@falai/agent 2.7.0 → 3.0.0

package/src/providers/AnthropicProvider.ts

@@ -1,573 +1,70 @@
 /**
- * Anthropic (Claude) provider implementation with retry and backup models
+ * Anthropic (Claude).
+ *
+ * Anthropic has no native schema mode, so a structured request carries the
+ * schema as an extra system block — placed after the cached one, where a
+ * per-call schema cannot invalidate the system prompt's cache. That, the
+ * thinking budget, its 529, and its cache-token accounting all live in
+ * `@providerkit/core`; this file is the constructor.
  */
 
-import Anthropic from "@anthropic-ai/sdk";
-import type {
-  MessageCreateParamsNonStreaming,
-  Tool,
-} from "@anthropic-ai/sdk/resources/messages";
+import { createAnthropicProvider } from "@providerkit/core";
 
-import type {
-  AiProvider,
-  GenerateMessageInput,
-  GenerateMessageOutput,
-  GenerateMessageStreamChunk,
-  AgentStructuredResponse,
-} from "../types";
-import type { ProviderCapabilities } from "../types/ai";
-import type { HistoryItem } from "../types/history";
-import { withTimeoutAndRetry, withStreamRetry, withBackupFallback, streamWithBackupFallback, backupFallbackLogging, resolveRetryConfig, logger, assertUsableCompletion, combineAbortSignals } from "../utils";
-import {
-  classifyProviderError,
-  isBackupEligible,
-  isRetriableProviderError,
-  toProviderError,
-  type ErrorClassificationOptions,
-} from "./errorClassification";
+import type { ProviderCapabilities } from "../types/ai.js";
+import { ProviderAdapter, type RequestConfig } from "./ProviderAdapter.js";
 
-/**
- * Configuration options for Anthropic provider
- * Uses types from @anthropic-ai/sdk package
- */
 export interface AnthropicProviderOptions {
-  /** Anthropic API key. Optional when `client` is injected (tests). */
-  apiKey?: string;
-  /** Model to use (required) - e.g., "claude-sonnet-5", "claude-opus-5" */
+  /** Anthropic API key */
+  apiKey: string;
+  /** Model to use (required) — e.g. "claude-sonnet-5", "claude-opus-5" */
   model: string;
-  /** Backup models to try if primary fails (default: []) */
+  /** Backup models to try if the primary fails (default: []) */
   backupModels?: string[];
-  /** Default parameters - uses MessageCreateParamsNonStreaming from @anthropic-ai/sdk */
-  config?: Partial<Omit<MessageCreateParamsNonStreaming, "model" | "messages">>;
-  /** Retry configuration */
-  retryConfig?: {
-    timeout?: number;
-    retries?: number;
-  };
+  /** Any endpoint speaking the Anthropic Messages dialect. */
+  baseUrl?: string;
+  /** Request defaults sent with every call */
+  config?: RequestConfig;
+  /** Idle-stream deadline and retry budget */
+  retryConfig?: { timeout?: number; retries?: number };
   /**
-   * Pre-configured SDK client. Overrides the internally-constructed one.
-   * Intended for tests injecting scripted transports; production callers
-   * should pass `apiKey` instead.
+   * Replacement `fetch`, for tests that script the wire.
+   *
+   * v3 note: this replaces the injected SDK client. There is no SDK now, so a
+   * test drives the same bytes the provider really receives rather than an
+   * SDK's idea of them.
    */
-  client?: Anthropic;
+  fetchImpl?: typeof fetch;
 }
 
-/**
- * Anthropic-specific error classification signals (HTTP 529 plus
- * overloaded_error/api_error types and the "overloaded" code).
- */
-const CLASSIFICATION_OPTIONS: ErrorClassificationOptions = {
-  overloadedStatuses: [529],
-  overloadedTypes: ["overloaded_error", "api_error"],
-  overloadedCodes: ["overloaded"],
-};
-
-/**
- * Determines if an error should trigger backup model usage.
- * Derived from the normalized error classification.
- */
-const shouldUseBackupModel = (error: unknown): boolean =>
-  isBackupEligible(classifyProviderError(error, CLASSIFICATION_OPTIONS));
-
-/**
- * Anthropic provider implementation with backup models and retry logic
- */
-export class AnthropicProvider implements AiProvider {
+export class AnthropicProvider extends ProviderAdapter {
   public readonly name = "anthropic";
   public readonly capabilities: ProviderCapabilities = {
     supportsTools: true,
-    supportsNativeJsonSchema: false, // JSON output is enforced via a prompt instruction, not a native schema mode
+    supportsNativeJsonSchema: false,
     supportsStreaming: true,
     supportsStreamingToolCalls: true,
-    // No cache_control is attached to any request yet — reporting true would
-    // make consumers believe repeated system prompts are cached when they are
-    // not. Flip this when caching is actually implemented.
