com/v1"), route("openai-chat", "https://api.openai.com/v1")], ["https://platform.openai.com/docs/api-reference/introduction", "https://platform.openai.com/docs/api-reference/models/list"], "OPENAI_API_KEY"),
+  preset("azure-openai", "Azure OpenAI (v1)", [
+    route("openai-responses", undefined, { authStyle: "api-key", catalogFormat: "none", notes: ["Pass the Azure OpenAI v1 resource endpoint ending in /openai/v1. The request model is your deployment name. Azure's model-definition list is not a deployment catalog, so configure manual deployment models or an explicit deployment catalog parser; Switcher does not synthesize deployment paths or api-version query parameters."] }),
+    route("openai-chat", undefined, { authStyle: "api-key", catalogFormat: "none", notes: ["Pass the Azure OpenAI v1 resource endpoint ending in /openai/v1. Chat Completions is POST /chat/completions and accepts the literal api-key header. The request model is your deployment name; configure manual deployment models or an explicit deployment catalog parser because GET /models does not establish deployment names."] })
+  ], ["https://learn.microsoft.com/en-us/rest/api/aifoundry/azureopenai/models", "https://learn.microsoft.com/en-us/rest/api/microsoft-foundry/azureopenai/chat", "https://learn.microsoft.com/en-us/rest/api/aifoundry/azureopenai/responses"], "AZURE_OPENAI_API_KEY"),
   preset("xai", "xAI", ["openai-chat", "openai-responses", "anthropic-messages"].map((protocol) => route(protocol, "https://api.x.ai/v1")), ["https://api.x.ai/docs/", "https://docs.x.ai/developers/model-capabilities/text/generate-text"], "XAI_API_KEY"),
   preset("ollama", "Ollama", ["openai-chat", "openai-responses"].map((protocol) => route(protocol, "http://127.0.0.1:11434/v1", {
     catalogBaseUrl: "http://127.0.0.1:11434",
@@ -229,6 +300,8 @@ function providerFromPreset(presetId, options = {}) {
   const baseUrl = options.baseUrl ?? selected.baseUrl;
   if (!baseUrl)
     throw new Fault(400, "endpoint_required", "This preset requires an explicit --url for its inference endpoint.");
+  if (presetId === "azure-openai" && !/\/openai\/v1$/.test(endpoint(baseUrl)))
+    throw new Fault(400, "invalid_url", "Azure OpenAI v1 requires an explicit endpoint ending in /openai/v1; deployment and api-version URLs are unsupported.");
   if (options.baseUrl && selected.baseUrl && new URL(endpoint(options.baseUrl)).origin !== new URL(selected.baseUrl).origin && preset2.credentialEnv && !options.credentialEnv)
     throw new Fault(422, "credential_authority", "An endpoint on another origin requires an explicit --credential-env reference.");
   const suffix = selected.protocol === "anthropic-messages" ? "messages" : selected.protocol === "openai-responses" ? "responses" : "chat";
@@ -265,6 +338,26 @@ class SwitcherError extends Error {
     this.requestId = requestId;
   }
 }
+function apiError(status, data, apiKey) {
+  const object = (value) => value !== null && typeof value === "object" && !Array.isArray(value);
+  const error = object(data) && object(data.error) ? data.error : {};
+  const escape = (value) => value.replace(/[.*+?^${}()|[\]\\]/g, "\\$&");
+  const base64 = Buffer.from(apiKey, "utf8").toString("base64");
+  const literal = [apiKey, JSON.stringify(apiKey).slice(1, -1), base64, base64.replace(/=+$/, ""), Buffer.from(apiKey, "utf8").toString("base64url")];
+  const encoded = [encodeURIComponent(apiKey), new URLSearchParams({ key: apiKey }).toString().slice(4)];
+  const patterns = [...new Set(literal)].sort((a, b) => b.length - a.length).map(escape);
+  for (const value of encoded) {
+    patterns.unshift(escape(value).replace(/%[0-9A-F]{2}/g, (part) => part.replace(/[A-F]/g, (letter) => `[${letter}${letter.toLowerCase()}]`)));
+  }
+  const reflected = new RegExp(patterns.join("|"), "g");
+  const redact = (value) => value.replace(reflected, "[REDACTED]");
+  const identifier = (value, pattern) => typeof value === "string" && pattern.test(value) && redact(value) === value ? value : undefined;
+  const code = identifier(error.code, /^[A-Za-z][A-Za-z0-9_.-]{0,63}$/) ?? "api_error";
+  const requestId = identifier(error.requestId, /^[A-Za-z0-9][A-Za-z0-9._:-]{0,127}$/);
+  const fallback = `Switcher API returned HTTP ${status}.`;
+  const message = typeof error.message === "string" && error.message.length <= 4096 ? redact(error.message).replace(/[\x00-\x1f\x7f-\x9f]/g, " ").trim().slice(0, 2048) || fallback : fallback;
+  return new SwitcherError(status, code, message, requestId);
+}
 
 class SwitcherClient {
   options;
@@ -301,7 +394,7 @@ class SwitcherClient {
       throw new SwitcherError(response.status, "invalid_response", "Switcher API returned invalid JSON.");
     }
     if (!response.ok)
-      throw new SwitcherError(response.status, data?.error?.code ?? "api_error", data?.error?.message ?? `Switcher API returned HTTP ${response.status}.`, data?.error?.requestId);
+      throw apiError(response.status, data, apiKey);
     return data;
   }
   query(options = {}) {
@@ -368,7 +461,7 @@ class SwitcherClient {
     return this.request("GET", `/v1/runs/${encodeURIComponent(id)}`);
   }
   createRun(input, idempotencyKey) {
-    return this.request("POST", "/v1/runs", input, { idempotencyKey });
+    return this.request("POST", "/v1/runs", { ...input, modelPolicyVersion: 1 }, { idempotencyKey });
   }
   finishRun(id, version, input, idempotencyKey) {
     return this.request("PATCH", `/v1/runs/${encodeURIComponent(id)}`, input, { version, idempotencyKey });