-    supportsPromptCaching: false,
+    supportsPromptCaching: true,
   };
-  private client: Anthropic;
-  private primaryModel: string;
-  private backupModels: string[];
-  private config?: Partial<
-    Omit<MessageCreateParamsNonStreaming, "model" | "messages">
-  >;
-  private retryConfig: { timeout: number; retries: number };
 
   constructor(options: AnthropicProviderOptions) {
-    const { apiKey, model, backupModels = [], config, retryConfig, client } = options;
-
-    if (!client && !apiKey) {
-      throw new Error("Anthropic API key is required");
-    }
-
-    if (!model) {
-      throw new Error("Model is required. Example: 'claude-sonnet-5'");
-    }
-
-    this.client = client ?? new Anthropic({
-      apiKey,
-    });
-    this.primaryModel = model;
-    this.backupModels = backupModels;
-    this.config = config;
-    this.retryConfig = resolveRetryConfig(retryConfig);
-  }
-
-  /**
-   * Build Anthropic-formatted messages from HistoryItem[] array.
-   * System messages are extracted separately (Anthropic uses a `system` param).
-   * Tool results are mapped to Anthropic's tool_result content blocks.
-   * Assistant tool_calls are mapped to tool_use content blocks.
-   */
-  private buildAnthropicMessages(history: HistoryItem[]): {
-    // eslint-disable-next-line @typescript-eslint/no-explicit-any
-    messages: any[];
-    systemMessages: string[];
-  } {
-    // eslint-disable-next-line @typescript-eslint/no-explicit-any
-    const messages: any[] = [];
-    const systemMessages: string[] = [];
-
-    for (const item of history) {
-      switch (item.role) {
-        case "system":
-          systemMessages.push(item.content);
-          break;
-        case "user":
-          messages.push({ role: "user", content: item.content });
-          break;
-        case "assistant":
-          if (item.tool_calls && item.tool_calls.length > 0) {
-            const content: Array<Record<string, unknown>> = [];
-            if (item.content) {
-              content.push({ type: "text", text: item.content });
-            }
-            for (const tc of item.tool_calls) {
-              content.push({
-                type: "tool_use",
-                id: tc.id,
-                name: tc.name,
-                input: tc.arguments,
-              });
-            }
-            messages.push({ role: "assistant", content });
-          } else {
-            messages.push({ role: "assistant", content: item.content || "" });
-          }
-          break;
-        case "tool":
-          // Anthropic tool results are sent as user messages with tool_result content blocks
-          messages.push({
-            role: "user",
-            content: [
-              {
-                type: "tool_result",
-                tool_use_id: item.tool_call_id,
-                content: typeof item.content === "string" ? item.content : JSON.stringify(item.content),
-              },
-            ],
-          });
-          break;
-      }
-    }
-
-    return { messages, systemMessages };
-  }
-
-  async generateMessage<
-    TContext = unknown,
-    TStructured = AgentStructuredResponse
-  >(
-    input: GenerateMessageInput<TContext>
-  ): Promise<GenerateMessageOutput<TStructured>> {
-    return this.generateWithBackup<TContext, TStructured>(input);
-  }
-
-  async *generateMessageStream<
-    TContext = unknown,
-    TStructured = AgentStructuredResponse
-  >(
-    input: GenerateMessageInput<TContext>
-  ): AsyncGenerator<GenerateMessageStreamChunk<TStructured>> {
-    yield* this.generateStreamWithBackup<TContext, TStructured>(input);
-  }
-
-  private async generateWithBackup<
-    TContext = unknown,
-    TStructured = AgentStructuredResponse
-  >(
-    input: GenerateMessageInput<TContext>
-  ): Promise<GenerateMessageOutput<TStructured>> {
-    const observer = backupFallbackLogging("[ANTHROPIC]", shouldUseBackupModel);
-
-    try {
-      return await withBackupFallback({
-        models: [this.primaryModel, ...this.backupModels],
-        attempt: (model) =>
-          this.generateWithModel<TContext, TStructured>(model, input),
-        shouldTryBackup: shouldUseBackupModel,
-        ...observer.callbacks,
-      });
-    } catch (error: unknown) {
-      observer.logExhausted(error);
-      throw toProviderError(error, this.name, CLASSIFICATION_OPTIONS);
-    }
-  }
-
-  private async generateWithModel<
-    TContext = unknown,
-    TStructured = AgentStructuredResponse
-  >(
-    model: string,
-    input: GenerateMessageInput<TContext>
-  ): Promise<GenerateMessageOutput<TStructured>> {
-    const operation = async (signal: AbortSignal): Promise<GenerateMessageOutput> => {
-      // Anthropic requires max_tokens to be specified
-      const maxTokens = input.parameters?.maxOutputTokens || 4096;
-
-      // Build messages from history
-      const { messages: historyMessages, systemMessages } = this.buildAnthropicMessages(input.history);
-
-      // Append the current prompt as the final user message
-      historyMessages.push({
-        role: "user",
-        content: input.prompt,
-      });
-
-      const params: MessageCreateParamsNonStreaming = {
-        model,
-        max_tokens: maxTokens,
-        messages: historyMessages,
-        ...this.config,
-      };
-
-      // Set system messages from history if present
-      if (systemMessages.length > 0) {
-        if (typeof this.config?.system === "string") {
-          params.system = `${this.config.system}\n\n${systemMessages.join("\n\n")}`;
-        } else if (Array.isArray(this.config?.system)) {
-          params.system = [
-            ...this.config.system,
-            ...systemMessages.map(s => ({ type: "text" as const, text: s })),
-          ];
-        } else {
-          params.system = systemMessages.join("\n\n");
-        }
-      }
-
-      // Add tools if provided
-      if (input.tools && input.tools.length > 0) {
-        params.tools = input.tools.map((tool) => ({
-          name: tool.name || tool.id,
-          description: tool.description || "",
-          input_schema: tool.parameters as Tool["input_schema"], // JSON schema
-        }));
-      }
-
-      // Handle schema: Anthropic doesn't have a native schema mode, so embed constraints
-      if (input.parameters?.jsonSchema) {
-        const systemPrompt =
-          "You must respond with valid JSON only and it MUST match the provided schema.";
-
-        // Merge with existing system if present
-        if (typeof this.config?.system === "string") {
-          params.system = `${this.config.system}\n\n${systemPrompt}`;
-        } else if (Array.isArray(this.config?.system)) {
-          params.system = [
-            ...this.config.system,
-            {
-              type: "text" as const,
-              text: systemPrompt,
-            },
-          ];
-        } else {
-          params.system = systemPrompt;
-        }
-      }
-
-      const response = await this.client.messages.create(params, {
-        signal: combineAbortSignals(input.signal, signal),
-      });
-
-      // Extract text and tool calls from response
-      const textContent = response.content.find(
-        (block) => block.type === "text"
-      );
-      const message = textContent?.type === "text" ? textContent.text : "";
-
-      // Extract tool calls from response
-      const toolCalls: Array<{
-        toolName: string;
-        arguments: Record<string, unknown>;
-      }> = [];
-
-      // Check for tool_use content blocks
-      for (const block of response.content) {
-        if (block.type === "tool_use") {
-          toolCalls.push({
-            toolName: block.name,
-            arguments: block.input as Record<string, unknown>,
-          });
-        }
-      }
-
-      // Only throw error if we have no text AND no function calls
-      if (!message && toolCalls.length === 0) {
-        throw new Error("No response from Anthropic");
-      }
-
-      // Parse JSON response if schema was provided
-      let structured: AgentStructuredResponse | undefined;
-      if (input.parameters?.jsonSchema) {
-        try {
-          structured = JSON.parse(message) as AgentStructuredResponse;
-        } catch (error) {
-          logger.warn("[ANTHROPIC] Failed to parse JSON response:", error);
-          // Fall back to treating the message as plain text
-        }
-      }
-
-      // If tools were used, include them in structured response
-      if (toolCalls.length > 0) {
-        structured = {
-          ...(structured || {}),
-          message: structured?.message || message,
-          toolCalls,
-        } as AgentStructuredResponse;
-      }
-
-      // A parsed-but-blank message with no tool calls is as empty as no text;
-      // the shared guard throws so withTimeoutAndRetry retries instead of
-      // returning {"message":""}.
-      assertUsableCompletion(structured, message, toolCalls.length, "Anthropic");
-
-      return {
-        message,
-        metadata: {
-          model: response.model,
-          stopReason: response.stop_reason,
-          tokensUsed:
-            response.usage.input_tokens + response.usage.output_tokens,
-          promptTokens: response.usage.input_tokens,
-          completionTokens: response.usage.output_tokens,
-        },
-        structured,
-      };
-    };
-
-    return withTimeoutAndRetry(
-      operation,
-      this.retryConfig.timeout,
-      this.retryConfig.retries,
-      `Anthropic ${model}`,
-      (error) => isRetriableProviderError(error, CLASSIFICATION_OPTIONS)
-    ) as Promise<GenerateMessageOutput<TStructured>>;
-  }
-
-  private async *generateStreamWithBackup<
-    TContext = unknown,
-    TStructured = AgentStructuredResponse
-  >(
-    input: GenerateMessageInput<TContext>
-  ): AsyncGenerator<GenerateMessageStreamChunk<TStructured>> {
-    const observer = backupFallbackLogging("[ANTHROPIC]", shouldUseBackupModel, { streaming: true });
-
-    try {
-      yield* streamWithBackupFallback({
-        models: [this.primaryModel, ...this.backupModels],
-        attempt: (model) =>
-          withStreamRetry(
-            (signal) => this.generateStreamWithModel<TContext, TStructured>(model, input, signal),
-            { maxRetries: this.retryConfig.retries, firstChunkTimeoutMs: this.retryConfig.timeout, operationName: `Anthropic ${model} stream`, isRetriable: (error) => isRetriableProviderError(error, CLASSIFICATION_OPTIONS) }
-          ),
-        shouldTryBackup: shouldUseBackupModel,
-        ...observer.callbacks,
-      });
-    } catch (error: unknown) {
-      observer.logExhausted(error);
-      throw toProviderError(error, this.name, CLASSIFICATION_OPTIONS);
+    if (!options.apiKey) throw new Error("Anthropic API key is required");
+    if (!options.model) {
+      throw new Error("Model is required. Example: 'claude-sonnet-5' or 'claude-opus-5'");
     }
-  }
-
-  private async *generateStreamWithModel<
-    TContext = unknown,
-    TStructured = AgentStructuredResponse
-  >(
-    model: string,
-    input: GenerateMessageInput<TContext>,
-    attemptSignal?: AbortSignal
-  ): AsyncGenerator<GenerateMessageStreamChunk<TStructured>> {
-    // Anthropic requires max_tokens to be specified
-    const maxTokens = input.parameters?.maxOutputTokens || 4096;
-
-    // Build messages from history
-    const { messages: historyMessages, systemMessages } = this.buildAnthropicMessages(input.history);
 
-    // Append the current prompt as the final user message
-    historyMessages.push({
-      role: "user" as const,
-      content: input.prompt,
+    super({
+      provider: createAnthropicProvider({
+        apiKey: options.apiKey,
+        model: options.model,
+        id: "anthropic",
+        ...(options.baseUrl ? { baseUrl: options.baseUrl } : {}),
+        ...(options.config?.maxTokens ? { maxTokens: options.config.maxTokens } : {}),
+        ...(options.fetchImpl ? { fetchImpl: options.fetchImpl } : {}),
+      }),
+      model: options.model,
+      ...(options.config ? { defaults: options.config } : {}),
+      ...(options.backupModels ? { backupModels: options.backupModels } : {}),
+      ...(options.retryConfig ? { retryConfig: options.retryConfig } : {}),
     });
-
-    const params = {
-      model,
-      max_tokens: maxTokens,
-      messages: historyMessages,
-      stream: true,
-      ...this.config,
-    };
-
-    // Set system messages from history if present
-    if (systemMessages.length > 0) {
-      if (typeof this.config?.system === "string") {
-        params.system = `${this.config.system}\n\n${systemMessages.join("\n\n")}`;
-      } else if (Array.isArray(this.config?.system)) {
-        params.system = [
-          ...this.config.system,
-          ...systemMessages.map(s => ({ type: "text" as const, text: s })),
-        ];
-      } else {
-        params.system = systemMessages.join("\n\n");
-      }
-    }
-
-    // Add tools if provided
-    if (input.tools && input.tools.length > 0) {
-      params.tools = input.tools.map((tool) => ({
-        name: tool.name || tool.id,
-        description: tool.description || "",
-        input_schema: tool.parameters as Tool["input_schema"], // JSON schema
-      }));
-    }
-
-    // Handle schema in streaming: embed constraint
-    if (input.parameters?.jsonSchema) {
-      const systemPrompt =
-        "You must respond with valid JSON only and it MUST match the provided schema.";
-
-      if (typeof this.config?.system === "string") {
-        params.system = `${this.config.system}\n\n${systemPrompt}`;
-      } else if (Array.isArray(this.config?.system)) {
-        params.system = [
-          ...this.config.system,
-          {
-            type: "text" as const,
-            text: systemPrompt,
-          },
-        ];
-      } else {
-        params.system = systemPrompt;
-      }
-    }
-
-    const stream = this.client.messages.stream(params, {
-      signal: combineAbortSignals(input.signal, attemptSignal),
-    });
-
-    let accumulated = "";
-    let currentModel = model;
-    let stopReason: string | undefined;
-    let inputTokens = 0;
-    let outputTokens = 0;
-    // Tool_use blocks arrive empty: content_block_start carries `input: {}`,
-    // the real arguments stream in later as input_json_delta partial_json
-    // fragments. Accumulate per block index and parse once at stream end.
-    const pendingToolBlocks = new Map<
-      number,
-      { toolName: string; argumentsBuffer: string }
-    >();
-
-    for await (const chunk of stream) {
-      if (chunk.type === "message_start") {
-        currentModel = chunk.message.model;
-        inputTokens = chunk.message.usage.input_tokens;
-      } else if (chunk.type === "content_block_start") {
-        if (chunk.content_block.type === "tool_use") {
-          pendingToolBlocks.set(chunk.index, {
-            toolName: chunk.content_block.name,
-            argumentsBuffer: "",
-          });
-        }
-      } else if (chunk.type === "content_block_delta") {
-        if (chunk.delta.type === "text_delta") {
-          const delta = chunk.delta.text;
-          accumulated += delta;
-          yield {
-            delta,
-            accumulated,
-            done: false,
-          } as GenerateMessageStreamChunk<TStructured>;
-        } else if (chunk.delta.type === "input_json_delta") {
-          const block = pendingToolBlocks.get(chunk.index);
-          if (block) {
-            block.argumentsBuffer += chunk.delta.partial_json;
-          }
-        }
-      } else if (chunk.type === "message_delta") {
-        stopReason = chunk.delta.stop_reason || undefined;
-        outputTokens = chunk.usage.output_tokens;
-      }
-    }
-
-    const toolCalls: Array<{
-      toolName: string;
-      arguments: Record<string, unknown>;
-    }> = [];
-    for (const [, block] of [...pendingToolBlocks.entries()].sort(
-      ([a], [b]) => a - b
-    )) {
-      let args: Record<string, unknown> = {};
-      if (block.argumentsBuffer) {
-        try {
-          args = JSON.parse(block.argumentsBuffer) as Record<string, unknown>;
-        } catch (error) {
-          logger.warn(
-            "[ANTHROPIC] Failed to parse streamed tool call arguments:",
-            error
-          );
-        }
-      }
-      toolCalls.push({ toolName: block.toolName, arguments: args });
-    }
-
-    // Parse JSON response if schema was provided
-    let structured: AgentStructuredResponse | undefined;
-    if (input.parameters?.jsonSchema && accumulated) {
-      try {
-        structured = JSON.parse(accumulated) as AgentStructuredResponse;
-      } catch (error) {
-        logger.warn(
-          "[ANTHROPIC] Failed to parse JSON response in stream:",
-          error
-        );
-      }
-    }
-
-    // If tools were used, include them in structured response
-    if (toolCalls.length > 0) {
-      structured = {
-        ...(structured || {}),
-        message: structured?.message || accumulated,
-        toolCalls,
-      } as AgentStructuredResponse;
-    }
-
-    // Empty-completion guard — same definition as the non-streaming path, so the
-    // stream retries / falls back to backup instead of emitting an empty message.
-    assertUsableCompletion(structured, accumulated, toolCalls.length, "Anthropic");
-
-    // Yield final chunk
-    yield {
-      delta: "",
-      accumulated,
-      done: true,
-      metadata: {
-        model: currentModel,
-        stopReason,
-        tokensUsed: inputTokens + outputTokens,
-        promptTokens: inputTokens,
-        completionTokens: outputTokens,
-      },
-      structured: structured as TStructured,
-    };
   }
 }